mariadb/sql/sql_base.cc

9212 lines
298 KiB
C++
Raw Normal View History

/* Copyright (c) 2000, 2011, Oracle and/or its affiliates. All rights reserved.
2000-07-31 21:29:14 +02:00
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
2000-07-31 21:29:14 +02:00
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
2000-07-31 21:29:14 +02:00
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
/* Basic functions needed by many modules */
2000-07-31 21:29:14 +02:00
#include "sql_base.h" // setup_table_map
#include "my_global.h" /* NO_EMBEDDED_ACCESS_CHECKS */
#include "sql_priv.h"
#include "unireg.h"
#include "debug_sync.h"
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
#include "lock.h" // mysql_lock_remove,
// mysql_unlock_tables,
// mysql_lock_have_duplicate
#include "sql_show.h" // append_identifier
#include "strfunc.h" // find_type
#include "parse_file.h" // sql_parse_prepare, File_parser
#include "sql_view.h" // mysql_make_view, VIEW_ANY_ACL
#include "sql_parse.h" // check_table_access
#include "sql_insert.h" // kill_delayed_threads
#include "sql_acl.h" // *_ACL, check_grant_all_columns,
// check_column_grant_in_table_ref,
// get_column_grant
#include "sql_partition.h" // ALTER_PARTITION_PARAM_TYPE
#include "sql_derived.h" // mysql_derived_prepare,
// mysql_handle_derived,
// mysql_derived_filling
#include "sql_handler.h" // mysql_ha_flush
#include "sql_partition.h" // ALTER_PARTITION_PARAM_TYPE
#include "log_event.h" // Query_log_event
#include "sql_select.h"
#include "sp_head.h"
#include "sp.h"
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
#include "sp_cache.h"
#include "sql_trigger.h"
#include "transaction.h"
Backport of revno 2630.28.10, 2630.28.31, 2630.28.26, 2630.33.1, 2630.39.1, 2630.28.29, 2630.34.3, 2630.34.2, 2630.34.1, 2630.29.29, 2630.29.28, 2630.31.1, 2630.28.13, 2630.28.10, 2617.23.14 and some other minor revisions. This patch implements: WL#4264 "Backup: Stabilize Service Interface" -- all the server prerequisites except si_objects.{h,cc} themselves (they can be just copied over, when needed). WL#4435: Support OUT-parameters in prepared statements. (and all issues in the initial patches for these two tasks, that were discovered in pushbuild and during testing). Bug#39519: mysql_stmt_close() should flush all data associated with the statement. After execution of a prepared statement, send OUT parameters of the invoked stored procedure, if any, to the client. When using the binary protocol, send the parameters in an additional result set over the wire. When using the text protocol, assign out parameters to the user variables from the CALL(@var1, @var2, ...) specification. The following refactoring has been made: - Protocol::send_fields() was renamed to Protocol::send_result_set_metadata(); - A new Protocol::send_result_set_row() was introduced to incapsulate common functionality for sending row data. - Signature of Protocol::prepare_for_send() was changed: this operation does not need a list of items, the number of items is fully sufficient. The following backward incompatible changes have been made: - CLIENT_MULTI_RESULTS is now enabled by default in the client; - CLIENT_PS_MULTI_RESUTLS is now enabled by default in the client.
2009-10-21 22:02:06 +02:00
#include "sql_prepare.h"
2000-07-31 21:29:14 +02:00
#include <m_ctype.h>
#include <my_dir.h>
#include <hash.h>
#include "rpl_filter.h"
#include "sql_table.h" // build_table_filename
#include "datadict.h" // dd_frm_type()
#include "sql_hset.h" // Hash_set
#ifdef __WIN__
2000-07-31 21:29:14 +02:00
#include <io.h>
#endif
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
bool
No_such_table_error_handler::handle_condition(THD *,
uint sql_errno,
const char*,
MYSQL_ERROR::enum_warning_level,
const char*,
MYSQL_ERROR ** cond_hdl)
{
*cond_hdl= NULL;
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
if (sql_errno == ER_NO_SUCH_TABLE)
{
m_handled_errors++;
2007-07-20 17:46:13 +02:00
return TRUE;
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
}
m_unhandled_errors++;
2007-07-20 17:46:13 +02:00
return FALSE;
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
}
bool No_such_table_error_handler::safely_trapped_errors()
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
{
/*
If m_unhandled_errors != 0, something else, unanticipated, happened,
so the error is not trapped but returned to the caller.
Multiple ER_NO_SUCH_TABLE can be raised in case of views.
*/
return ((m_handled_errors > 0) && (m_unhandled_errors == 0));
}
/**
@defgroup Data_Dictionary Data Dictionary
@{
*/
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/**
Protects table_def_hash, used and unused lists in the
TABLE_SHARE object, LRU lists of used TABLEs and used
TABLE_SHAREs, refresh_version and the table id counter.
*/
mysql_mutex_t LOCK_open;
#ifdef HAVE_PSI_INTERFACE
static PSI_mutex_key key_LOCK_open;
static PSI_mutex_info all_tdc_mutexes[]= {
{ &key_LOCK_open, "LOCK_open", PSI_FLAG_GLOBAL }
};
/**
Initialize performance schema instrumentation points
used by the table cache.
*/
static void init_tdc_psi_keys(void)
{
const char *category= "sql";
int count;
if (PSI_server == NULL)
return;
count= array_elements(all_tdc_mutexes);
PSI_server->register_mutex(category, all_tdc_mutexes, count);
}
#endif /* HAVE_PSI_INTERFACE */
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/**
Total number of TABLE instances for tables in the table definition cache
(both in use by threads and not in use). This value is accessible to user
as "Open_tables" status variable.
*/
uint table_cache_count= 0;
/**
List that contains all TABLE instances for tables in the table definition
cache that are not in use by any thread. Recently used TABLE instances are
appended to the end of the list. Thus the beginning of the list contains
tables which have been least recently used.
*/
TABLE *unused_tables;
HASH table_def_cache;
static TABLE_SHARE *oldest_unused_share, end_of_unused_share;
static bool table_def_inited= 0;
static bool table_def_shutdown_in_progress= 0;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
static bool check_and_update_table_version(THD *thd, TABLE_LIST *tables,
TABLE_SHARE *table_share);
static bool open_table_entry_fini(THD *thd, TABLE_SHARE *share, TABLE *entry);
static bool auto_repair_table(THD *thd, TABLE_LIST *table_list);
2000-07-31 21:29:14 +02:00
static void free_cache_entry(TABLE *entry);
* Mixed replication mode * : 1) Fix for BUG#19630 "stored function inserting into two auto_increment breaks statement-based binlog": a stored function inserting into two such tables may fail to replicate (inserting wrong data in the slave's copy of the second table) if the slave's second table had an internal auto_increment counter different from master's. Because the auto_increment value autogenerated by master for the 2nd table does not go into binlog, only the first does, so the slave lacks information. To fix this, if running in mixed binlogging mode, if the stored function or trigger plans to update two different tables both having auto_increment columns, we switch to row-based for the whole function. We don't have a simple solution for statement-based binlogging mode, there the bug remains and will be documented as a known problem. Re-enabling rpl_switch_stm_row_mixed. 2) Fix for BUG#20630 "Mixed binlogging mode does not work with stored functions, triggers, views", which was a documented limitation (in mixed mode, we didn't detect that a stored function's execution needed row-based binlogging (due to some UUID() call for example); same for triggers, same for views (a view created from a SELECT UUID(), and doing INSERT INTO sometable SELECT theview; would not replicate row-based). This is implemented by, after parsing a routine's body, remembering in sp_head that this routine needs row-based binlogging. Then when this routine is used, the caller is marked to require row-based binlogging too. Same for views: when we parse a view and detect that its SELECT needs row-based binary logging, we mark the calling LEX as such. 3) Fix for BUG#20499 "mixed mode with temporary table breaks binlog": a temporary table containing e.g. UUID has its changes not binlogged, so any query updating a permanent table with data from the temporary table will run wrongly on slave. Solution: in mixed mode we don't switch back from row-based to statement-based when there exists temporary tables. 4) Attempt to test mysqlbinlog on a binlog generated by mysqlbinlog; impossible due to BUG#11312 and BUG#20329, but test is in place for when they are fixed.
2006-07-09 17:00:47 +02:00
static bool
has_write_table_with_auto_increment(TABLE_LIST *tables);
2000-07-31 21:29:14 +02:00
uint cached_open_tables(void)
2000-07-31 21:29:14 +02:00
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
return table_cache_count;
2000-07-31 21:29:14 +02:00
}
2000-07-31 21:29:14 +02:00
#ifdef EXTRA_DEBUG
static void check_unused(void)
{
uint count= 0, open_files= 0, idx= 0;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
TABLE *cur_link, *start_link, *entry;
TABLE_SHARE *share;
2000-07-31 21:29:14 +02:00
if ((start_link=cur_link=unused_tables))
{
do
{
if (cur_link != cur_link->next->prev || cur_link != cur_link->prev->next)
{
DBUG_PRINT("error",("Unused_links aren't linked properly")); /* purecov: inspected */
return; /* purecov: inspected */
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
} while (count++ < table_cache_count &&
2000-07-31 21:29:14 +02:00
(cur_link=cur_link->next) != start_link);
if (cur_link != start_link)
{
DBUG_PRINT("error",("Unused_links aren't connected")); /* purecov: inspected */
}
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
for (idx=0 ; idx < table_def_cache.records ; idx++)
2000-07-31 21:29:14 +02:00
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
share= (TABLE_SHARE*) my_hash_element(&table_def_cache, idx);
I_P_List_iterator<TABLE, TABLE_share> it(share->free_tables);
while ((entry= it++))
{
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
/* We must not have TABLEs in the free list that have their file closed. */
DBUG_ASSERT(entry->db_stat && entry->file);
/* Merge children should be detached from a merge parent */
DBUG_ASSERT(! entry->file->extra(HA_EXTRA_IS_ATTACHED_CHILDREN));
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (entry->in_use)
{
DBUG_PRINT("error",("Used table is in share's list of unused tables")); /* purecov: inspected */
}
2000-07-31 21:29:14 +02:00
count--;
open_files++;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
it.init(share->used_tables);
while ((entry= it++))
{
if (!entry->in_use)
{
DBUG_PRINT("error",("Unused table is in share's list of used tables")); /* purecov: inspected */
}
open_files++;
}
2000-07-31 21:29:14 +02:00
}
if (count != 0)
{
DBUG_PRINT("error",("Unused_links doesn't match open_cache: diff: %d", /* purecov: inspected */
2000-07-31 21:29:14 +02:00
count)); /* purecov: inspected */
}
}
#else
#define check_unused()
#endif
/*
Create a table cache key
SYNOPSIS
create_table_def_key()
thd Thread handler
key Create key here (must be of size MAX_DBKEY_LENGTH)
table_list Table definition
tmp_table Set if table is a tmp table
IMPLEMENTATION
The table cache_key is created from:
db_name + \0
table_name + \0
if the table is a tmp table, we add the following to make each tmp table
unique on the slave:
4 bytes for master thread id
4 bytes pseudo thread id
RETURN
Length of key
*/
uint create_table_def_key(THD *thd, char *key,
const TABLE_LIST *table_list,
bool tmp_table)
{
uint key_length= (uint) (strmov(strmov(key, table_list->db)+1,
table_list->table_name)-key)+1;
if (tmp_table)
{
int4store(key + key_length, thd->server_id);
int4store(key + key_length + 4, thd->variables.pseudo_thread_id);
key_length+= TMP_TABLE_KEY_EXTRA;
}
return key_length;
}
/*****************************************************************************
Functions to handle table definition cach (TABLE_SHARE)
*****************************************************************************/
WL#3817: Simplify string / memory area types and make things more consistent (first part) The following type conversions was done: - Changed byte to uchar - Changed gptr to uchar* - Change my_string to char * - Change my_size_t to size_t - Change size_s to size_t Removed declaration of byte, gptr, my_string, my_size_t and size_s. Following function parameter changes was done: - All string functions in mysys/strings was changed to use size_t instead of uint for string lengths. - All read()/write() functions changed to use size_t (including vio). - All protocoll functions changed to use size_t instead of uint - Functions that used a pointer to a string length was changed to use size_t* - Changed malloc(), free() and related functions from using gptr to use void * as this requires fewer casts in the code and is more in line with how the standard functions work. - Added extra length argument to dirname_part() to return the length of the created string. - Changed (at least) following functions to take uchar* as argument: - db_dump() - my_net_write() - net_write_command() - net_store_data() - DBUG_DUMP() - decimal2bin() & bin2decimal() - Changed my_compress() and my_uncompress() to use size_t. Changed one argument to my_uncompress() from a pointer to a value as we only return one value (makes function easier to use). - Changed type of 'pack_data' argument to packfrm() to avoid casts. - Changed in readfrm() and writefrom(), ha_discover and handler::discover() the type for argument 'frmdata' to uchar** to avoid casts. - Changed most Field functions to use uchar* instead of char* (reduced a lot of casts). - Changed field->val_xxx(xxx, new_ptr) to take const pointers. Other changes: - Removed a lot of not needed casts - Added a few new cast required by other changes - Added some cast to my_multi_malloc() arguments for safety (as string lengths needs to be uint, not size_t). - Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done explicitely as this conflict was often hided by casting the function to hash_get_key). - Changed some buffers to memory regions to uchar* to avoid casts. - Changed some string lengths from uint to size_t. - Changed field->ptr to be uchar* instead of char*. This allowed us to get rid of a lot of casts. - Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar - Include zlib.h in some files as we needed declaration of crc32() - Changed MY_FILE_ERROR to be (size_t) -1. - Changed many variables to hold the result of my_read() / my_write() to be size_t. This was needed to properly detect errors (which are returned as (size_t) -1). - Removed some very old VMS code - Changed packfrm()/unpackfrm() to not be depending on uint size (portability fix) - Removed windows specific code to restore cursor position as this causes slowdown on windows and we should not mix read() and pread() calls anyway as this is not thread safe. Updated function comment to reflect this. Changed function that depended on original behavior of my_pwrite() to itself restore the cursor position (one such case). - Added some missing checking of return value of malloc(). - Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow. - Changed type of table_def::m_size from my_size_t to ulong to reflect that m_size is the number of elements in the array, not a string/memory length. - Moved THD::max_row_length() to table.cc (as it's not depending on THD). Inlined max_row_length_blob() into this function. - More function comments - Fixed some compiler warnings when compiled without partitions. - Removed setting of LEX_STRING() arguments in declaration (portability fix). - Some trivial indentation/variable name changes. - Some trivial code simplifications: - Replaced some calls to alloc_root + memcpy to use strmake_root()/strdup_root(). - Changed some calls from memdup() to strmake() (Safety fix) - Simpler loops in client-simple.c
2007-05-10 11:59:39 +02:00
extern "C" uchar *table_def_key(const uchar *record, size_t *length,
my_bool not_used __attribute__((unused)))
{
TABLE_SHARE *entry=(TABLE_SHARE*) record;
*length= entry->table_cache_key.length;
WL#3817: Simplify string / memory area types and make things more consistent (first part) The following type conversions was done: - Changed byte to uchar - Changed gptr to uchar* - Change my_string to char * - Change my_size_t to size_t - Change size_s to size_t Removed declaration of byte, gptr, my_string, my_size_t and size_s. Following function parameter changes was done: - All string functions in mysys/strings was changed to use size_t instead of uint for string lengths. - All read()/write() functions changed to use size_t (including vio). - All protocoll functions changed to use size_t instead of uint - Functions that used a pointer to a string length was changed to use size_t* - Changed malloc(), free() and related functions from using gptr to use void * as this requires fewer casts in the code and is more in line with how the standard functions work. - Added extra length argument to dirname_part() to return the length of the created string. - Changed (at least) following functions to take uchar* as argument: - db_dump() - my_net_write() - net_write_command() - net_store_data() - DBUG_DUMP() - decimal2bin() & bin2decimal() - Changed my_compress() and my_uncompress() to use size_t. Changed one argument to my_uncompress() from a pointer to a value as we only return one value (makes function easier to use). - Changed type of 'pack_data' argument to packfrm() to avoid casts. - Changed in readfrm() and writefrom(), ha_discover and handler::discover() the type for argument 'frmdata' to uchar** to avoid casts. - Changed most Field functions to use uchar* instead of char* (reduced a lot of casts). - Changed field->val_xxx(xxx, new_ptr) to take const pointers. Other changes: - Removed a lot of not needed casts - Added a few new cast required by other changes - Added some cast to my_multi_malloc() arguments for safety (as string lengths needs to be uint, not size_t). - Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done explicitely as this conflict was often hided by casting the function to hash_get_key). - Changed some buffers to memory regions to uchar* to avoid casts. - Changed some string lengths from uint to size_t. - Changed field->ptr to be uchar* instead of char*. This allowed us to get rid of a lot of casts. - Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar - Include zlib.h in some files as we needed declaration of crc32() - Changed MY_FILE_ERROR to be (size_t) -1. - Changed many variables to hold the result of my_read() / my_write() to be size_t. This was needed to properly detect errors (which are returned as (size_t) -1). - Removed some very old VMS code - Changed packfrm()/unpackfrm() to not be depending on uint size (portability fix) - Removed windows specific code to restore cursor position as this causes slowdown on windows and we should not mix read() and pread() calls anyway as this is not thread safe. Updated function comment to reflect this. Changed function that depended on original behavior of my_pwrite() to itself restore the cursor position (one such case). - Added some missing checking of return value of malloc(). - Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow. - Changed type of table_def::m_size from my_size_t to ulong to reflect that m_size is the number of elements in the array, not a string/memory length. - Moved THD::max_row_length() to table.cc (as it's not depending on THD). Inlined max_row_length_blob() into this function. - More function comments - Fixed some compiler warnings when compiled without partitions. - Removed setting of LEX_STRING() arguments in declaration (portability fix). - Some trivial indentation/variable name changes. - Some trivial code simplifications: - Replaced some calls to alloc_root + memcpy to use strmake_root()/strdup_root(). - Changed some calls from memdup() to strmake() (Safety fix) - Simpler loops in client-simple.c
2007-05-10 11:59:39 +02:00
return (uchar*) entry->table_cache_key.str;
}
static void table_def_free_entry(TABLE_SHARE *share)
{
DBUG_ENTER("table_def_free_entry");
2010-02-02 00:22:16 +01:00
mysql_mutex_assert_owner(&LOCK_open);
if (share->prev)
{
/* remove from old_unused_share list */
*share->prev= share->next;
share->next->prev= share->prev;
}
free_table_share(share);
DBUG_VOID_RETURN;
}
bool table_def_init(void)
{
table_def_inited= 1;
#ifdef HAVE_PSI_INTERFACE
init_tdc_psi_keys();
#endif
mysql_mutex_init(key_LOCK_open, &LOCK_open, MY_MUTEX_INIT_FAST);
oldest_unused_share= &end_of_unused_share;
end_of_unused_share.prev= &oldest_unused_share;
return my_hash_init(&table_def_cache, &my_charset_bin, table_def_size,
0, 0, table_def_key,
(my_hash_free_key) table_def_free_entry, 0) != 0;
}
/**
Notify table definition cache that process of shutting down server
has started so it has to keep number of TABLE and TABLE_SHARE objects
minimal in order to reduce number of references to pluggable engines.
*/
void table_def_start_shutdown(void)
{
if (table_def_inited)
{
2010-02-02 00:22:16 +01:00
mysql_mutex_lock(&LOCK_open);
/*
Ensure that TABLE and TABLE_SHARE objects which are created for
tables that are open during process of plugins' shutdown are
immediately released. This keeps number of references to engine
plugins minimal and allows shutdown to proceed smoothly.
*/
table_def_shutdown_in_progress= TRUE;
2010-02-02 00:22:16 +01:00
mysql_mutex_unlock(&LOCK_open);
/* Free all cached but unused TABLEs and TABLE_SHAREs. */
close_cached_tables(NULL, NULL, FALSE, LONG_TIMEOUT);
}
}
void table_def_free(void)
{
DBUG_ENTER("table_def_free");
if (table_def_inited)
{
table_def_inited= 0;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/* Free table definitions. */
my_hash_free(&table_def_cache);
mysql_mutex_destroy(&LOCK_open);
}
DBUG_VOID_RETURN;
}
uint cached_table_definitions(void)
{
return table_def_cache.records;
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/*
Auxiliary routines for manipulating with per-share used/unused and
global unused lists of TABLE objects and table_cache_count counter.
Responsible for preserving invariants between those lists, counter
and TABLE::in_use member.
In fact those routines implement sort of implicit table cache as
part of table definition cache.
*/
/**
Add newly created TABLE object for table share which is going
to be used right away.
*/
static void table_def_add_used_table(THD *thd, TABLE *table)
{
DBUG_ASSERT(table->in_use == thd);
table->s->used_tables.push_front(table);
table_cache_count++;
}
/**
Prepare used or unused TABLE instance for destruction by removing
it from share's and global list.
*/
static void table_def_remove_table(TABLE *table)
{
if (table->in_use)
{
/* Remove from per-share chain of used TABLE objects. */
table->s->used_tables.remove(table);
}
else
{
/* Remove from per-share chain of unused TABLE objects. */
table->s->free_tables.remove(table);
/* And global unused chain. */
table->next->prev=table->prev;
table->prev->next=table->next;
if (table == unused_tables)
{
unused_tables=unused_tables->next;
if (table == unused_tables)
unused_tables=0;
}
check_unused();
}
table_cache_count--;
}
/**
Mark already existing TABLE instance as used.
*/
static void table_def_use_table(THD *thd, TABLE *table)
{
DBUG_ASSERT(!table->in_use);
/* Unlink table from list of unused tables for this share. */
table->s->free_tables.remove(table);
/* Unlink able from global unused tables list. */
if (table == unused_tables)
{ // First unused
unused_tables=unused_tables->next; // Remove from link
if (table == unused_tables)
unused_tables=0;
}
table->prev->next=table->next; /* Remove from unused list */
table->next->prev=table->prev;
check_unused();
/* Add table to list of used tables for this share. */
table->s->used_tables.push_front(table);
table->in_use= thd;
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
/* The ex-unused table must be fully functional. */
DBUG_ASSERT(table->db_stat && table->file);
/* The children must be detached from the table. */
DBUG_ASSERT(! table->file->extra(HA_EXTRA_IS_ATTACHED_CHILDREN));
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
/**
Mark already existing used TABLE instance as unused.
*/
static void table_def_unuse_table(TABLE *table)
{
DBUG_ASSERT(table->in_use);
/* We shouldn't put the table to 'unused' list if the share is old. */
DBUG_ASSERT(! table->s->has_old_version());
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
table->in_use= 0;
/* Remove table from the list of tables used in this share. */
table->s->used_tables.remove(table);
/* Add table to the list of unused TABLE objects for this share. */
table->s->free_tables.push_front(table);
/* Also link it last in the global list of unused TABLE objects. */
if (unused_tables)
{
table->next=unused_tables;
table->prev=unused_tables->prev;
unused_tables->prev=table;
table->prev->next=table;
}
else
unused_tables=table->next=table->prev=table;
check_unused();
}
/*
Get TABLE_SHARE for a table.
get_table_share()
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
thd Thread handle
table_list Table that should be opened
key Table cache key
key_length Length of key
db_flags Flags to open_table_def():
OPEN_VIEW
error out: Error code from open_table_def()
IMPLEMENTATION
Get a table definition from the table definition cache.
If it doesn't exist, create a new from the table definition file.
NOTES
We must have wrlock on LOCK_open when we come here
(To be changed later)
RETURN
0 Error
# Share for table
*/
TABLE_SHARE *get_table_share(THD *thd, TABLE_LIST *table_list, char *key,
2010-02-03 14:43:03 +01:00
uint key_length, uint db_flags, int *error,
my_hash_value_type hash_value)
{
TABLE_SHARE *share;
DBUG_ENTER("get_table_share");
*error= 0;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/*
To be able perform any operation on table we should own
some kind of metadata lock on it.
*/
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
DBUG_ASSERT(thd->mdl_context.is_lock_owner(MDL_key::TABLE,
table_list->db,
table_list->table_name,
MDL_SHARED));
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/* Read table definition from cache */
2010-02-03 14:43:03 +01:00
if ((share= (TABLE_SHARE*) my_hash_search_using_hash_value(&table_def_cache,
hash_value, (uchar*) key, key_length)))
goto found;
if (!(share= alloc_table_share(table_list, key, key_length)))
{
DBUG_RETURN(0);
}
/*
We assign a new table id under the protection of LOCK_open.
We do this instead of creating a new mutex
and using it for the sole purpose of serializing accesses to a
static variable, we assign the table id here. We assign it to the
share before inserting it into the table_def_cache to be really
sure that it cannot be read from the cache without having a table
id assigned.
CAVEAT. This means that the table cannot be used for
binlogging/replication purposes, unless get_table_share() has been
called directly or indirectly.
*/
assign_new_table_id(share);
WL#3817: Simplify string / memory area types and make things more consistent (first part) The following type conversions was done: - Changed byte to uchar - Changed gptr to uchar* - Change my_string to char * - Change my_size_t to size_t - Change size_s to size_t Removed declaration of byte, gptr, my_string, my_size_t and size_s. Following function parameter changes was done: - All string functions in mysys/strings was changed to use size_t instead of uint for string lengths. - All read()/write() functions changed to use size_t (including vio). - All protocoll functions changed to use size_t instead of uint - Functions that used a pointer to a string length was changed to use size_t* - Changed malloc(), free() and related functions from using gptr to use void * as this requires fewer casts in the code and is more in line with how the standard functions work. - Added extra length argument to dirname_part() to return the length of the created string. - Changed (at least) following functions to take uchar* as argument: - db_dump() - my_net_write() - net_write_command() - net_store_data() - DBUG_DUMP() - decimal2bin() & bin2decimal() - Changed my_compress() and my_uncompress() to use size_t. Changed one argument to my_uncompress() from a pointer to a value as we only return one value (makes function easier to use). - Changed type of 'pack_data' argument to packfrm() to avoid casts. - Changed in readfrm() and writefrom(), ha_discover and handler::discover() the type for argument 'frmdata' to uchar** to avoid casts. - Changed most Field functions to use uchar* instead of char* (reduced a lot of casts). - Changed field->val_xxx(xxx, new_ptr) to take const pointers. Other changes: - Removed a lot of not needed casts - Added a few new cast required by other changes - Added some cast to my_multi_malloc() arguments for safety (as string lengths needs to be uint, not size_t). - Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done explicitely as this conflict was often hided by casting the function to hash_get_key). - Changed some buffers to memory regions to uchar* to avoid casts. - Changed some string lengths from uint to size_t. - Changed field->ptr to be uchar* instead of char*. This allowed us to get rid of a lot of casts. - Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar - Include zlib.h in some files as we needed declaration of crc32() - Changed MY_FILE_ERROR to be (size_t) -1. - Changed many variables to hold the result of my_read() / my_write() to be size_t. This was needed to properly detect errors (which are returned as (size_t) -1). - Removed some very old VMS code - Changed packfrm()/unpackfrm() to not be depending on uint size (portability fix) - Removed windows specific code to restore cursor position as this causes slowdown on windows and we should not mix read() and pread() calls anyway as this is not thread safe. Updated function comment to reflect this. Changed function that depended on original behavior of my_pwrite() to itself restore the cursor position (one such case). - Added some missing checking of return value of malloc(). - Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow. - Changed type of table_def::m_size from my_size_t to ulong to reflect that m_size is the number of elements in the array, not a string/memory length. - Moved THD::max_row_length() to table.cc (as it's not depending on THD). Inlined max_row_length_blob() into this function. - More function comments - Fixed some compiler warnings when compiled without partitions. - Removed setting of LEX_STRING() arguments in declaration (portability fix). - Some trivial indentation/variable name changes. - Some trivial code simplifications: - Replaced some calls to alloc_root + memcpy to use strmake_root()/strdup_root(). - Changed some calls from memdup() to strmake() (Safety fix) - Simpler loops in client-simple.c
2007-05-10 11:59:39 +02:00
if (my_hash_insert(&table_def_cache, (uchar*) share))
{
free_table_share(share);
DBUG_RETURN(0); // return error
}
if (open_table_def(thd, share, db_flags))
{
*error= share->error;
(void) my_hash_delete(&table_def_cache, (uchar*) share);
DBUG_RETURN(0);
}
share->ref_count++; // Mark in use
DBUG_PRINT("exit", ("share: 0x%lx ref_count: %u",
(ulong) share, share->ref_count));
DBUG_RETURN(share);
found:
/*
We found an existing table definition. Return it if we didn't get
an error when reading the table definition from file.
*/
if (share->error)
{
/* Table definition contained an error */
open_table_error(share, share->error, share->open_errno, share->errarg);
DBUG_RETURN(0);
}
if (share->is_view && !(db_flags & OPEN_VIEW))
{
open_table_error(share, 1, ENOENT, 0);
DBUG_RETURN(0);
}
++share->ref_count;
if (share->ref_count == 1 && share->prev)
{
/*
Share was not used before and it was in the old_unused_share list
Unlink share from this list
*/
DBUG_PRINT("info", ("Unlinking from not used list"));
*share->prev= share->next;
share->next->prev= share->prev;
share->next= 0;
share->prev= 0;
}
/* Free cache if too big */
while (table_def_cache.records > table_def_size &&
oldest_unused_share->next)
my_hash_delete(&table_def_cache, (uchar*) oldest_unused_share);
DBUG_PRINT("exit", ("share: 0x%lx ref_count: %u",
(ulong) share, share->ref_count));
DBUG_RETURN(share);
}
/**
Get a table share. If it didn't exist, try creating it from engine
For arguments and return values, see get_table_share()
*/
static TABLE_SHARE *
get_table_share_with_discover(THD *thd, TABLE_LIST *table_list,
char *key, uint key_length,
uint db_flags, int *error,
my_hash_value_type hash_value)
2010-02-03 14:43:03 +01:00
{
TABLE_SHARE *share;
bool exists;
DBUG_ENTER("get_table_share_with_create");
2010-02-03 14:43:03 +01:00
share= get_table_share(thd, table_list, key, key_length, db_flags, error,
hash_value);
/*
If share is not NULL, we found an existing share.
If share is NULL, and there is no error, we're inside
pre-locking, which silences 'ER_NO_SUCH_TABLE' errors
with the intention to silently drop non-existing tables
from the pre-locking list. In this case we still need to try
auto-discover before returning a NULL share.
Or, we're inside SHOW CREATE VIEW, which
also installs a silencer for ER_NO_SUCH_TABLE error.
If share is NULL and the error is ER_NO_SUCH_TABLE, this is
the same as above, only that the error was not silenced by
pre-locking or SHOW CREATE VIEW.
In both these cases it won't harm to try to discover the
table.
Finally, if share is still NULL, it's a real error and we need
to abort.
@todo Rework alternative ways to deal with ER_NO_SUCH TABLE.
*/
if (share || (thd->is_error() && thd->stmt_da->sql_errno() != ER_NO_SUCH_TABLE))
DBUG_RETURN(share);
*error= 0;
/* Table didn't exist. Check if some engine can provide it */
if (ha_check_if_table_exists(thd, table_list->db, table_list->table_name,
&exists))
{
thd->clear_error();
/* Conventionally, the storage engine API does not report errors. */
my_error(ER_OUT_OF_RESOURCES, MYF(0));
}
else if (! exists)
{
/*
No such table in any engine.
Hide "Table doesn't exist" errors if the table belongs to a view.
The check for thd->is_error() is necessary to not push an
unwanted error in case the error was already silenced.
@todo Rework the alternative ways to deal with ER_NO_SUCH TABLE.
*/
if (thd->is_error())
{
if (table_list->parent_l)
{
thd->clear_error();
my_error(ER_WRONG_MRG_TABLE, MYF(0));
}
else if (table_list->belong_to_view)
{
TABLE_LIST *view= table_list->belong_to_view;
thd->clear_error();
my_error(ER_VIEW_INVALID, MYF(0),
view->view_db.str, view->view_name.str);
}
}
}
else
{
thd->clear_error();
*error= 7; /* Run auto-discover. */
}
DBUG_RETURN(NULL);
}
/**
Mark that we are not using table share anymore.
@param share Table share
If the share has no open tables and (we have done a refresh or
if we have already too many open table shares) then delete the
definition.
*/
void release_table_share(TABLE_SHARE *share)
{
DBUG_ENTER("release_table_share");
DBUG_PRINT("enter",
("share: 0x%lx table: %s.%s ref_count: %u version: %lu",
(ulong) share, share->db.str, share->table_name.str,
share->ref_count, share->version));
mysql_mutex_assert_owner(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
DBUG_ASSERT(share->ref_count);
if (!--share->ref_count)
{
if (share->has_old_version() || table_def_shutdown_in_progress)
my_hash_delete(&table_def_cache, (uchar*) share);
else
{
/* Link share last in used_table_share list */
DBUG_PRINT("info",("moving share to unused list"));
DBUG_ASSERT(share->next == 0);
share->prev= end_of_unused_share.prev;
*end_of_unused_share.prev= share;
end_of_unused_share.prev= &share->next;
share->next= &end_of_unused_share;
if (table_def_cache.records > table_def_size)
{
/* Delete the least used share to preserve LRU order. */
my_hash_delete(&table_def_cache, (uchar*) oldest_unused_share);
}
}
}
DBUG_VOID_RETURN;
}
/*
Check if table definition exits in cache
SYNOPSIS
get_cached_table_share()
db Database name
table_name Table name
RETURN
0 Not cached
# TABLE_SHARE for table
*/
TABLE_SHARE *get_cached_table_share(const char *db, const char *table_name)
{
char key[NAME_LEN*2+2];
TABLE_LIST table_list;
uint key_length;
mysql_mutex_assert_owner(&LOCK_open);
table_list.db= (char*) db;
table_list.table_name= (char*) table_name;
key_length= create_table_def_key((THD*) 0, key, &table_list, 0);
return (TABLE_SHARE*) my_hash_search(&table_def_cache,
(uchar*) key, key_length);
}
/*
Create a list for all open tables matching SQL expression
SYNOPSIS
list_open_tables()
thd Thread THD
wild SQL like expression
NOTES
One gets only a list of tables for which one has any kind of privilege.
db and table names are allocated in result struct, so one doesn't need
a lock on LOCK_open when traversing the return list.
RETURN VALUES
NULL Error (Probably OOM)
# Pointer to list of names of open tables.
*/
OPEN_TABLE_LIST *list_open_tables(THD *thd, const char *db, const char *wild)
{
int result = 0;
2001-07-11 09:36:22 +02:00
OPEN_TABLE_LIST **start_list, *open_list;
TABLE_LIST table_list;
DBUG_ENTER("list_open_tables");
2001-07-11 09:36:22 +02:00
mysql_mutex_lock(&LOCK_open);
bzero((char*) &table_list,sizeof(table_list));
2001-07-11 09:36:22 +02:00
start_list= &open_list;
open_list=0;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
for (uint idx=0 ; result == 0 && idx < table_def_cache.records; idx++)
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
TABLE_SHARE *share= (TABLE_SHARE *)my_hash_element(&table_def_cache, idx);
2001-07-11 09:36:22 +02:00
if (db && my_strcasecmp(system_charset_info, db, share->db.str))
continue;
if (wild && wild_compare(share->table_name.str, wild, 0))
continue;
2001-07-11 09:36:22 +02:00
/* Check if user has SELECT privilege for any column in the table */
table_list.db= share->db.str;
table_list.table_name= share->table_name.str;
2001-07-11 09:36:22 +02:00
table_list.grant.privilege=0;
2004-06-25 15:52:01 +02:00
if (check_table_access(thd,SELECT_ACL,&table_list, TRUE, 1, TRUE))
2001-07-11 09:36:22 +02:00
continue;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
2001-07-11 09:36:22 +02:00
if (!(*start_list = (OPEN_TABLE_LIST *)
sql_alloc(sizeof(**start_list)+share->table_cache_key.length)))
{
2001-07-11 09:36:22 +02:00
open_list=0; // Out of memory
break;
}
strmov((*start_list)->table=
strmov(((*start_list)->db= (char*) ((*start_list)+1)),
share->db.str)+1,
share->table_name.str);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
(*start_list)->in_use= 0;
I_P_List_iterator<TABLE, TABLE_share> it(share->used_tables);
while (it++)
++(*start_list)->in_use;
(*start_list)->locked= 0; /* Obsolete. */
2001-07-11 09:36:22 +02:00
start_list= &(*start_list)->next;
*start_list=0;
}
mysql_mutex_unlock(&LOCK_open);
2001-07-11 09:36:22 +02:00
DBUG_RETURN(open_list);
}
2000-07-31 21:29:14 +02:00
/*****************************************************************************
* Functions to free open table cache
****************************************************************************/
void intern_close_table(TABLE *table)
{ // Free all structures
DBUG_ENTER("intern_close_table");
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
DBUG_PRINT("tcache", ("table: '%s'.'%s' 0x%lx",
table->s ? table->s->db.str : "?",
table->s ? table->s->table_name.str : "?",
(long) table));
2000-07-31 21:29:14 +02:00
free_io_cache(table);
delete table->triggers;
if (table->file) // Not true if placeholder
(void) closefrm(table, 1); // close file
DBUG_VOID_RETURN;
2000-07-31 21:29:14 +02:00
}
/*
Remove table from the open table cache
SYNOPSIS
free_cache_entry()
table Table to remove
NOTE
We need to have a lock on LOCK_open when calling this
*/
2000-07-31 21:29:14 +02:00
static void free_cache_entry(TABLE *table)
{
DBUG_ENTER("free_cache_entry");
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/* This should be done before releasing table share. */
table_def_remove_table(table);
2000-07-31 21:29:14 +02:00
intern_close_table(table);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(table);
2000-07-31 21:29:14 +02:00
DBUG_VOID_RETURN;
}
/* Free resources allocated by filesort() and read_record() */
2000-07-31 21:29:14 +02:00
void free_io_cache(TABLE *table)
{
DBUG_ENTER("free_io_cache");
if (table->sort.io_cache)
2000-07-31 21:29:14 +02:00
{
close_cached_file(table->sort.io_cache);
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(table->sort.io_cache);
table->sort.io_cache=0;
2000-07-31 21:29:14 +02:00
}
DBUG_VOID_RETURN;
2000-07-31 21:29:14 +02:00
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/**
Auxiliary function which allows to kill delayed threads for
particular table identified by its share.
@param share Table share.
@pre Caller should have LOCK_open mutex.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
*/
static void kill_delayed_threads_for_table(TABLE_SHARE *share)
{
I_P_List_iterator<TABLE, TABLE_share> it(share->used_tables);
TABLE *tab;
2010-02-02 00:22:16 +01:00
mysql_mutex_assert_owner(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
while ((tab= it++))
{
THD *in_use= tab->in_use;
if ((in_use->system_thread & SYSTEM_THREAD_DELAYED_INSERT) &&
! in_use->killed)
{
in_use->killed= THD::KILL_CONNECTION;
2010-02-02 00:22:16 +01:00
mysql_mutex_lock(&in_use->mysys_var->mutex);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (in_use->mysys_var->current_cond)
{
2010-02-02 00:22:16 +01:00
mysql_mutex_lock(in_use->mysys_var->current_mutex);
mysql_cond_broadcast(in_use->mysys_var->current_cond);
mysql_mutex_unlock(in_use->mysys_var->current_mutex);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
2010-02-02 00:22:16 +01:00
mysql_mutex_unlock(&in_use->mysys_var->mutex);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
}
}
/*
Close all tables which aren't in use by any thread
@param thd Thread context
@param tables List of tables to remove from the cache
@param wait_for_refresh Wait for a impending flush
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
@param timeout Timeout for waiting for flush to be completed.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@note THD can be NULL, but then wait_for_refresh must be FALSE
and tables must be NULL.
@note When called as part of FLUSH TABLES WITH READ LOCK this function
ignores metadata locks held by other threads. In order to avoid
situation when FLUSH TABLES WITH READ LOCK sneaks in at the moment
when some write-locked table is being reopened (by FLUSH TABLES or
ALTER TABLE) we have to rely on additional global shared metadata
lock taken by thread trying to obtain global read lock.
*/
2000-07-31 21:29:14 +02:00
bool close_cached_tables(THD *thd, TABLE_LIST *tables,
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
bool wait_for_refresh, ulong timeout)
2000-07-31 21:29:14 +02:00
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
bool result= FALSE;
bool found= TRUE;
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
struct timespec abstime;
2000-07-31 21:29:14 +02:00
DBUG_ENTER("close_cached_tables");
DBUG_ASSERT(thd || (!wait_for_refresh && !tables));
2000-07-31 21:29:14 +02:00
mysql_mutex_lock(&LOCK_open);
if (!tables)
2000-07-31 21:29:14 +02:00
{
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
/*
Force close of all open tables.
Note that code in TABLE_SHARE::wait_for_old_version() assumes that
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
incrementing of refresh_version and removal of unused tables and
shares from TDC happens atomically under protection of LOCK_open,
or putting it another way that TDC does not contain old shares
which don't have any tables used.
*/
refresh_version++;
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
DBUG_PRINT("tcache", ("incremented global refresh_version to: %lu",
refresh_version));
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
kill_delayed_threads();
/*
Get rid of all unused TABLE and TABLE_SHARE instances. By doing
this we automatically close all tables which were marked as "old".
*/
while (unused_tables)
free_cache_entry(unused_tables);
/* Free table shares which were not freed implicitly by loop above. */
while (oldest_unused_share->next)
(void) my_hash_delete(&table_def_cache, (uchar*) oldest_unused_share);
2000-07-31 21:29:14 +02:00
}
else
2000-07-31 21:29:14 +02:00
{
bool found=0;
2004-07-16 00:15:55 +02:00
for (TABLE_LIST *table= tables; table; table= table->next_local)
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
TABLE_SHARE *share= get_cached_table_share(table->db, table->table_name);
if (share)
{
kill_delayed_threads_for_table(share);
/* tdc_remove_table() also sets TABLE_SHARE::version to 0. */
tdc_remove_table(thd, TDC_RT_REMOVE_UNUSED, table->db,
table->table_name, TRUE);
found=1;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
}
if (!found)
wait_for_refresh=0; // Nothing to wait for
2000-07-31 21:29:14 +02:00
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!wait_for_refresh)
DBUG_RETURN(result);
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
set_timespec(abstime, timeout);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (thd->locked_tables_mode)
2000-07-31 21:29:14 +02:00
{
/*
If we are under LOCK TABLES, we need to reopen the tables without
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
opening a door for any concurrent threads to sneak in and get
lock on our tables. To achieve this we use exclusive metadata
locks.
2000-07-31 21:29:14 +02:00
*/
TABLE_LIST *tables_to_reopen= (tables ? tables :
thd->locked_tables_list.locked_tables());
/* Close open HANLER instances to avoid self-deadlock. */
mysql_ha_flush_tables(thd, tables_to_reopen);
for (TABLE_LIST *table_list= tables_to_reopen; table_list;
table_list= table_list->next_global)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
/* A check that the table was locked for write is done by the caller. */
TABLE *table= find_table_for_mdl_upgrade(thd, table_list->db,
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
table_list->table_name, TRUE);
/* May return NULL if this table has already been closed via an alias. */
if (! table)
continue;
if (wait_while_table_is_used(thd, table, HA_EXTRA_FORCE_REOPEN))
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
result= TRUE;
goto err_with_reopen;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
close_all_tables_for_name(thd, table->s, FALSE);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
}
/* Wait until all threads have closed all the tables we are flushing. */
DBUG_PRINT("info", ("Waiting for other threads to close their open tables"));
2000-07-31 21:29:14 +02:00
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
while (found && ! thd->killed)
{
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
TABLE_SHARE *share;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
found= FALSE;
/*
To a self-deadlock or deadlocks with other FLUSH threads
waiting on our open HANDLERs, we have to flush them.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
*/
mysql_ha_flush(thd);
DEBUG_SYNC(thd, "after_flush_unlock");
2010-02-02 00:22:16 +01:00
mysql_mutex_lock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!tables)
2000-07-31 21:29:14 +02:00
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
for (uint idx=0 ; idx < table_def_cache.records ; idx++)
2000-07-31 21:29:14 +02:00
{
share= (TABLE_SHARE*) my_hash_element(&table_def_cache, idx);
if (share->has_old_version())
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
found= TRUE;
break;
}
}
}
else
{
for (TABLE_LIST *table= tables; table; table= table->next_local)
{
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
share= get_cached_table_share(table->db, table->table_name);
if (share && share->has_old_version())
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
found= TRUE;
break;
}
2000-07-31 21:29:14 +02:00
}
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (found)
{
/*
The method below temporarily unlocks LOCK_open and frees
share's memory.
*/
if (share->wait_for_old_version(thd, &abstime,
MDL_wait_for_subgraph::DEADLOCK_WEIGHT_DDL))
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
{
mysql_mutex_unlock(&LOCK_open);
result= TRUE;
goto err_with_reopen;
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
err_with_reopen:
if (thd->locked_tables_mode)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
2000-07-31 21:29:14 +02:00
/*
No other thread has the locked tables open; reopen them and get the
old locks. This should always succeed (unless some external process
has removed the tables)
*/
thd->locked_tables_list.reopen_tables(thd);
/*
Since downgrade_exclusive_lock() won't do anything with shared
metadata lock it is much simpler to go through all open tables rather
than picking only those tables that were flushed.
*/
for (TABLE *tab= thd->open_tables; tab; tab= tab->next)
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
tab->mdl_ticket->downgrade_exclusive_lock(MDL_SHARED_NO_READ_WRITE);
2000-07-31 21:29:14 +02:00
}
DBUG_RETURN(result);
}
2010-08-12 18:44:41 +02:00
/**
Close all tables which match specified connection string or
if specified string is NULL, then any table with a connection string.
*/
2010-08-12 18:44:41 +02:00
bool close_cached_connection_tables(THD *thd, LEX_STRING *connection)
{
uint idx;
TABLE_LIST tmp, *tables= NULL;
bool result= FALSE;
DBUG_ENTER("close_cached_connections");
DBUG_ASSERT(thd);
bzero(&tmp, sizeof(TABLE_LIST));
mysql_mutex_lock(&LOCK_open);
for (idx= 0; idx < table_def_cache.records; idx++)
{
TABLE_SHARE *share= (TABLE_SHARE *) my_hash_element(&table_def_cache, idx);
/* Ignore if table is not open or does not have a connect_string */
if (!share->connect_string.length || !share->ref_count)
continue;
/* Compare the connection string */
if (connection &&
(connection->length > share->connect_string.length ||
(connection->length < share->connect_string.length &&
(share->connect_string.str[connection->length] != '/' &&
share->connect_string.str[connection->length] != '\\')) ||
strncasecmp(connection->str, share->connect_string.str,
connection->length)))
continue;
/* close_cached_tables() only uses these elements */
tmp.db= share->db.str;
tmp.table_name= share->table_name.str;
tmp.next_local= tables;
tables= (TABLE_LIST *) memdup_root(thd->mem_root, (char*)&tmp,
sizeof(TABLE_LIST));
}
mysql_mutex_unlock(&LOCK_open);
if (tables)
result= close_cached_tables(thd, tables, FALSE, LONG_TIMEOUT);
DBUG_RETURN(result);
}
/**
Mark all temporary tables which were used by the current statement or
substatement as free for reuse, but only if the query_id can be cleared.
@param thd thread context
@remark For temp tables associated with a open SQL HANDLER the query_id
is not reset until the HANDLER is closed.
*/
static void mark_temp_tables_as_free_for_reuse(THD *thd)
{
for (TABLE *table= thd->temporary_tables ; table ; table= table->next)
{
if ((table->query_id == thd->query_id) && ! table->open_by_handler)
A prerequisite patch for the fix for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". Introduce a notion of a sentinel to MDL_context. A sentinel is a ticket that separates all tickets in the context into two groups: before and after it. Currently we can have (and need) only one designated sentinel -- it separates all locks taken by LOCK TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK and all other locks, which must be released at COMMIT or ROLLBACK. The tricky part is maintaining the sentinel up to date when someone release its corresponding ticket. This can happen, e.g. if someone issues DROP TABLE under LOCK TABLES (generally, see all calls to release_all_locks_for_name()). MDL_context::release_ticket() is modified to take care of it. ****** A fix and a test case for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". An attempt to mix HANDLER SQL statements, which are transaction- agnostic, an open multi-statement transaction, and DDL against the involved tables (in a concurrent connection) could lead to a deadlock. The deadlock would occur when HANDLER OPEN or HANDLER READ would have to wait on a conflicting metadata lock. If the connection that issued HANDLER statement also had other metadata locks (say, acquired in scope of a transaction), a classical deadlock situation of mutual wait could occur. Incompatible change: entering LOCK TABLES mode automatically closes all open HANDLERs in the current connection. Incompatible change: previously an attempt to wait on a lock in a connection that has an open HANDLER statement could wait indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK is produced. The idea of the fix is to merge thd->handler_mdl_context with the main mdl_context of the connection, used for transactional locks. This makes deadlock detection possible, since all waits with locks are "visible" and available to analysis in a single MDL context of the connection. Since HANDLER locks and transactional locks have a different life cycle -- HANDLERs are explicitly open and closed, and so are HANDLER locks, explicitly acquired and released, whereas transactional locks "accumulate" till the end of a transaction and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT, a concept of "sentinel" was introduced to MDL_context. All locks, HANDLER and others, reside in the same linked list. However, a selected element of the list separates locks with different life cycle. HANDLER locks always reside at the end of the list, after the sentinel. Transactional locks are prepended to the beginning of the list, before the sentinel. Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only release those locks that reside before the sentinel. HANDLER locks must be released explicitly as part of HANDLER CLOSE statement, or an implicit close. The same approach with sentinel is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES statement has never worked together, the implementation is made simple and only maintains one sentinel, which is used either for HANDLER locks, or for LOCK TABLES locks.
2009-12-22 17:09:15 +01:00
mark_tmp_table_for_reuse(table);
}
}
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
A prerequisite patch for the fix for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". Introduce a notion of a sentinel to MDL_context. A sentinel is a ticket that separates all tickets in the context into two groups: before and after it. Currently we can have (and need) only one designated sentinel -- it separates all locks taken by LOCK TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK and all other locks, which must be released at COMMIT or ROLLBACK. The tricky part is maintaining the sentinel up to date when someone release its corresponding ticket. This can happen, e.g. if someone issues DROP TABLE under LOCK TABLES (generally, see all calls to release_all_locks_for_name()). MDL_context::release_ticket() is modified to take care of it. ****** A fix and a test case for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". An attempt to mix HANDLER SQL statements, which are transaction- agnostic, an open multi-statement transaction, and DDL against the involved tables (in a concurrent connection) could lead to a deadlock. The deadlock would occur when HANDLER OPEN or HANDLER READ would have to wait on a conflicting metadata lock. If the connection that issued HANDLER statement also had other metadata locks (say, acquired in scope of a transaction), a classical deadlock situation of mutual wait could occur. Incompatible change: entering LOCK TABLES mode automatically closes all open HANDLERs in the current connection. Incompatible change: previously an attempt to wait on a lock in a connection that has an open HANDLER statement could wait indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK is produced. The idea of the fix is to merge thd->handler_mdl_context with the main mdl_context of the connection, used for transactional locks. This makes deadlock detection possible, since all waits with locks are "visible" and available to analysis in a single MDL context of the connection. Since HANDLER locks and transactional locks have a different life cycle -- HANDLERs are explicitly open and closed, and so are HANDLER locks, explicitly acquired and released, whereas transactional locks "accumulate" till the end of a transaction and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT, a concept of "sentinel" was introduced to MDL_context. All locks, HANDLER and others, reside in the same linked list. However, a selected element of the list separates locks with different life cycle. HANDLER locks always reside at the end of the list, after the sentinel. Transactional locks are prepended to the beginning of the list, before the sentinel. Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only release those locks that reside before the sentinel. HANDLER locks must be released explicitly as part of HANDLER CLOSE statement, or an implicit close. The same approach with sentinel is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES statement has never worked together, the implementation is made simple and only maintains one sentinel, which is used either for HANDLER locks, or for LOCK TABLES locks.
2009-12-22 17:09:15 +01:00
/**
Reset a single temporary table.
Effectively this "closes" one temporary table,
in a session.
@param table Temporary table.
*/
void mark_tmp_table_for_reuse(TABLE *table)
{
DBUG_ASSERT(table->s->tmp_table);
table->query_id= 0;
table->file->ha_reset();
/* Detach temporary MERGE children from temporary parent. */
DBUG_ASSERT(table->file);
table->file->extra(HA_EXTRA_DETACH_CHILDREN);
/*
Reset temporary table lock type to it's default value (TL_WRITE).
Statements such as INSERT INTO .. SELECT FROM tmp, CREATE TABLE
.. SELECT FROM tmp and UPDATE may under some circumstances modify
the lock type of the tables participating in the statement. This
isn't a problem for non-temporary tables since their lock type is
reset at every open, but the same does not occur for temporary
tables for historical reasons.
Furthermore, the lock type of temporary tables is not really that
important because they can only be used by one query at a time and
not even twice in a query -- a temporary table is represented by
only one TABLE object. Nonetheless, it's safer from a maintenance
point of view to reset the lock type of this singleton TABLE object
as to not cause problems when the table is reused.
Even under LOCK TABLES mode its okay to reset the lock type as
LOCK TABLES is allowed (but ignored) for a temporary table.
*/
table->reginfo.lock_type= TL_WRITE;
}
/*
Mark all tables in the list which were used by current substatement
as free for reuse.
SYNOPSIS
mark_used_tables_as_free_for_reuse()
thd - thread context
table - head of the list of tables
DESCRIPTION
Marks all tables in the list which were used by current substatement
(they are marked by its query_id) as free for reuse.
NOTE
The reason we reset query_id is that it's not enough to just test
if table->query_id != thd->query_id to know if a table is in use.
For example
SELECT f1_that_uses_t1() FROM t1;
In f1_that_uses_t1() we will see one instance of t1 where query_id is
set to query_id of original query.
*/
static void mark_used_tables_as_free_for_reuse(THD *thd, TABLE *table)
{
for (; table ; table= table->next)
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
{
DBUG_ASSERT(table->pos_in_locked_tables == NULL ||
table->pos_in_locked_tables->table == table);
if (table->query_id == thd->query_id)
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
{
table->query_id= 0;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
table->file->ha_reset();
}
}
}
/**
Auxiliary function to close all tables in the open_tables list.
@param thd Thread context.
@remark It should not ordinarily be called directly.
*/
static void close_open_tables(THD *thd)
{
mysql_mutex_assert_not_owner(&LOCK_open);
DBUG_PRINT("info", ("thd->open_tables: 0x%lx", (long) thd->open_tables));
while (thd->open_tables)
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
(void) close_thread_table(thd, &thd->open_tables);
}
/**
Close all open instances of the table but keep the MDL lock.
Works both under LOCK TABLES and in the normal mode.
Removes all closed instances of the table from the table cache.
@param thd thread handle
@param[in] share table share, but is just a handy way to
access the table cache key
@param[in] remove_from_locked_tables
TRUE if the table is being dropped or renamed.
In that case the documented behaviour is to
implicitly remove the table from LOCK TABLES
list.
@pre Must be called with an X MDL lock on the table.
*/
void
close_all_tables_for_name(THD *thd, TABLE_SHARE *share,
bool remove_from_locked_tables)
{
char key[MAX_DBKEY_LENGTH];
uint key_length= share->table_cache_key.length;
const char *db= key;
const char *table_name= db + share->db.length + 1;
memcpy(key, share->table_cache_key.str, key_length);
2010-02-02 00:22:16 +01:00
mysql_mutex_assert_not_owner(&LOCK_open);
for (TABLE **prev= &thd->open_tables; *prev; )
{
TABLE *table= *prev;
if (table->s->table_cache_key.length == key_length &&
!memcmp(table->s->table_cache_key.str, key, key_length))
{
thd->locked_tables_list.unlink_from_list(thd,
table->pos_in_locked_tables,
remove_from_locked_tables);
/*
Does nothing if the table is not locked.
This allows one to use this function after a table
has been unlocked, e.g. in partition management.
*/
mysql_lock_remove(thd, thd->lock, table);
/* Inform handler that table will be dropped after close */
if (table->db_stat) /* Not true for partitioned tables. */
table->file->extra(HA_EXTRA_PREPARE_FOR_DROP);
close_thread_table(thd, prev);
}
else
{
/* Step to next entry in open_tables list. */
prev= &table->next;
}
}
/* Remove the table share from the cache. */
tdc_remove_table(thd, TDC_RT_REMOVE_ALL, db, table_name,
FALSE);
}
/*
Close all tables used by the current substatement, or all tables
used by this thread if we are on the upper level.
2000-07-31 21:29:14 +02:00
SYNOPSIS
close_thread_tables()
thd Thread handler
IMPLEMENTATION
Unlocks tables and frees derived tables.
Put all normal tables used by thread in free list.
It will only close/mark as free for reuse tables opened by this
substatement, it will also check if we are closing tables after
execution of complete query (i.e. we are on upper level) and will
leave prelocked mode if needed.
*/
2000-07-31 21:29:14 +02:00
void close_thread_tables(THD *thd)
2000-07-31 21:29:14 +02:00
{
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
TABLE *table;
2000-07-31 21:29:14 +02:00
DBUG_ENTER("close_thread_tables");
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
#ifdef EXTRA_DEBUG
DBUG_PRINT("tcache", ("open tables:"));
for (table= thd->open_tables; table; table= table->next)
DBUG_PRINT("tcache", ("table: '%s'.'%s' 0x%lx", table->s->db.str,
table->s->table_name.str, (long) table));
#endif
#if defined(ENABLED_DEBUG_SYNC)
/* debug_sync may not be initialized for some slave threads */
if (thd->debug_sync_control)
DEBUG_SYNC(thd, "before_close_thread_tables");
#endif
DBUG_ASSERT(thd->transaction.stmt.is_empty() || thd->in_sub_stmt ||
(thd->state_flags & Open_tables_state::BACKUPS_AVAIL));
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
/* Detach MERGE children after every statement. Even under LOCK TABLES. */
for (table= thd->open_tables; table; table= table->next)
{
/* Table might be in use by some outer statement. */
DBUG_PRINT("tcache", ("table: '%s' query_id: %lu",
table->s->table_name.str, (ulong) table->query_id));
if (thd->locked_tables_mode <= LTM_LOCK_TABLES ||
table->query_id == thd->query_id)
{
DBUG_ASSERT(table->file);
table->file->extra(HA_EXTRA_DETACH_CHILDREN);
}
}
/*
We are assuming here that thd->derived_tables contains ONLY derived
tables for this substatement. i.e. instead of approach which uses
query_id matching for determining which of the derived tables belong
to this substatement we rely on the ability of substatements to
save/restore thd->derived_tables during their execution.
TODO: Probably even better approach is to simply associate list of
derived tables with (sub-)statement instead of thread and destroy
them at the end of its execution.
*/
if (thd->derived_tables)
{
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
TABLE *next;
/*
2005-01-16 13:16:23 +01:00
Close all derived tables generated in queries like
SELECT * FROM (SELECT * FROM t1)
*/
for (table= thd->derived_tables ; table ; table= next)
{
next= table->next;
free_tmp_table(thd, table);
}
thd->derived_tables= 0;
}
/*
Mark all temporary tables used by this statement as free for reuse.
*/
mark_temp_tables_as_free_for_reuse(thd);
if (thd->locked_tables_mode)
{
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
/* Ensure we are calling ha_reset() for all used tables */
mark_used_tables_as_free_for_reuse(thd, thd->open_tables);
/*
We are under simple LOCK TABLES or we're inside a sub-statement
of a prelocked statement, so should not do anything else.
Note that even if we are in LTM_LOCK_TABLES mode and statement
requires prelocking (e.g. when we are closing tables after
failing ot "open" all tables required for statement execution)
we will exit this function a few lines below.
*/
if (! thd->lex->requires_prelocking())
DBUG_VOID_RETURN;
/*
We are in the top-level statement of a prelocked statement,
so we have to leave the prelocked mode now with doing implicit
UNLOCK TABLES if needed.
*/
if (thd->locked_tables_mode == LTM_PRELOCKED_UNDER_LOCK_TABLES)
thd->locked_tables_mode= LTM_LOCK_TABLES;
if (thd->locked_tables_mode == LTM_LOCK_TABLES)
DBUG_VOID_RETURN;
A prerequisite patch for the fix for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". Introduce a notion of a sentinel to MDL_context. A sentinel is a ticket that separates all tickets in the context into two groups: before and after it. Currently we can have (and need) only one designated sentinel -- it separates all locks taken by LOCK TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK and all other locks, which must be released at COMMIT or ROLLBACK. The tricky part is maintaining the sentinel up to date when someone release its corresponding ticket. This can happen, e.g. if someone issues DROP TABLE under LOCK TABLES (generally, see all calls to release_all_locks_for_name()). MDL_context::release_ticket() is modified to take care of it. ****** A fix and a test case for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". An attempt to mix HANDLER SQL statements, which are transaction- agnostic, an open multi-statement transaction, and DDL against the involved tables (in a concurrent connection) could lead to a deadlock. The deadlock would occur when HANDLER OPEN or HANDLER READ would have to wait on a conflicting metadata lock. If the connection that issued HANDLER statement also had other metadata locks (say, acquired in scope of a transaction), a classical deadlock situation of mutual wait could occur. Incompatible change: entering LOCK TABLES mode automatically closes all open HANDLERs in the current connection. Incompatible change: previously an attempt to wait on a lock in a connection that has an open HANDLER statement could wait indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK is produced. The idea of the fix is to merge thd->handler_mdl_context with the main mdl_context of the connection, used for transactional locks. This makes deadlock detection possible, since all waits with locks are "visible" and available to analysis in a single MDL context of the connection. Since HANDLER locks and transactional locks have a different life cycle -- HANDLERs are explicitly open and closed, and so are HANDLER locks, explicitly acquired and released, whereas transactional locks "accumulate" till the end of a transaction and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT, a concept of "sentinel" was introduced to MDL_context. All locks, HANDLER and others, reside in the same linked list. However, a selected element of the list separates locks with different life cycle. HANDLER locks always reside at the end of the list, after the sentinel. Transactional locks are prepended to the beginning of the list, before the sentinel. Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only release those locks that reside before the sentinel. HANDLER locks must be released explicitly as part of HANDLER CLOSE statement, or an implicit close. The same approach with sentinel is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES statement has never worked together, the implementation is made simple and only maintains one sentinel, which is used either for HANDLER locks, or for LOCK TABLES locks.
2009-12-22 17:09:15 +01:00
thd->leave_locked_tables_mode();
/* Fallthrough */
}
2000-07-31 21:29:14 +02:00
if (thd->lock)
{
/*
For RBR we flush the pending event just before we unlock all the
tables. This means that we are at the end of a topmost
statement, so we ensure that the STMT_END_F flag is set on the
pending event. For statements that are *inside* stored
functions, the pending event will not be flushed: that will be
handled either before writing a query log event (inside
binlog_query()) or when preparing a pending event.
*/
(void)thd->binlog_flush_pending_rows_event(TRUE);
mysql_unlock_tables(thd, thd->lock);
thd->lock=0;
2000-07-31 21:29:14 +02:00
}
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
/*
Closing a MERGE child before the parent would be fatal if the
other thread tries to abort the MERGE lock in between.
*/
if (thd->open_tables)
close_open_tables(thd);
2000-07-31 21:29:14 +02:00
DBUG_VOID_RETURN;
}
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
2001-07-11 09:36:22 +02:00
/* move one table to free list */
bool close_thread_table(THD *thd, TABLE **table_ptr)
{
bool found_old_table= 0;
TABLE *table= *table_ptr;
DBUG_ENTER("close_thread_table");
DBUG_ASSERT(table->key_read == 0);
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
DBUG_ASSERT(!table->file || table->file->inited == handler::NONE);
mysql_mutex_assert_not_owner(&LOCK_open);
/*
The metadata lock must be released after giving back
the table to the table cache.
*/
DBUG_ASSERT(thd->mdl_context.is_lock_owner(MDL_key::TABLE,
table->s->db.str,
table->s->table_name.str,
MDL_SHARED));
table->mdl_ticket= NULL;
2001-07-11 09:36:22 +02:00
mysql_mutex_lock(&thd->LOCK_thd_data);
2001-07-11 09:36:22 +02:00
*table_ptr=table->next;
mysql_mutex_unlock(&thd->LOCK_thd_data);
if (! table->needs_reopen())
{
/* Avoid having MERGE tables with attached children in unused_tables. */
table->file->extra(HA_EXTRA_DETACH_CHILDREN);
/* Free memory and reset for next loop. */
free_field_buffers_larger_than(table, MAX_TDC_BLOB_SIZE);
table->file->ha_reset();
}
mysql_mutex_lock(&LOCK_open);
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
if (table->s->has_old_version() || table->needs_reopen() ||
table_def_shutdown_in_progress)
2001-07-11 09:36:22 +02:00
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
free_cache_entry(table);
found_old_table= 1;
2001-07-11 09:36:22 +02:00
}
else
{
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
DBUG_ASSERT(table->file);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
table_def_unuse_table(table);
/*
We free the least used table, not the subject table,
to keep the LRU order.
*/
if (table_cache_count > table_cache_size)
free_cache_entry(unused_tables);
2001-07-11 09:36:22 +02:00
}
mysql_mutex_unlock(&LOCK_open);
2001-07-11 09:36:22 +02:00
DBUG_RETURN(found_old_table);
}
2000-07-31 21:29:14 +02:00
/* close_temporary_tables' internal, 4 is due to uint4korr definition */
static inline uint tmpkeyval(THD *thd, TABLE *table)
{
return uint4korr(table->s->table_cache_key.str + table->s->table_cache_key.length - 4);
}
/*
Close all temporary tables created by 'CREATE TEMPORARY TABLE' for thread
creates one DROP TEMPORARY TABLE binlog event for each pseudo-thread
*/
2000-07-31 21:29:14 +02:00
bool close_temporary_tables(THD *thd)
2000-07-31 21:29:14 +02:00
{
DBUG_ENTER("close_temporary_tables");
TABLE *table;
2009-06-17 16:56:44 +02:00
TABLE *next= NULL;
TABLE *prev_table;
WL#4738 streamline/simplify @@variable creation process Bug#16565 mysqld --help --verbose does not order variablesBug#20413 sql_slave_skip_counter is not shown in show variables Bug#20415 Output of mysqld --help --verbose is incomplete Bug#25430 variable not found in SELECT @@global.ft_max_word_len; Bug#32902 plugin variables don't know their names Bug#34599 MySQLD Option and Variable Reference need to be consistent in formatting! Bug#34829 No default value for variable and setting default does not raise error Bug#34834 ? Is accepted as a valid sql mode Bug#34878 Few variables have default value according to documentation but error occurs Bug#34883 ft_boolean_syntax cant be assigned from user variable to global var. Bug#37187 `INFORMATION_SCHEMA`.`GLOBAL_VARIABLES`: inconsistent status Bug#40988 log_output_basic.test succeeded though syntactically false. Bug#41010 enum-style command-line options are not honoured (maria.maria-recover fails) Bug#42103 Setting key_buffer_size to a negative value may lead to very large allocations Bug#44691 Some plugins configured as MYSQL_PLUGIN_MANDATORY in can be disabled Bug#44797 plugins w/o command-line options have no disabling option in --help Bug#46314 string system variables don't support expressions Bug#46470 sys_vars.max_binlog_cache_size_basic_32 is broken Bug#46586 When using the plugin interface the type "set" for options caused a crash. Bug#47212 Crash in DBUG_PRINT in mysqltest.cc when trying to print octal number Bug#48758 mysqltest crashes on sys_vars.collation_server_basic in gcov builds Bug#49417 some complaints about mysqld --help --verbose output Bug#49540 DEFAULT value of binlog_format isn't the default value Bug#49640 ambiguous option '--skip-skip-myisam' (double skip prefix) Bug#49644 init_connect and \0 Bug#49645 init_slave and multi-byte characters Bug#49646 mysql --show-warnings crashes when server dies
2009-12-22 10:35:56 +01:00
/* Assume thd->variables.option_bits has OPTION_QUOTE_SHOW_CREATE */
bool was_quote_show= TRUE;
bool error= 0;
if (!thd->temporary_tables)
DBUG_RETURN(FALSE);
if (!mysql_bin_log.is_open())
{
TABLE *tmp_next;
for (table= thd->temporary_tables; table; table= tmp_next)
{
tmp_next= table->next;
close_temporary(table, 1, 1);
}
thd->temporary_tables= 0;
DBUG_RETURN(FALSE);
}
/* Better add "if exists", in case a RESET MASTER has been done */
const char stub[]= "DROP /*!40005 TEMPORARY */ TABLE IF EXISTS ";
uint stub_len= sizeof(stub) - 1;
char buf[256];
String s_query= String(buf, sizeof(buf), system_charset_info);
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
bool found_user_tables= FALSE;
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
memcpy(buf, stub, stub_len);
/*
Insertion sort of temp tables by pseudo_thread_id to build ordered list
of sublists of equal pseudo_thread_id
*/
for (prev_table= thd->temporary_tables, table= prev_table->next;
table;
prev_table= table, table= table->next)
{
TABLE *prev_sorted /* same as for prev_table */, *sorted;
if (is_user_table(table))
{
if (!found_user_tables)
found_user_tables= true;
for (prev_sorted= NULL, sorted= thd->temporary_tables; sorted != table;
prev_sorted= sorted, sorted= sorted->next)
{
if (!is_user_table(sorted) ||
tmpkeyval(thd, sorted) > tmpkeyval(thd, table))
{
/* move into the sorted part of the list from the unsorted */
prev_table->next= table->next;
table->next= sorted;
if (prev_sorted)
{
prev_sorted->next= table;
}
else
{
thd->temporary_tables= table;
}
table= prev_table;
break;
}
}
}
}
/* We always quote db,table names though it is slight overkill */
if (found_user_tables &&
WL#4738 streamline/simplify @@variable creation process Bug#16565 mysqld --help --verbose does not order variablesBug#20413 sql_slave_skip_counter is not shown in show variables Bug#20415 Output of mysqld --help --verbose is incomplete Bug#25430 variable not found in SELECT @@global.ft_max_word_len; Bug#32902 plugin variables don't know their names Bug#34599 MySQLD Option and Variable Reference need to be consistent in formatting! Bug#34829 No default value for variable and setting default does not raise error Bug#34834 ? Is accepted as a valid sql mode Bug#34878 Few variables have default value according to documentation but error occurs Bug#34883 ft_boolean_syntax cant be assigned from user variable to global var. Bug#37187 `INFORMATION_SCHEMA`.`GLOBAL_VARIABLES`: inconsistent status Bug#40988 log_output_basic.test succeeded though syntactically false. Bug#41010 enum-style command-line options are not honoured (maria.maria-recover fails) Bug#42103 Setting key_buffer_size to a negative value may lead to very large allocations Bug#44691 Some plugins configured as MYSQL_PLUGIN_MANDATORY in can be disabled Bug#44797 plugins w/o command-line options have no disabling option in --help Bug#46314 string system variables don't support expressions Bug#46470 sys_vars.max_binlog_cache_size_basic_32 is broken Bug#46586 When using the plugin interface the type "set" for options caused a crash. Bug#47212 Crash in DBUG_PRINT in mysqltest.cc when trying to print octal number Bug#48758 mysqltest crashes on sys_vars.collation_server_basic in gcov builds Bug#49417 some complaints about mysqld --help --verbose output Bug#49540 DEFAULT value of binlog_format isn't the default value Bug#49640 ambiguous option '--skip-skip-myisam' (double skip prefix) Bug#49644 init_connect and \0 Bug#49645 init_slave and multi-byte characters Bug#49646 mysql --show-warnings crashes when server dies
2009-12-22 10:35:56 +01:00
!(was_quote_show= test(thd->variables.option_bits & OPTION_QUOTE_SHOW_CREATE)))
{
WL#4738 streamline/simplify @@variable creation process Bug#16565 mysqld --help --verbose does not order variablesBug#20413 sql_slave_skip_counter is not shown in show variables Bug#20415 Output of mysqld --help --verbose is incomplete Bug#25430 variable not found in SELECT @@global.ft_max_word_len; Bug#32902 plugin variables don't know their names Bug#34599 MySQLD Option and Variable Reference need to be consistent in formatting! Bug#34829 No default value for variable and setting default does not raise error Bug#34834 ? Is accepted as a valid sql mode Bug#34878 Few variables have default value according to documentation but error occurs Bug#34883 ft_boolean_syntax cant be assigned from user variable to global var. Bug#37187 `INFORMATION_SCHEMA`.`GLOBAL_VARIABLES`: inconsistent status Bug#40988 log_output_basic.test succeeded though syntactically false. Bug#41010 enum-style command-line options are not honoured (maria.maria-recover fails) Bug#42103 Setting key_buffer_size to a negative value may lead to very large allocations Bug#44691 Some plugins configured as MYSQL_PLUGIN_MANDATORY in can be disabled Bug#44797 plugins w/o command-line options have no disabling option in --help Bug#46314 string system variables don't support expressions Bug#46470 sys_vars.max_binlog_cache_size_basic_32 is broken Bug#46586 When using the plugin interface the type "set" for options caused a crash. Bug#47212 Crash in DBUG_PRINT in mysqltest.cc when trying to print octal number Bug#48758 mysqltest crashes on sys_vars.collation_server_basic in gcov builds Bug#49417 some complaints about mysqld --help --verbose output Bug#49540 DEFAULT value of binlog_format isn't the default value Bug#49640 ambiguous option '--skip-skip-myisam' (double skip prefix) Bug#49644 init_connect and \0 Bug#49645 init_slave and multi-byte characters Bug#49646 mysql --show-warnings crashes when server dies
2009-12-22 10:35:56 +01:00
thd->variables.option_bits |= OPTION_QUOTE_SHOW_CREATE;
}
/* scan sorted tmps to generate sequence of DROP */
for (table= thd->temporary_tables; table; table= next)
2000-07-31 21:29:14 +02:00
{
if (is_user_table(table))
{
bool save_thread_specific_used= thd->thread_specific_used;
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
my_thread_id save_pseudo_thread_id= thd->variables.pseudo_thread_id;
/* Set pseudo_thread_id to be that of the processed table */
thd->variables.pseudo_thread_id= tmpkeyval(thd, table);
String db;
db.append(table->s->db.str);
/* Loop forward through all tables that belong to a common database
within the sublist of common pseudo_thread_id to create single
DROP query
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
*/
for (s_query.length(stub_len);
table && is_user_table(table) &&
tmpkeyval(thd, table) == thd->variables.pseudo_thread_id &&
table->s->db.length == db.length() &&
strcmp(table->s->db.str, db.ptr()) == 0;
table= next)
{
/*
We are going to add ` around the table names and possible more
due to special characters
*/
append_identifier(thd, &s_query, table->s->table_name.str,
strlen(table->s->table_name.str));
s_query.append(',');
next= table->next;
close_temporary(table, 1, 1);
}
thd->clear_error();
CHARSET_INFO *cs_save= thd->variables.character_set_client;
thd->variables.character_set_client= system_charset_info;
thd->thread_specific_used= TRUE;
Query_log_event qinfo(thd, s_query.ptr(),
s_query.length() - 1 /* to remove trailing ',' */,
FALSE, TRUE, FALSE, 0);
qinfo.db= db.ptr();
qinfo.db_len= db.length();
thd->variables.character_set_client= cs_save;
thd->stmt_da->can_overwrite_status= TRUE;
if ((error= (mysql_bin_log.write(&qinfo) || error)))
{
/*
If we're here following THD::cleanup, thence the connection
has been closed already. So lets print a message to the
error log instead of pushing yet another error into the
stmt_da.
Also, we keep the error flag so that we propagate the error
up in the stack. This way, if we're the SQL thread we notice
that close_temporary_tables failed. (Actually, the SQL
thread only calls close_temporary_tables while applying old
Start_log_event_v3 events.)
*/
sql_print_error("Failed to write the DROP statement for "
"temporary tables to binary log");
}
thd->stmt_da->can_overwrite_status= FALSE;
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
thd->variables.pseudo_thread_id= save_pseudo_thread_id;
thd->thread_specific_used= save_thread_specific_used;
}
else
{
next= table->next;
close_temporary(table, 1, 1);
}
2005-02-16 23:23:20 +01:00
}
if (!was_quote_show)
WL#4738 streamline/simplify @@variable creation process Bug#16565 mysqld --help --verbose does not order variablesBug#20413 sql_slave_skip_counter is not shown in show variables Bug#20415 Output of mysqld --help --verbose is incomplete Bug#25430 variable not found in SELECT @@global.ft_max_word_len; Bug#32902 plugin variables don't know their names Bug#34599 MySQLD Option and Variable Reference need to be consistent in formatting! Bug#34829 No default value for variable and setting default does not raise error Bug#34834 ? Is accepted as a valid sql mode Bug#34878 Few variables have default value according to documentation but error occurs Bug#34883 ft_boolean_syntax cant be assigned from user variable to global var. Bug#37187 `INFORMATION_SCHEMA`.`GLOBAL_VARIABLES`: inconsistent status Bug#40988 log_output_basic.test succeeded though syntactically false. Bug#41010 enum-style command-line options are not honoured (maria.maria-recover fails) Bug#42103 Setting key_buffer_size to a negative value may lead to very large allocations Bug#44691 Some plugins configured as MYSQL_PLUGIN_MANDATORY in can be disabled Bug#44797 plugins w/o command-line options have no disabling option in --help Bug#46314 string system variables don't support expressions Bug#46470 sys_vars.max_binlog_cache_size_basic_32 is broken Bug#46586 When using the plugin interface the type "set" for options caused a crash. Bug#47212 Crash in DBUG_PRINT in mysqltest.cc when trying to print octal number Bug#48758 mysqltest crashes on sys_vars.collation_server_basic in gcov builds Bug#49417 some complaints about mysqld --help --verbose output Bug#49540 DEFAULT value of binlog_format isn't the default value Bug#49640 ambiguous option '--skip-skip-myisam' (double skip prefix) Bug#49644 init_connect and \0 Bug#49645 init_slave and multi-byte characters Bug#49646 mysql --show-warnings crashes when server dies
2009-12-22 10:35:56 +01:00
thd->variables.option_bits&= ~OPTION_QUOTE_SHOW_CREATE; /* restore option */
2000-07-31 21:29:14 +02:00
thd->temporary_tables=0;
DBUG_RETURN(error);
2000-07-31 21:29:14 +02:00
}
/*
Find table in list.
SYNOPSIS
find_table_in_list()
table Pointer to table list
offset Offset to which list in table structure to use
db_name Data base name
table_name Table name
2004-07-16 00:15:55 +02:00
NOTES:
This is called by find_table_in_local_list() and
find_table_in_global_list().
2004-07-16 00:15:55 +02:00
RETURN VALUES
NULL Table not found
# Pointer to found table.
*/
TABLE_LIST *find_table_in_list(TABLE_LIST *table,
TABLE_LIST *TABLE_LIST::*link,
const char *db_name,
const char *table_name)
2004-07-16 00:15:55 +02:00
{
for (; table; table= table->*link )
{
if ((table->table == 0 || table->table->s->tmp_table == NO_TMP_TABLE) &&
strcmp(table->db, db_name) == 0 &&
strcmp(table->table_name, table_name) == 0)
break;
}
2004-07-16 00:15:55 +02:00
return table;
}
/**
Test that table is unique (It's only exists once in the table list)
@param thd thread handle
@param table table which should be checked
@param table_list list of tables
@param check_alias whether to check tables' aliases
NOTE: to exclude derived tables from check we use following mechanism:
a) during derived table processing set THD::derived_tables_processing
b) JOIN::prepare set SELECT::exclude_from_table_unique_test if
THD::derived_tables_processing set. (we can't use JOIN::execute
because for PS we perform only JOIN::prepare, but we can't set this
flag in JOIN::prepare if we are not sure that we are in derived table
processing loop, because multi-update call fix_fields() for some its
items (which mean JOIN::prepare for subqueries) before unique_table
call to detect which tables should be locked for write).
c) find_dup_table skip all tables which belong to SELECT with
SELECT::exclude_from_table_unique_test set.
Also SELECT::exclude_from_table_unique_test used to exclude from check
tables of main SELECT of multi-delete and multi-update
We also skip tables with TABLE_LIST::prelocking_placeholder set,
because we want to allow SELECTs from them, and their modification
will rise the error anyway.
TODO: when we will have table/view change detection we can do this check
only once for PS/SP
@retval !=0 found duplicate
@retval 0 if table is unique
*/
static
TABLE_LIST* find_dup_table(THD *thd, TABLE_LIST *table, TABLE_LIST *table_list,
bool check_alias)
{
2004-09-10 01:22:44 +02:00
TABLE_LIST *res;
const char *d_name, *t_name, *t_alias;
DBUG_ENTER("find_dup_table");
2005-02-26 23:46:40 +01:00
DBUG_PRINT("enter", ("table alias: %s", table->alias));
/*
If this function called for query which update table (INSERT/UPDATE/...)
then we have in table->table pointer to TABLE object which we are
updating even if it is VIEW so we need TABLE_LIST of this TABLE object
to get right names (even if lower_case_table_names used).
If this function called for CREATE command that we have not opened table
(table->table equal to 0) and right names is in current TABLE_LIST
object.
*/
if (table->table)
{
/* All MyISAMMRG children are plain MyISAM tables. */
DBUG_ASSERT(table->table->file->ht->db_type != DB_TYPE_MRG_MYISAM);
/* temporary table is always unique */
if (table->table && table->table->s->tmp_table != NO_TMP_TABLE)
DBUG_RETURN(0);
table= table->find_underlying_table(table->table);
/*
as far as we have table->table we have to find real TABLE_LIST of
it in underlying tables
*/
DBUG_ASSERT(table);
}
d_name= table->db;
t_name= table->table_name;
t_alias= table->alias;
DBUG_PRINT("info", ("real table: %s.%s", d_name, t_name));
for (;;)
2004-09-10 01:22:44 +02:00
{
/*
Table is unique if it is present only once in the global list
of tables and once in the list of table locks.
*/
if (! (res= find_table_in_global_list(table_list, d_name, t_name)))
break;
/* Skip if same underlying table. */
if (res->table && (res->table == table->table))
goto next;
/* Skip if table alias does not match. */
if (check_alias)
{
if (lower_case_table_names ?
2007-03-02 11:25:41 +01:00
my_strcasecmp(files_charset_info, t_alias, res->alias) :
strcmp(t_alias, res->alias))
goto next;
}
/*
Skip if marked to be excluded (could be a derived table) or if
entry is a prelocking placeholder.
*/
if (res->select_lex &&
!res->select_lex->exclude_from_table_unique_test &&
!res->prelocking_placeholder)
break;
/*
If we found entry of this table or table of SELECT which already
processed in derived table or top select of multi-update/multi-delete
(exclude_from_table_unique_test) or prelocking placeholder.
*/
next:
table_list= res->next_global;
DBUG_PRINT("info",
("found same copy of table or table which we should skip"));
2004-09-10 01:22:44 +02:00
}
DBUG_RETURN(res);
}
/**
Test that the subject table of INSERT/UPDATE/DELETE/CREATE
or (in case of MyISAMMRG) one of its children are not used later
in the query.
For MyISAMMRG tables, it is assumed that all the underlying
tables of @c table (if any) are listed right after it and that
their @c parent_l field points at the main table.
@retval non-NULL The table list element for the table that
represents the duplicate.
@retval NULL No duplicates found.
*/
TABLE_LIST*
unique_table(THD *thd, TABLE_LIST *table, TABLE_LIST *table_list,
bool check_alias)
{
TABLE_LIST *dup;
if (table->table && table->table->file->ht->db_type == DB_TYPE_MRG_MYISAM)
{
TABLE_LIST *child;
dup= NULL;
/* Check duplicates of all merge children. */
for (child= table->next_global; child && child->parent_l == table;
child= child->next_global)
{
if ((dup= find_dup_table(thd, child, child->next_global, check_alias)))
break;
}
}
else
dup= find_dup_table(thd, table, table_list, check_alias);
return dup;
}
/*
Issue correct error message in case we found 2 duplicate tables which
prevent some update operation
SYNOPSIS
update_non_unique_table_error()
update table which we try to update
operation name of update operation
duplicate duplicate table which we found
NOTE:
here we hide view underlying tables if we have them
*/
void update_non_unique_table_error(TABLE_LIST *update,
const char *operation,
TABLE_LIST *duplicate)
{
update= update->top_table();
duplicate= duplicate->top_table();
if (!update->view || !duplicate->view ||
update->view == duplicate->view ||
update->view_name.length != duplicate->view_name.length ||
update->view_db.length != duplicate->view_db.length ||
my_strcasecmp(table_alias_charset,
update->view_name.str, duplicate->view_name.str) != 0 ||
my_strcasecmp(table_alias_charset,
update->view_db.str, duplicate->view_db.str) != 0)
{
/*
it is not the same view repeated (but it can be parts of the same copy
of view), so we have to hide underlying tables.
*/
if (update->view)
{
/* Issue the ER_NON_INSERTABLE_TABLE error for an INSERT */
if (update->view == duplicate->view)
my_error(!strncmp(operation, "INSERT", 6) ?
ER_NON_INSERTABLE_TABLE : ER_NON_UPDATABLE_TABLE, MYF(0),
update->alias, operation);
else
my_error(ER_VIEW_PREVENT_UPDATE, MYF(0),
(duplicate->view ? duplicate->alias : update->alias),
operation, update->alias);
return;
}
if (duplicate->view)
{
my_error(ER_VIEW_PREVENT_UPDATE, MYF(0), duplicate->alias, operation,
update->alias);
return;
}
}
my_error(ER_UPDATE_TABLE_USED, MYF(0), update->alias);
}
/**
Find temporary table specified by database and table names in the
THD::temporary_tables list.
@return TABLE instance if a temporary table has been found; NULL otherwise.
*/
TABLE *find_temporary_table(THD *thd, const char *db, const char *table_name)
2000-07-31 21:29:14 +02:00
{
TABLE_LIST tl;
2000-07-31 21:29:14 +02:00
tl.db= (char*) db;
tl.table_name= (char*) table_name;
return find_temporary_table(thd, &tl);
}
/**
Find a temporary table specified by TABLE_LIST instance in the
THD::temporary_tables list.
@return TABLE instance if a temporary table has been found; NULL otherwise.
*/
TABLE *find_temporary_table(THD *thd, const TABLE_LIST *tl)
{
char key[MAX_DBKEY_LENGTH];
uint key_length= create_table_def_key(thd, key, tl, 1);
return find_temporary_table(thd, key, key_length);
}
/**
Find a temporary table specified by a key in the THD::temporary_tables list.
@return TABLE instance if a temporary table has been found; NULL otherwise.
*/
TABLE *find_temporary_table(THD *thd,
const char *table_key,
uint table_key_length)
{
for (TABLE *table= thd->temporary_tables; table; table= table->next)
2000-07-31 21:29:14 +02:00
{
if (table->s->table_cache_key.length == table_key_length &&
!memcmp(table->s->table_cache_key.str, table_key, table_key_length))
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
{
return table;
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
}
2000-07-31 21:29:14 +02:00
}
return NULL;
2000-07-31 21:29:14 +02:00
}
2007-07-27 14:37:29 +02:00
/**
Drop a temporary table.
Try to locate the table in the list of thd->temporary_tables.
If the table is found:
- if the table is being used by some outer statement, fail.
- if the table is locked with LOCK TABLES or by prelocking,
unlock it and remove it from the list of locked tables
(THD::lock). Currently only transactional temporary tables
are locked.
2007-07-27 14:37:29 +02:00
- Close the temporary table, remove its .FRM
- remove the table from the list of temporary tables
This function is used to drop user temporary tables, as well as
internal tables created in CREATE TEMPORARY TABLE ... SELECT
or ALTER TABLE. Even though part of the work done by this function
is redundant when the table is internal, as long as we
link both internal and user temporary tables into the same
thd->temporary_tables list, it's impossible to tell here whether
we're dealing with an internal or a user temporary table.
If is_trans is not null, we return the type of the table:
either transactional (e.g. innodb) as TRUE or non-transactional
(e.g. myisam) as FALSE.
@retval 0 the table was found and dropped successfully.
@retval 1 the table was not found in the list of temporary tables
of this thread
@retval -1 the table is in use by a outer query
*/
int drop_temporary_table(THD *thd, TABLE_LIST *table_list, bool *is_trans)
2000-07-31 21:29:14 +02:00
{
TABLE *table;
DBUG_ENTER("drop_temporary_table");
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
DBUG_PRINT("tmptable", ("closing table: '%s'.'%s'",
table_list->db, table_list->table_name));
2000-07-31 21:29:14 +02:00
if (!(table= find_temporary_table(thd, table_list)))
DBUG_RETURN(1);
/* Table might be in use by some outer statement. */
if (table->query_id && table->query_id != thd->query_id)
{
my_error(ER_CANT_REOPEN_TABLE, MYF(0), table->alias);
DBUG_RETURN(-1);
}
if (is_trans != NULL)
*is_trans= table->file->has_transactions();
2007-07-27 14:37:29 +02:00
/*
If LOCK TABLES list is not empty and contains this table,
unlock the table and remove the table from this list.
*/
mysql_lock_remove(thd, thd->lock, table);
close_temporary_table(thd, table, 1, 1);
DBUG_RETURN(0);
2000-07-31 21:29:14 +02:00
}
/*
unlink from thd->temporary tables and close temporary table
*/
void close_temporary_table(THD *thd, TABLE *table,
bool free_share, bool delete_table)
{
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
DBUG_ENTER("close_temporary_table");
DBUG_PRINT("tmptable", ("closing table: '%s'.'%s' 0x%lx alias: '%s'",
table->s->db.str, table->s->table_name.str,
(long) table, table->alias));
if (table->prev)
{
table->prev->next= table->next;
if (table->prev->next)
table->next->prev= table->prev;
}
else
{
/* removing the item from the list */
DBUG_ASSERT(table == thd->temporary_tables);
/*
slave must reset its temporary list pointer to zero to exclude
passing non-zero value to end_slave via rli->save_temporary_tables
when no temp tables opened, see an invariant below.
*/
thd->temporary_tables= table->next;
if (thd->temporary_tables)
table->next->prev= 0;
}
if (thd->slave_thread)
{
/* natural invariant of temporary_tables */
DBUG_ASSERT(slave_open_temp_tables || !thd->temporary_tables);
slave_open_temp_tables--;
}
close_temporary(table, free_share, delete_table);
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
DBUG_VOID_RETURN;
}
/*
Close and delete a temporary table
NOTE
This dosn't unlink table from thd->temporary
If this is needed, use close_temporary_table()
*/
void close_temporary(TABLE *table, bool free_share, bool delete_table)
{
handlerton *table_type= table->s->db_type();
DBUG_ENTER("close_temporary");
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
DBUG_PRINT("tmptable", ("closing table: '%s'.'%s'",
table->s->db.str, table->s->table_name.str));
free_io_cache(table);
closefrm(table, 0);
if (delete_table)
rm_temporary_table(table_type, table->s->path.str);
if (free_share)
{
free_table_share(table->s);
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(table);
}
DBUG_VOID_RETURN;
}
/*
Used by ALTER TABLE when the table is a temporary one. It changes something
only if the ALTER contained a RENAME clause (otherwise, table_name is the old
name).
Prepares a table cache key, which is the concatenation of db, table_name and
thd->slave_proxy_id, separated by '\0'.
*/
2000-11-26 07:30:02 +01:00
bool rename_temporary_table(THD* thd, TABLE *table, const char *db,
2000-07-31 21:29:14 +02:00
const char *table_name)
{
char *key;
uint key_length;
TABLE_SHARE *share= table->s;
TABLE_LIST table_list;
DBUG_ENTER("rename_temporary_table");
if (!(key=(char*) alloc_root(&share->mem_root, MAX_DBKEY_LENGTH)))
DBUG_RETURN(1); /* purecov: inspected */
table_list.db= (char*) db;
table_list.table_name= (char*) table_name;
key_length= create_table_def_key(thd, key, &table_list, 1);
share->set_table_cache_key(key, key_length);
DBUG_RETURN(0);
2000-07-31 21:29:14 +02:00
}
/**
Force all other threads to stop using the table by upgrading
metadata lock on it and remove unused TABLE instances from cache.
@param thd Thread handler
@param table Table to remove from cache
@param function HA_EXTRA_PREPARE_FOR_DROP if table is to be deleted
HA_EXTRA_FORCE_REOPEN if table is not be used
HA_EXTRA_PREPARE_FOR_RENAME if table is to be renamed
@note When returning, the table will be unusable for other threads
until metadata lock is downgraded.
@retval FALSE Success.
@retval TRUE Failure (e.g. because thread was killed).
*/
bool wait_while_table_is_used(THD *thd, TABLE *table,
enum ha_extra_function function)
{
DBUG_ENTER("wait_while_table_is_used");
DBUG_PRINT("enter", ("table: '%s' share: 0x%lx db_stat: %u version: %lu",
table->s->table_name.str, (ulong) table->s,
table->db_stat, table->s->version));
Bug #45225 Locking: hang if drop table with no timeout This patch introduces timeouts for metadata locks. The timeout is specified in seconds using the new dynamic system variable "lock_wait_timeout" which has both GLOBAL and SESSION scopes. Allowed values range from 1 to 31536000 seconds (= 1 year). The default value is 1 year. The new server parameter "lock-wait-timeout" can be used to set the default value parameter upon server startup. "lock_wait_timeout" applies to all statements that use metadata locks. These include DML and DDL operations on tables, views, stored procedures and stored functions. They also include LOCK TABLES, FLUSH TABLES WITH READ LOCK and HANDLER statements. The patch also changes thr_lock.c code (table data locks used by MyISAM and other simplistic engines) to use the same system variable. InnoDB row locks are unaffected. One exception to the handling of the "lock_wait_timeout" variable is delayed inserts. All delayed inserts are executed with a timeout of 1 year regardless of the setting for the global variable. As the connection issuing the delayed insert gets no notification of delayed insert timeouts, we want to avoid unnecessary timeouts. It's important to note that the timeout value is used for each lock acquired and that one statement can take more than one lock. A statement can therefore block for longer than the lock_wait_timeout value before reporting a timeout error. When lock timeout occurs, ER_LOCK_WAIT_TIMEOUT is reported. Test case added to lock_multi.test.
2010-02-11 11:23:39 +01:00
if (thd->mdl_context.upgrade_shared_lock_to_exclusive(
table->mdl_ticket, thd->variables.lock_wait_timeout))
DBUG_RETURN(TRUE);
tdc_remove_table(thd, TDC_RT_REMOVE_NOT_OWN,
table->s->db.str, table->s->table_name.str,
FALSE);
/* extra() call must come only after all instances above are closed */
(void) table->file->extra(function);
DBUG_RETURN(FALSE);
}
/**
Close a and drop a just created table in CREATE TABLE ... SELECT.
@param thd Thread handle
@param table TABLE object for the table to be dropped
@param db_name Name of database for this table
@param table_name Name of this table
This routine assumes that the table to be closed is open only
by the calling thread, so we needn't wait until other threads
close the table. It also assumes that the table is first
in thd->open_ables and a data lock on it, if any, has been
released. To sum up, it's tuned to work with
CREATE TABLE ... SELECT and CREATE TABLE .. SELECT only.
Note, that currently CREATE TABLE ... SELECT is not supported
under LOCK TABLES. This function, still, can be called in
prelocked mode, e.g. if we do CREATE TABLE .. SELECT f1();
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
*/
void drop_open_table(THD *thd, TABLE *table, const char *db_name,
const char *table_name)
{
DBUG_ENTER("drop_open_table");
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
if (table->s->tmp_table)
close_temporary_table(thd, table, 1, 1);
else
{
DBUG_ASSERT(table == thd->open_tables);
2007-05-15 15:58:26 +02:00
handlerton *table_type= table->s->db_type();
table->file->extra(HA_EXTRA_PREPARE_FOR_DROP);
close_thread_table(thd, &thd->open_tables);
/* Remove the table share from the table cache. */
tdc_remove_table(thd, TDC_RT_REMOVE_ALL, db_name, table_name,
FALSE);
/* Remove the table from the storage engine and rm the .frm. */
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
quick_rm_table(table_type, db_name, table_name, 0);
}
DBUG_VOID_RETURN;
2000-07-31 21:29:14 +02:00
}
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
/**
2007-10-16 21:37:31 +02:00
Check that table exists in table definition cache, on disk
or in some storage engine.
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
2007-08-15 15:43:08 +02:00
@param thd Thread context
@param table Table list element
@param[out] exists Out parameter which is set to TRUE if table
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
exists and to FALSE otherwise.
@note This function acquires LOCK_open internally.
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
@note If there is no .FRM file for the table but it exists in one
of engines (e.g. it was created on another node of NDB cluster)
this function will fetch and create proper .FRM file for it.
@retval TRUE Some error occurred
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
@retval FALSE No error. 'exists' out parameter set accordingly.
*/
bool check_if_table_exists(THD *thd, TABLE_LIST *table, bool *exists)
{
char path[FN_REFLEN + 1];
TABLE_SHARE *share;
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
DBUG_ENTER("check_if_table_exists");
*exists= TRUE;
DBUG_ASSERT(thd->mdl_context.
is_lock_owner(MDL_key::TABLE, table->db,
table->table_name, MDL_SHARED));
mysql_mutex_lock(&LOCK_open);
share= get_cached_table_share(table->db, table->table_name);
mysql_mutex_unlock(&LOCK_open);
if (share)
goto end;
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
build_table_filename(path, sizeof(path) - 1, table->db, table->table_name,
reg_ext, 0);
if (!access(path, F_OK))
goto end;
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
/* .FRM file doesn't exist. Check if some engine can provide it. */
if (ha_check_if_table_exists(thd, table->db, table->table_name, exists))
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
{
my_printf_error(ER_OUT_OF_RESOURCES, "Failed to open '%-.64s', error while "
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
"unpacking from engine", MYF(0), table->table_name);
DBUG_RETURN(TRUE);
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
}
end:
DBUG_RETURN(FALSE);
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
}
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
/**
An error handler which converts, if possible, ER_LOCK_DEADLOCK error
that can occur when we are trying to acquire a metadata lock to
a request for back-off and re-start of open_tables() process.
*/
class MDL_deadlock_handler : public Internal_error_handler
{
public:
MDL_deadlock_handler(Open_table_context *ot_ctx_arg)
: m_ot_ctx(ot_ctx_arg), m_is_active(FALSE)
{}
virtual ~MDL_deadlock_handler() {}
virtual bool handle_condition(THD *thd,
uint sql_errno,
const char* sqlstate,
MYSQL_ERROR::enum_warning_level level,
const char* msg,
MYSQL_ERROR ** cond_hdl);
private:
/** Open table context to be used for back-off request. */
Open_table_context *m_ot_ctx;
/**
Indicates that we are already in the process of handling
ER_LOCK_DEADLOCK error. Allows to re-emit the error from
the error handler without falling into infinite recursion.
*/
bool m_is_active;
};
bool MDL_deadlock_handler::handle_condition(THD *,
uint sql_errno,
const char*,
MYSQL_ERROR::enum_warning_level,
const char*,
MYSQL_ERROR ** cond_hdl)
{
*cond_hdl= NULL;
if (! m_is_active && sql_errno == ER_LOCK_DEADLOCK)
{
/* Disable the handler to avoid infinite recursion. */
m_is_active= TRUE;
(void) m_ot_ctx->request_backoff_action(
Open_table_context::OT_BACKOFF_AND_RETRY,
NULL);
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
m_is_active= FALSE;
/*
If the above back-off request failed, a new instance of
ER_LOCK_DEADLOCK error was emitted. Thus the current
instance of error condition can be treated as handled.
*/
return TRUE;
}
return FALSE;
}
/**
Try to acquire an MDL lock for a table being opened.
@param[in,out] thd Session context, to report errors.
@param[out] ot_ctx Open table context, to hold the back off
state. If we failed to acquire a lock
due to a lock conflict, we add the
failed request to the open table context.
@param[in,out] mdl_request A request for an MDL lock.
If we managed to acquire a ticket
(no errors or lock conflicts occurred),
contains a reference to it on
return. However, is not modified if MDL
lock type- modifying flags were provided.
@param[in] flags flags MYSQL_OPEN_FORCE_SHARED_MDL,
MYSQL_OPEN_FORCE_SHARED_HIGH_PRIO_MDL or
MYSQL_OPEN_FAIL_ON_MDL_CONFLICT
@sa open_table().
@param[out] mdl_ticket Only modified if there was no error.
If we managed to acquire an MDL
lock, contains a reference to the
ticket, otherwise is set to NULL.
@retval TRUE An error occurred.
@retval FALSE No error, but perhaps a lock conflict, check mdl_ticket.
*/
static bool
open_table_get_mdl_lock(THD *thd, Open_table_context *ot_ctx,
MDL_request *mdl_request,
uint flags,
MDL_ticket **mdl_ticket)
{
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
MDL_request mdl_request_shared;
if (flags & (MYSQL_OPEN_FORCE_SHARED_MDL |
MYSQL_OPEN_FORCE_SHARED_HIGH_PRIO_MDL))
{
/*
MYSQL_OPEN_FORCE_SHARED_MDL flag means that we are executing
PREPARE for a prepared statement and want to override
the type-of-operation aware metadata lock which was set
in the parser/during view opening with a simple shared
metadata lock.
This is necessary to allow concurrent execution of PREPARE
and LOCK TABLES WRITE statement against the same table.
MYSQL_OPEN_FORCE_SHARED_HIGH_PRIO_MDL flag means that we open
the table in order to get information about it for one of I_S
queries and also want to override the type-of-operation aware
shared metadata lock which was set earlier (e.g. during view
opening) with a high-priority shared metadata lock.
This is necessary to avoid unnecessary waiting and extra
ER_WARN_I_S_SKIPPED_TABLE warnings when accessing I_S tables.
These two flags are mutually exclusive.
*/
DBUG_ASSERT(!(flags & MYSQL_OPEN_FORCE_SHARED_MDL) ||
!(flags & MYSQL_OPEN_FORCE_SHARED_HIGH_PRIO_MDL));
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
mdl_request_shared.init(&mdl_request->key,
(flags & MYSQL_OPEN_FORCE_SHARED_MDL) ?
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
MDL_SHARED : MDL_SHARED_HIGH_PRIO,
MDL_TRANSACTION);
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
mdl_request= &mdl_request_shared;
}
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
if (flags & MYSQL_OPEN_FAIL_ON_MDL_CONFLICT)
{
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
/*
When table is being open in order to get data for I_S table,
we might have some tables not only open but also locked (e.g. when
this happens under LOCK TABLES or in a stored function).
As a result by waiting on a conflicting metadata lock to go away
we may create a deadlock which won't entirely belong to the
MDL subsystem and thus won't be detectable by this subsystem's
deadlock detector.
To avoid such situation we skip the trouble-making table if
there is a conflicting lock.
*/
if (thd->mdl_context.try_acquire_lock(mdl_request))
return TRUE;
if (mdl_request->ticket == NULL)
{
my_error(ER_WARN_I_S_SKIPPED_TABLE, MYF(0),
mdl_request->key.db_name(), mdl_request->key.name());
return TRUE;
}
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
}
else
{
/*
We are doing a normal table open. Let us try to acquire a metadata
lock on the table. If there is a conflicting lock, acquire_lock()
will wait for it to go away. Sometimes this waiting may lead to a
deadlock, with the following results:
1) If a deadlock is entirely within MDL subsystem, it is
detected by the deadlock detector of this subsystem.
ER_LOCK_DEADLOCK error is produced. Then, the error handler
that is installed prior to the call to acquire_lock() attempts
to request a back-off and retry. Upon success, ER_LOCK_DEADLOCK
error is suppressed, otherwise propagated up the calling stack.
2) Otherwise, a deadlock may occur when the wait-for graph
includes edges not visible to the MDL deadlock detector.
One such example is a wait on an InnoDB row lock, e.g. when:
conn C1 gets SR MDL lock on t1 with SELECT * FROM t1
conn C2 gets a row lock on t2 with SELECT * FROM t2 FOR UPDATE
conn C3 gets in and waits on C1 with DROP TABLE t0, t1
conn C2 continues and blocks on C3 with SELECT * FROM t0
conn C1 deadlocks by waiting on C2 by issuing SELECT * FROM
t2 LOCK IN SHARE MODE.
Such circular waits are currently only resolved by timeouts,
e.g. @@innodb_lock_wait_timeout or @@lock_wait_timeout.
*/
MDL_deadlock_handler mdl_deadlock_handler(ot_ctx);
thd->push_internal_handler(&mdl_deadlock_handler);
bool result= thd->mdl_context.acquire_lock(mdl_request,
ot_ctx->get_timeout());
thd->pop_internal_handler();
if (result && !ot_ctx->can_recover_from_failed_open())
return TRUE;
}
*mdl_ticket= mdl_request->ticket;
return FALSE;
}
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
/**
Check if table's share is being removed from the table definition
cache and, if yes, wait until the flush is complete.
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
@param thd Thread context.
@param table_list Table which share should be checked.
@param timeout Timeout for waiting.
@param deadlock_weight Weight of this wait for deadlock detector.
@retval FALSE Success. Share is up to date or has been flushed.
@retval TRUE Error (OOM, our was killed, the wait resulted
in a deadlock or timeout). Reported.
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
*/
static bool
tdc_wait_for_old_version(THD *thd, const char *db, const char *table_name,
ulong wait_timeout, uint deadlock_weight)
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
{
TABLE_SHARE *share;
bool res= FALSE;
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
mysql_mutex_lock(&LOCK_open);
if ((share= get_cached_table_share(db, table_name)) &&
share->has_old_version())
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
{
struct timespec abstime;
set_timespec(abstime, wait_timeout);
res= share->wait_for_old_version(thd, &abstime, deadlock_weight);
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
}
mysql_mutex_unlock(&LOCK_open);
return res;
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
}
/*
Open a table.
SYNOPSIS
open_table()
thd Thread context.
table_list Open first table in list.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
action INOUT Pointer to variable of enum_open_table_action type
which will be set according to action which is
required to remedy problem appeared during attempt
to open table.
flags Bitmap of flags to modify how open works:
MYSQL_OPEN_IGNORE_FLUSH - Open table even if
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
someone has done a flush or there is a pending
exclusive metadata lock requests against it
(i.e. request high priority metadata lock).
No version number checking is done.
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
MYSQL_OPEN_TEMPORARY_ONLY - Open only temporary
table not the base table or view.
MYSQL_OPEN_TAKE_UPGRADABLE_MDL - Obtain upgradable
metadata lock for tables on which we are going to
take some kind of write table-level lock.
IMPLEMENTATION
Uses a cache of open tables to find a table not in use.
Backport of revno ## 2617.31.1, 2617.31.3, 2617.31.4, 2617.31.5, 2617.31.12, 2617.31.15, 2617.31.15, 2617.31.16, 2617.43.1 - initial changeset that introduced the fix for Bug#989 and follow up fixes for all test suite failures introduced in the initial changeset. ------------------------------------------------------------ revno: 2617.31.1 committer: Davi Arnaut <Davi.Arnaut@Sun.COM> branch nick: 4284-6.0 timestamp: Fri 2009-03-06 19:17:00 -0300 message: Bug#989: If DROP TABLE while there's an active transaction, wrong binlog order WL#4284: Transactional DDL locking Currently the MySQL server does not keep metadata locks on schema objects for the duration of a transaction, thus failing to guarantee the integrity of the schema objects being used during the transaction and to protect then from concurrent DDL operations. This also poses a problem for replication as a DDL operation might be replicated even thought there are active transactions using the object being modified. The solution is to defer the release of metadata locks until a active transaction is either committed or rolled back. This prevents other statements from modifying the table for the entire duration of the transaction. This provides commitment ordering for guaranteeing serializability across multiple transactions. - Incompatible change: If MySQL's metadata locking system encounters a lock conflict, the usual schema is to use the try and back-off technique to avoid deadlocks -- this schema consists in releasing all locks and trying to acquire them all in one go. But in a transactional context this algorithm can't be utilized as its not possible to release locks acquired during the course of the transaction without breaking the transaction commitments. To avoid deadlocks in this case, the ER_LOCK_DEADLOCK will be returned if a lock conflict is encountered during a transaction. Let's consider an example: A transaction has two statements that modify table t1, then table t2, and then commits. The first statement of the transaction will acquire a shared metadata lock on table t1, and it will be kept utill COMMIT to ensure serializability. At the moment when the second statement attempts to acquire a shared metadata lock on t2, a concurrent ALTER or DROP statement might have locked t2 exclusively. The prescription of the current locking protocol is that the acquirer of the shared lock backs off -- gives up all his current locks and retries. This implies that the entire multi-statement transaction has to be rolled back. - Incompatible change: FLUSH commands such as FLUSH PRIVILEGES and FLUSH TABLES WITH READ LOCK won't cause locked tables to be implicitly unlocked anymore.
2009-12-05 00:02:48 +01:00
If TABLE_LIST::open_strategy is set to OPEN_IF_EXISTS, the table is opened
only if it exists. If the open strategy is OPEN_STUB, the underlying table
is never opened. In both cases, metadata locks are always taken according
to the lock strategy.
RETURN
TRUE Open failed. "action" parameter may contain type of action
needed to remedy problem before retrying again.
FALSE Success. Members of TABLE_LIST structure are filled properly (e.g.
TABLE_LIST::table is set for real tables and TABLE_LIST::view is
set for views).
*/
2000-07-31 21:29:14 +02:00
bool open_table(THD *thd, TABLE_LIST *table_list, MEM_ROOT *mem_root,
Open_table_context *ot_ctx)
2000-07-31 21:29:14 +02:00
{
reg1 TABLE *table;
char key[MAX_DBKEY_LENGTH];
uint key_length;
2004-07-16 00:15:55 +02:00
char *alias= table_list->alias;
uint flags= ot_ctx->get_flags();
MDL_ticket *mdl_ticket;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
int error;
TABLE_SHARE *share;
2009-11-10 15:09:44 +01:00
my_hash_value_type hash_value;
2000-07-31 21:29:14 +02:00
DBUG_ENTER("open_table");
/* an open table operation needs a lot of the stack space */
WL#3817: Simplify string / memory area types and make things more consistent (first part) The following type conversions was done: - Changed byte to uchar - Changed gptr to uchar* - Change my_string to char * - Change my_size_t to size_t - Change size_s to size_t Removed declaration of byte, gptr, my_string, my_size_t and size_s. Following function parameter changes was done: - All string functions in mysys/strings was changed to use size_t instead of uint for string lengths. - All read()/write() functions changed to use size_t (including vio). - All protocoll functions changed to use size_t instead of uint - Functions that used a pointer to a string length was changed to use size_t* - Changed malloc(), free() and related functions from using gptr to use void * as this requires fewer casts in the code and is more in line with how the standard functions work. - Added extra length argument to dirname_part() to return the length of the created string. - Changed (at least) following functions to take uchar* as argument: - db_dump() - my_net_write() - net_write_command() - net_store_data() - DBUG_DUMP() - decimal2bin() & bin2decimal() - Changed my_compress() and my_uncompress() to use size_t. Changed one argument to my_uncompress() from a pointer to a value as we only return one value (makes function easier to use). - Changed type of 'pack_data' argument to packfrm() to avoid casts. - Changed in readfrm() and writefrom(), ha_discover and handler::discover() the type for argument 'frmdata' to uchar** to avoid casts. - Changed most Field functions to use uchar* instead of char* (reduced a lot of casts). - Changed field->val_xxx(xxx, new_ptr) to take const pointers. Other changes: - Removed a lot of not needed casts - Added a few new cast required by other changes - Added some cast to my_multi_malloc() arguments for safety (as string lengths needs to be uint, not size_t). - Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done explicitely as this conflict was often hided by casting the function to hash_get_key). - Changed some buffers to memory regions to uchar* to avoid casts. - Changed some string lengths from uint to size_t. - Changed field->ptr to be uchar* instead of char*. This allowed us to get rid of a lot of casts. - Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar - Include zlib.h in some files as we needed declaration of crc32() - Changed MY_FILE_ERROR to be (size_t) -1. - Changed many variables to hold the result of my_read() / my_write() to be size_t. This was needed to properly detect errors (which are returned as (size_t) -1). - Removed some very old VMS code - Changed packfrm()/unpackfrm() to not be depending on uint size (portability fix) - Removed windows specific code to restore cursor position as this causes slowdown on windows and we should not mix read() and pread() calls anyway as this is not thread safe. Updated function comment to reflect this. Changed function that depended on original behavior of my_pwrite() to itself restore the cursor position (one such case). - Added some missing checking of return value of malloc(). - Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow. - Changed type of table_def::m_size from my_size_t to ulong to reflect that m_size is the number of elements in the array, not a string/memory length. - Moved THD::max_row_length() to table.cc (as it's not depending on THD). Inlined max_row_length_blob() into this function. - More function comments - Fixed some compiler warnings when compiled without partitions. - Removed setting of LEX_STRING() arguments in declaration (portability fix). - Some trivial indentation/variable name changes. - Some trivial code simplifications: - Replaced some calls to alloc_root + memcpy to use strmake_root()/strdup_root(). - Changed some calls from memdup() to strmake() (Safety fix) - Simpler loops in client-simple.c
2007-05-10 11:59:39 +02:00
if (check_stack_overrun(thd, STACK_MIN_SIZE_FOR_OPEN, (uchar *)&alias))
DBUG_RETURN(TRUE);
2000-07-31 21:29:14 +02:00
if (thd->killed)
DBUG_RETURN(TRUE);
key_length= (create_table_def_key(thd, key, table_list, 1) -
TMP_TABLE_KEY_EXTRA);
/*
Unless requested otherwise, try to resolve this table in the list
of temporary tables of this thread. In MySQL temporary tables
are always thread-local and "shadow" possible base tables with the
same name. This block implements the behaviour.
TODO: move this block into a separate function.
*/
if (table_list->open_type != OT_BASE_ONLY &&
! (flags & MYSQL_OPEN_SKIP_TEMPORARY))
2000-07-31 21:29:14 +02:00
{
2004-07-16 00:15:55 +02:00
for (table= thd->temporary_tables; table ; table=table->next)
2000-07-31 21:29:14 +02:00
{
if (table->s->table_cache_key.length == key_length +
TMP_TABLE_KEY_EXTRA &&
!memcmp(table->s->table_cache_key.str, key,
2004-10-29 18:26:52 +02:00
key_length + TMP_TABLE_KEY_EXTRA))
2000-07-31 21:29:14 +02:00
{
/*
We're trying to use the same temporary table twice in a query.
Right now we don't support this because a temporary table
is always represented by only one TABLE object in THD, and
it can not be cloned. Emit an error for an unsupported behaviour.
*/
if (table->query_id)
2004-07-16 00:15:55 +02:00
{
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
DBUG_PRINT("error",
("query_id: %lu server_id: %u pseudo_thread_id: %lu",
(ulong) table->query_id, (uint) thd->server_id,
(ulong) thd->variables.pseudo_thread_id));
my_error(ER_CANT_REOPEN_TABLE, MYF(0), table->alias);
DBUG_RETURN(TRUE);
2004-07-16 00:15:55 +02:00
}
table->query_id= thd->query_id;
thd->thread_specific_used= TRUE;
2004-10-29 18:26:52 +02:00
DBUG_PRINT("info",("Using temporary table"));
goto reset;
2000-07-31 21:29:14 +02:00
}
}
}
if (table_list->open_type == OT_TEMPORARY_ONLY ||
(flags & MYSQL_OPEN_TEMPORARY_ONLY))
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
{
if (table_list->open_strategy == TABLE_LIST::OPEN_NORMAL)
{
my_error(ER_NO_SUCH_TABLE, MYF(0), table_list->db, table_list->table_name);
DBUG_RETURN(TRUE);
}
else
DBUG_RETURN(FALSE);
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
}
/*
The table is not temporary - if we're in pre-locked or LOCK TABLES
mode, let's try to find the requested table in the list of pre-opened
and locked tables. If the table is not there, return an error - we can't
open not pre-opened tables in pre-locked/LOCK TABLES mode.
TODO: move this block into a separate function.
*/
if (thd->locked_tables_mode &&
! (flags & MYSQL_OPEN_GET_NEW_TABLE))
2000-07-31 21:29:14 +02:00
{ // Using table locks
TABLE *best_table= 0;
int best_distance= INT_MIN;
2000-07-31 21:29:14 +02:00
for (table=thd->open_tables; table ; table=table->next)
{
if (table->s->table_cache_key.length == key_length &&
!memcmp(table->s->table_cache_key.str, key, key_length))
2004-06-25 15:52:01 +02:00
{
if (!my_strcasecmp(system_charset_info, table->alias, alias) &&
table->query_id != thd->query_id && /* skip tables already used */
(thd->locked_tables_mode == LTM_LOCK_TABLES ||
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
table->query_id == 0))
{
int distance= ((int) table->reginfo.lock_type -
(int) table_list->lock_type);
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
/*
Find a table that either has the exact lock type requested,
or has the best suitable lock. In case there is no locked
table that has an equal or higher lock than requested,
we us the closest matching lock to be able to produce an error
message about wrong lock mode on the table. The best_table
is changed if bd < 0 <= d or bd < d < 0 or 0 <= d < bd.
distance < 0 - No suitable lock found
distance > 0 - we have lock mode higher then we require
distance == 0 - we have lock mode exactly which we need
*/
if ((best_distance < 0 && distance > best_distance) ||
(distance >= 0 && distance < best_distance))
{
best_distance= distance;
best_table= table;
if (best_distance == 0)
{
/*
We have found a perfect match and can finish iterating
through open tables list. Check for table use conflict
between calling statement and SP/trigger is done in
lock_tables().
*/
break;
}
}
}
2004-06-25 15:52:01 +02:00
}
2000-07-31 21:29:14 +02:00
}
if (best_table)
{
table= best_table;
table->query_id= thd->query_id;
DBUG_PRINT("info",("Using locked table"));
goto reset;
}
/*
Is this table a view and not a base table?
(it is work around to allow to open view with locked tables,
real fix will be made after definition cache will be made)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
Since opening of view which was not explicitly locked by LOCK
TABLES breaks metadata locking protocol (potentially can lead
to deadlocks) it should be disallowed.
*/
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
if (thd->mdl_context.is_lock_owner(MDL_key::TABLE,
table_list->db,
table_list->table_name,
MDL_SHARED))
{
char path[FN_REFLEN + 1];
enum legacy_db_type not_used;
build_table_filename(path, sizeof(path) - 1,
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
table_list->db, table_list->table_name, reg_ext, 0);
/*
Note that we can't be 100% sure that it is a view since it's
possible that we either simply have not found unused TABLE
instance in THD::open_tables list or were unable to open table
during prelocking process (in this case in theory we still
should hold shared metadata lock on it).
*/
if (dd_frm_type(thd, path, &not_used) == FRMTYPE_VIEW)
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!tdc_open_view(thd, table_list, alias, key, key_length,
mem_root, 0))
{
DBUG_ASSERT(table_list->view != 0);
DBUG_RETURN(FALSE); // VIEW
}
}
}
/*
No table in the locked tables list. In case of explicit LOCK TABLES
this can happen if a user did not include the able into the list.
In case of pre-locked mode locked tables list is generated automatically,
so we may only end up here if the table did not exist when
locked tables list was created.
*/
if (thd->locked_tables_mode == LTM_PRELOCKED)
my_error(ER_NO_SUCH_TABLE, MYF(0), table_list->db, table_list->alias);
else
my_error(ER_TABLE_NOT_LOCKED, MYF(0), alias);
DBUG_RETURN(TRUE);
}
/*
Non pre-locked/LOCK TABLES mode, and the table is not temporary.
This is the normal use case.
*/
if (! (flags & MYSQL_OPEN_HAS_MDL_LOCK))
{
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
/*
We are not under LOCK TABLES and going to acquire write-lock/
modify the base table. We need to acquire protection against
global read lock until end of this statement in order to have
this statement blocked by active FLUSH TABLES WITH READ LOCK.
We don't block acquire this protection under LOCK TABLES as
such protection already acquired at LOCK TABLES time and
not released until UNLOCK TABLES.
We don't block statements which modify only temporary tables
as these tables are not preserved by backup by any form of
backup which uses FLUSH TABLES WITH READ LOCK.
TODO: The fact that we sometimes acquire protection against
GRL only when we encounter table to be write-locked
slightly increases probability of deadlock.
This problem will be solved once Alik pushes his
temporary table refactoring patch and we can start
pre-acquiring metadata locks at the beggining of
open_tables() call.
*/
if (table_list->mdl_request.type >= MDL_SHARED_WRITE &&
! (flags & (MYSQL_OPEN_IGNORE_GLOBAL_READ_LOCK |
MYSQL_OPEN_FORCE_SHARED_MDL |
MYSQL_OPEN_FORCE_SHARED_HIGH_PRIO_MDL |
MYSQL_OPEN_SKIP_SCOPED_MDL_LOCK)) &&
! ot_ctx->has_protection_against_grl())
{
MDL_request protection_request;
MDL_deadlock_handler mdl_deadlock_handler(ot_ctx);
if (thd->global_read_lock.can_acquire_protection())
DBUG_RETURN(TRUE);
protection_request.init(MDL_key::GLOBAL, "", "", MDL_INTENTION_EXCLUSIVE,
MDL_STATEMENT);
/*
Install error handler which if possible will convert deadlock error
into request to back-off and restart process of opening tables.
*/
thd->push_internal_handler(&mdl_deadlock_handler);
bool result= thd->mdl_context.acquire_lock(&protection_request,
ot_ctx->get_timeout());
thd->pop_internal_handler();
if (result)
DBUG_RETURN(TRUE);
ot_ctx->set_has_protection_against_grl();
}
if (open_table_get_mdl_lock(thd, ot_ctx, &table_list->mdl_request,
flags, &mdl_ticket) ||
mdl_ticket == NULL)
{
DEBUG_SYNC(thd, "before_open_table_wait_refresh");
DBUG_RETURN(TRUE);
}
DEBUG_SYNC(thd, "after_open_table_mdl_shared");
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
else
{
/*
Grab reference to the MDL lock ticket that was acquired
by the caller.
*/
mdl_ticket= table_list->mdl_request.ticket;
}
Backport of: ------------------------------------------------------------ revno: 2617.23.18 committer: Davi Arnaut <Davi.Arnaut@Sun.COM> branch nick: 4284-6.0 timestamp: Mon 2009-03-02 18:18:26 -0300 message: Bug#989: If DROP TABLE while there's an active transaction, wrong binlog order WL#4284: Transactional DDL locking This is a prerequisite patch: These changes are intended to split lock requests from granted locks and to allow the memory and lifetime of granted locks to be managed within the MDL subsystem. Furthermore, tickets can now be shared and therefore are used to satisfy multiple lock requests, but only shared locks can be recursive. The problem is that the MDL subsystem morphs lock requests into granted locks locks but does not manage the memory and lifetime of lock requests, and hence, does not manage the memory of granted locks either. This can be problematic because it puts the burden of tracking references on the users of the subsystem and it can't be easily done in transactional contexts where the locks have to be kept around for the duration of a transaction. Another issue is that recursive locks (when the context trying to acquire a lock already holds a lock on the same object) requires that each time the lock is granted, a unique lock request/granted lock structure structure must be kept around until the lock is released. This can lead to memory leaks in transactional contexts as locks taken during the transaction should only be released at the end of the transaction. This also leads to unnecessary wake ups (broadcasts) in the MDL subsystem if the context still holds a equivalent of the lock being released. These issues are exacerbated due to the fact that WL#4284 low-level design says that the implementation should "2) Store metadata locks in transaction memory root, rather than statement memory root" but this is not possible because a memory root, as implemented in mysys, requires all objects allocated from it to be freed all at once. This patch combines review input and significant code contributions from Konstantin Osipov (kostja) and Dmitri Lenev (dlenev).
2009-12-04 00:29:40 +01:00
2010-02-03 14:43:03 +01:00
hash_value= my_calc_hash(&table_def_cache, (uchar*) key, key_length);
2000-07-31 21:29:14 +02:00
Backport of revno ## 2617.31.1, 2617.31.3, 2617.31.4, 2617.31.5, 2617.31.12, 2617.31.15, 2617.31.15, 2617.31.16, 2617.43.1 - initial changeset that introduced the fix for Bug#989 and follow up fixes for all test suite failures introduced in the initial changeset. ------------------------------------------------------------ revno: 2617.31.1 committer: Davi Arnaut <Davi.Arnaut@Sun.COM> branch nick: 4284-6.0 timestamp: Fri 2009-03-06 19:17:00 -0300 message: Bug#989: If DROP TABLE while there's an active transaction, wrong binlog order WL#4284: Transactional DDL locking Currently the MySQL server does not keep metadata locks on schema objects for the duration of a transaction, thus failing to guarantee the integrity of the schema objects being used during the transaction and to protect then from concurrent DDL operations. This also poses a problem for replication as a DDL operation might be replicated even thought there are active transactions using the object being modified. The solution is to defer the release of metadata locks until a active transaction is either committed or rolled back. This prevents other statements from modifying the table for the entire duration of the transaction. This provides commitment ordering for guaranteeing serializability across multiple transactions. - Incompatible change: If MySQL's metadata locking system encounters a lock conflict, the usual schema is to use the try and back-off technique to avoid deadlocks -- this schema consists in releasing all locks and trying to acquire them all in one go. But in a transactional context this algorithm can't be utilized as its not possible to release locks acquired during the course of the transaction without breaking the transaction commitments. To avoid deadlocks in this case, the ER_LOCK_DEADLOCK will be returned if a lock conflict is encountered during a transaction. Let's consider an example: A transaction has two statements that modify table t1, then table t2, and then commits. The first statement of the transaction will acquire a shared metadata lock on table t1, and it will be kept utill COMMIT to ensure serializability. At the moment when the second statement attempts to acquire a shared metadata lock on t2, a concurrent ALTER or DROP statement might have locked t2 exclusively. The prescription of the current locking protocol is that the acquirer of the shared lock backs off -- gives up all his current locks and retries. This implies that the entire multi-statement transaction has to be rolled back. - Incompatible change: FLUSH commands such as FLUSH PRIVILEGES and FLUSH TABLES WITH READ LOCK won't cause locked tables to be implicitly unlocked anymore.
2009-12-05 00:02:48 +01:00
if (table_list->open_strategy == TABLE_LIST::OPEN_IF_EXISTS)
2000-07-31 21:29:14 +02:00
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
bool exists;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (check_if_table_exists(thd, table_list, &exists))
DBUG_RETURN(TRUE);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!exists)
DBUG_RETURN(FALSE);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/* Table exists. Let us try to open it. */
}
Backport of revno ## 2617.31.1, 2617.31.3, 2617.31.4, 2617.31.5, 2617.31.12, 2617.31.15, 2617.31.15, 2617.31.16, 2617.43.1 - initial changeset that introduced the fix for Bug#989 and follow up fixes for all test suite failures introduced in the initial changeset. ------------------------------------------------------------ revno: 2617.31.1 committer: Davi Arnaut <Davi.Arnaut@Sun.COM> branch nick: 4284-6.0 timestamp: Fri 2009-03-06 19:17:00 -0300 message: Bug#989: If DROP TABLE while there's an active transaction, wrong binlog order WL#4284: Transactional DDL locking Currently the MySQL server does not keep metadata locks on schema objects for the duration of a transaction, thus failing to guarantee the integrity of the schema objects being used during the transaction and to protect then from concurrent DDL operations. This also poses a problem for replication as a DDL operation might be replicated even thought there are active transactions using the object being modified. The solution is to defer the release of metadata locks until a active transaction is either committed or rolled back. This prevents other statements from modifying the table for the entire duration of the transaction. This provides commitment ordering for guaranteeing serializability across multiple transactions. - Incompatible change: If MySQL's metadata locking system encounters a lock conflict, the usual schema is to use the try and back-off technique to avoid deadlocks -- this schema consists in releasing all locks and trying to acquire them all in one go. But in a transactional context this algorithm can't be utilized as its not possible to release locks acquired during the course of the transaction without breaking the transaction commitments. To avoid deadlocks in this case, the ER_LOCK_DEADLOCK will be returned if a lock conflict is encountered during a transaction. Let's consider an example: A transaction has two statements that modify table t1, then table t2, and then commits. The first statement of the transaction will acquire a shared metadata lock on table t1, and it will be kept utill COMMIT to ensure serializability. At the moment when the second statement attempts to acquire a shared metadata lock on t2, a concurrent ALTER or DROP statement might have locked t2 exclusively. The prescription of the current locking protocol is that the acquirer of the shared lock backs off -- gives up all his current locks and retries. This implies that the entire multi-statement transaction has to be rolled back. - Incompatible change: FLUSH commands such as FLUSH PRIVILEGES and FLUSH TABLES WITH READ LOCK won't cause locked tables to be implicitly unlocked anymore.
2009-12-05 00:02:48 +01:00
else if (table_list->open_strategy == TABLE_LIST::OPEN_STUB)
DBUG_RETURN(FALSE);
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
retry_share:
mysql_mutex_lock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!(share= get_table_share_with_discover(thd, table_list, key,
key_length, OPEN_VIEW,
&error,
hash_value)))
{
mysql_mutex_unlock(&LOCK_open);
/*
If thd->is_error() is not set, we either need discover
(error == 7), or the error was silenced by the prelocking
handler (error == 0), in which case we should skip this
table.
*/
if (error == 7 && !thd->is_error())
{
(void) ot_ctx->request_backoff_action(Open_table_context::OT_DISCOVER,
table_list);
}
DBUG_RETURN(TRUE);
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (share->is_view)
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/*
If parent_l of the table_list is non null then a merge table
has this view as child table, which is not supported.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
*/
if (table_list->parent_l)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
my_error(ER_WRONG_MRG_TABLE, MYF(0));
goto err_unlock;
2000-07-31 21:29:14 +02:00
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/*
This table is a view. Validate its metadata version: in particular,
that it was a view when the statement was prepared.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
*/
if (check_and_update_table_version(thd, table_list, share))
goto err_unlock;
if (table_list->i_s_requested_object & OPEN_TABLE_ONLY)
{
my_error(ER_NO_SUCH_TABLE, MYF(0), table_list->db,
table_list->table_name);
goto err_unlock;
}
/* Open view */
if (open_new_frm(thd, share, alias,
(uint) (HA_OPEN_KEYFILE | HA_OPEN_RNDFILE |
HA_GET_INDEX | HA_TRY_READ_ONLY),
READ_KEYINFO | COMPUTE_TYPES | EXTRA_RECORD,
thd->open_options,
0, table_list, mem_root))
goto err_unlock;
/* TODO: Don't free this */
release_table_share(share);
DBUG_ASSERT(table_list->view);
mysql_mutex_unlock(&LOCK_open);
DBUG_RETURN(FALSE);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
/*
Note that situation when we are trying to open a table for what
was a view during previous execution of PS will be handled in by
the caller. Here we should simply open our table even if
TABLE_LIST::view is true.
*/
if (table_list->i_s_requested_object & OPEN_VIEW_ONLY)
{
my_error(ER_NO_SUCH_TABLE, MYF(0), table_list->db,
table_list->table_name);
goto err_unlock;
}
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
if (!(flags & MYSQL_OPEN_IGNORE_FLUSH))
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
if (share->has_old_version())
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
/*
We already have an MDL lock. But we have encountered an old
version of table in the table definition cache which is possible
when someone changes the table version directly in the cache
without acquiring a metadata lock (e.g. this can happen during
"rolling" FLUSH TABLE(S)).
Release our reference to share, wait until old version of
share goes away and then try to get new version of table share.
*/
MDL_deadlock_handler mdl_deadlock_handler(ot_ctx);
bool wait_result;
release_table_share(share);
mysql_mutex_unlock(&LOCK_open);
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
thd->push_internal_handler(&mdl_deadlock_handler);
wait_result= tdc_wait_for_old_version(thd, table_list->db,
table_list->table_name,
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
ot_ctx->get_timeout(),
mdl_ticket->get_deadlock_weight());
thd->pop_internal_handler();
if (wait_result)
DBUG_RETURN(TRUE);
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
goto retry_share;
}
if (thd->open_tables && thd->open_tables->s->version != share->version)
{
/*
If the version changes while we're opening the tables,
we have to back off, close all the tables opened-so-far,
and try to reopen them. Note: refresh_version is currently
changed only during FLUSH TABLES.
*/
release_table_share(share);
2010-02-02 00:22:16 +01:00
mysql_mutex_unlock(&LOCK_open);
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
(void)ot_ctx->request_backoff_action(Open_table_context::OT_REOPEN_TABLES,
NULL);
DBUG_RETURN(TRUE);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
}
if (!share->free_tables.is_empty())
{
A prerequisite patch for the fix for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". Introduce a notion of a sentinel to MDL_context. A sentinel is a ticket that separates all tickets in the context into two groups: before and after it. Currently we can have (and need) only one designated sentinel -- it separates all locks taken by LOCK TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK and all other locks, which must be released at COMMIT or ROLLBACK. The tricky part is maintaining the sentinel up to date when someone release its corresponding ticket. This can happen, e.g. if someone issues DROP TABLE under LOCK TABLES (generally, see all calls to release_all_locks_for_name()). MDL_context::release_ticket() is modified to take care of it. ****** A fix and a test case for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". An attempt to mix HANDLER SQL statements, which are transaction- agnostic, an open multi-statement transaction, and DDL against the involved tables (in a concurrent connection) could lead to a deadlock. The deadlock would occur when HANDLER OPEN or HANDLER READ would have to wait on a conflicting metadata lock. If the connection that issued HANDLER statement also had other metadata locks (say, acquired in scope of a transaction), a classical deadlock situation of mutual wait could occur. Incompatible change: entering LOCK TABLES mode automatically closes all open HANDLERs in the current connection. Incompatible change: previously an attempt to wait on a lock in a connection that has an open HANDLER statement could wait indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK is produced. The idea of the fix is to merge thd->handler_mdl_context with the main mdl_context of the connection, used for transactional locks. This makes deadlock detection possible, since all waits with locks are "visible" and available to analysis in a single MDL context of the connection. Since HANDLER locks and transactional locks have a different life cycle -- HANDLERs are explicitly open and closed, and so are HANDLER locks, explicitly acquired and released, whereas transactional locks "accumulate" till the end of a transaction and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT, a concept of "sentinel" was introduced to MDL_context. All locks, HANDLER and others, reside in the same linked list. However, a selected element of the list separates locks with different life cycle. HANDLER locks always reside at the end of the list, after the sentinel. Transactional locks are prepended to the beginning of the list, before the sentinel. Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only release those locks that reside before the sentinel. HANDLER locks must be released explicitly as part of HANDLER CLOSE statement, or an implicit close. The same approach with sentinel is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES statement has never worked together, the implementation is made simple and only maintains one sentinel, which is used either for HANDLER locks, or for LOCK TABLES locks.
2009-12-22 17:09:15 +01:00
table= share->free_tables.front();
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
table_def_use_table(thd, table);
/* We need to release share as we have EXTRA reference to it in our hands. */
release_table_share(share);
2000-07-31 21:29:14 +02:00
}
else
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/* We have too many TABLE instances around let us try to get rid of them. */
while (table_cache_count > table_cache_size && unused_tables)
free_cache_entry(unused_tables);
2000-07-31 21:29:14 +02:00
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/* make a new table */
if (!(table=(TABLE*) my_malloc(sizeof(*table),MYF(MY_WME))))
goto err_lock;
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
error= open_table_from_share(thd, share, alias,
(uint) (HA_OPEN_KEYFILE |
HA_OPEN_RNDFILE |
HA_GET_INDEX |
HA_TRY_READ_ONLY),
(READ_KEYINFO | COMPUTE_TYPES |
EXTRA_RECORD),
thd->open_options, table, FALSE);
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (error)
{
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(table);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (error == 7)
(void) ot_ctx->request_backoff_action(Open_table_context::OT_DISCOVER,
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
table_list);
else if (share->crashed)
(void) ot_ctx->request_backoff_action(Open_table_context::OT_REPAIR,
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
table_list);
2000-07-31 21:29:14 +02:00
goto err_lock;
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (open_table_entry_fini(thd, share, table))
2000-07-31 21:29:14 +02:00
{
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
closefrm(table, 0);
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(table);
goto err_lock;
2000-07-31 21:29:14 +02:00
}
mysql_mutex_lock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/* Add table to the share's used tables list. */
table_def_add_used_table(thd, table);
2000-07-31 21:29:14 +02:00
}
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
table->mdl_ticket= mdl_ticket;
table->next= thd->open_tables; /* Link into simple list */
thd->set_open_tables(table);
2000-07-31 21:29:14 +02:00
table->reginfo.lock_type=TL_READ; /* Assume read */
reset:
/*
Check that there is no reference to a condtion from an earlier query
(cf. Bug#58553).
*/
DBUG_ASSERT(table->file->pushed_cond == NULL);
2004-07-16 00:15:55 +02:00
table_list->updatable= 1; // It is not derived table nor non-updatable VIEW
table_list->table= table;
table->init(thd, table_list);
DBUG_RETURN(FALSE);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
err_lock:
mysql_mutex_lock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
err_unlock:
release_table_share(share);
2010-02-02 00:22:16 +01:00
mysql_mutex_unlock(&LOCK_open);
DBUG_RETURN(TRUE);
2000-07-31 21:29:14 +02:00
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/**
Find table in the list of open tables.
@param list List of TABLE objects to be inspected.
@param db Database name
@param table_name Table name
@return Pointer to the TABLE object found, 0 if no table found.
*/
TABLE *find_locked_table(TABLE *list, const char *db, const char *table_name)
2000-07-31 21:29:14 +02:00
{
char key[MAX_DBKEY_LENGTH];
uint key_length=(uint) (strmov(strmov(key,db)+1,table_name)-key)+1;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
for (TABLE *table= list; table ; table=table->next)
2000-07-31 21:29:14 +02:00
{
if (table->s->table_cache_key.length == key_length &&
!memcmp(table->s->table_cache_key.str, key, key_length))
2000-07-31 21:29:14 +02:00
return table;
}
return(0);
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/**
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
Find instance of TABLE with upgradable or exclusive metadata
lock from the list of open tables, emit error if no such table
found.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@param thd Thread context
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@param db Database name.
@param table_name Name of table.
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
@param no_error Don't emit error if no suitable TABLE
instance were found.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@note This function checks if the connection holds a global IX
metadata lock. If no such lock is found, it is not safe to
upgrade the lock and ER_TABLE_NOT_LOCKED_FOR_WRITE will be
reported.
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
@return Pointer to TABLE instance with MDL_SHARED_NO_WRITE,
MDL_SHARED_NO_READ_WRITE, or MDL_EXCLUSIVE metadata
lock, NULL otherwise.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
*/
TABLE *find_table_for_mdl_upgrade(THD *thd, const char *db,
const char *table_name, bool no_error)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
TABLE *tab= find_locked_table(thd->open_tables, db, table_name);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!tab)
{
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
if (!no_error)
my_error(ER_TABLE_NOT_LOCKED, MYF(0), table_name);
return NULL;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
/*
It is not safe to upgrade the metadata lock without a global IX lock.
This can happen with FLUSH TABLES <list> WITH READ LOCK as we in these
cases don't take a global IX lock in order to be compatible with
global read lock.
*/
if (!thd->mdl_context.is_lock_owner(MDL_key::GLOBAL, "", "",
MDL_INTENTION_EXCLUSIVE))
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
if (!no_error)
my_error(ER_TABLE_NOT_LOCKED_FOR_WRITE, MYF(0), table_name);
return NULL;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
while (tab->mdl_ticket != NULL &&
!tab->mdl_ticket->is_upgradable_or_exclusive() &&
(tab= find_locked_table(tab->next, db, table_name)))
continue;
if (!tab && !no_error)
my_error(ER_TABLE_NOT_LOCKED_FOR_WRITE, MYF(0), table_name);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
return tab;
}
/***********************************************************************
class Locked_tables_list implementation. Declared in sql_class.h
************************************************************************/
2000-07-31 21:29:14 +02:00
/**
Enter LTM_LOCK_TABLES mode.
Enter the LOCK TABLES mode using all the tables that are
currently open and locked in this connection.
Initializes a TABLE_LIST instance for every locked table.
@param thd thread handle
@return TRUE if out of memory.
*/
bool
Locked_tables_list::init_locked_tables(THD *thd)
2000-07-31 21:29:14 +02:00
{
DBUG_ASSERT(thd->locked_tables_mode == LTM_NONE);
DBUG_ASSERT(m_locked_tables == NULL);
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
DBUG_ASSERT(m_reopen_array == NULL);
DBUG_ASSERT(m_locked_tables_count == 0);
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
for (TABLE *table= thd->open_tables; table;
table= table->next, m_locked_tables_count++)
{
TABLE_LIST *src_table_list= table->pos_in_table_list;
char *db, *table_name, *alias;
size_t db_len= src_table_list->db_length;
size_t table_name_len= src_table_list->table_name_length;
size_t alias_len= strlen(src_table_list->alias);
TABLE_LIST *dst_table_list;
if (! multi_alloc_root(&m_locked_tables_root,
&dst_table_list, sizeof(*dst_table_list),
&db, db_len + 1,
&table_name, table_name_len + 1,
&alias, alias_len + 1,
NullS))
{
unlock_locked_tables(0);
return TRUE;
}
memcpy(db, src_table_list->db, db_len + 1);
memcpy(table_name, src_table_list->table_name, table_name_len + 1);
memcpy(alias, src_table_list->alias, alias_len + 1);
/**
Sic: remember the *actual* table level lock type taken, to
acquire the exact same type in reopen_tables().
E.g. if the table was locked for write, src_table_list->lock_type is
TL_WRITE_DEFAULT, whereas reginfo.lock_type has been updated from
thd->update_lock_default.
*/
dst_table_list->init_one_table(db, db_len, table_name, table_name_len,
alias,
src_table_list->table->reginfo.lock_type);
dst_table_list->table= table;
dst_table_list->mdl_request.ticket= src_table_list->mdl_request.ticket;
/* Link last into the list of tables */
*(dst_table_list->prev_global= m_locked_tables_last)= dst_table_list;
m_locked_tables_last= &dst_table_list->next_global;
table->pos_in_locked_tables= dst_table_list;
}
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
if (m_locked_tables_count)
{
/**
Allocate an auxiliary array to pass to mysql_lock_tables()
in reopen_tables(). reopen_tables() is a critical
path and we don't want to complicate it with extra allocations.
*/
m_reopen_array= (TABLE**)alloc_root(&m_locked_tables_root,
sizeof(TABLE*) *
(m_locked_tables_count+1));
if (m_reopen_array == NULL)
{
unlock_locked_tables(0);
return TRUE;
}
}
A prerequisite patch for the fix for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". Introduce a notion of a sentinel to MDL_context. A sentinel is a ticket that separates all tickets in the context into two groups: before and after it. Currently we can have (and need) only one designated sentinel -- it separates all locks taken by LOCK TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK and all other locks, which must be released at COMMIT or ROLLBACK. The tricky part is maintaining the sentinel up to date when someone release its corresponding ticket. This can happen, e.g. if someone issues DROP TABLE under LOCK TABLES (generally, see all calls to release_all_locks_for_name()). MDL_context::release_ticket() is modified to take care of it. ****** A fix and a test case for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". An attempt to mix HANDLER SQL statements, which are transaction- agnostic, an open multi-statement transaction, and DDL against the involved tables (in a concurrent connection) could lead to a deadlock. The deadlock would occur when HANDLER OPEN or HANDLER READ would have to wait on a conflicting metadata lock. If the connection that issued HANDLER statement also had other metadata locks (say, acquired in scope of a transaction), a classical deadlock situation of mutual wait could occur. Incompatible change: entering LOCK TABLES mode automatically closes all open HANDLERs in the current connection. Incompatible change: previously an attempt to wait on a lock in a connection that has an open HANDLER statement could wait indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK is produced. The idea of the fix is to merge thd->handler_mdl_context with the main mdl_context of the connection, used for transactional locks. This makes deadlock detection possible, since all waits with locks are "visible" and available to analysis in a single MDL context of the connection. Since HANDLER locks and transactional locks have a different life cycle -- HANDLERs are explicitly open and closed, and so are HANDLER locks, explicitly acquired and released, whereas transactional locks "accumulate" till the end of a transaction and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT, a concept of "sentinel" was introduced to MDL_context. All locks, HANDLER and others, reside in the same linked list. However, a selected element of the list separates locks with different life cycle. HANDLER locks always reside at the end of the list, after the sentinel. Transactional locks are prepended to the beginning of the list, before the sentinel. Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only release those locks that reside before the sentinel. HANDLER locks must be released explicitly as part of HANDLER CLOSE statement, or an implicit close. The same approach with sentinel is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES statement has never worked together, the implementation is made simple and only maintains one sentinel, which is used either for HANDLER locks, or for LOCK TABLES locks.
2009-12-22 17:09:15 +01:00
thd->enter_locked_tables_mode(LTM_LOCK_TABLES);
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
return FALSE;
}
2000-07-31 21:29:14 +02:00
/**
Leave LTM_LOCK_TABLES mode if it's been entered.
Close all locked tables, free memory, and leave the mode.
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
@note This function is a no-op if we're not in LOCK TABLES.
*/
2000-07-31 21:29:14 +02:00
void
Locked_tables_list::unlock_locked_tables(THD *thd)
2000-07-31 21:29:14 +02:00
{
if (thd)
{
DBUG_ASSERT(!thd->in_sub_stmt &&
!(thd->state_flags & Open_tables_state::BACKUPS_AVAIL));
/*
Sic: we must be careful to not close open tables if
we're not in LOCK TABLES mode: unlock_locked_tables() is
sometimes called implicitly, expecting no effect on
open tables, e.g. from begin_trans().
*/
if (thd->locked_tables_mode != LTM_LOCK_TABLES)
return;
for (TABLE_LIST *table_list= m_locked_tables;
table_list; table_list= table_list->next_global)
{
/*
Clear the position in the list, the TABLE object will be
returned to the table cache.
*/
table_list->table->pos_in_locked_tables= NULL;
}
A prerequisite patch for the fix for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". Introduce a notion of a sentinel to MDL_context. A sentinel is a ticket that separates all tickets in the context into two groups: before and after it. Currently we can have (and need) only one designated sentinel -- it separates all locks taken by LOCK TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK and all other locks, which must be released at COMMIT or ROLLBACK. The tricky part is maintaining the sentinel up to date when someone release its corresponding ticket. This can happen, e.g. if someone issues DROP TABLE under LOCK TABLES (generally, see all calls to release_all_locks_for_name()). MDL_context::release_ticket() is modified to take care of it. ****** A fix and a test case for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". An attempt to mix HANDLER SQL statements, which are transaction- agnostic, an open multi-statement transaction, and DDL against the involved tables (in a concurrent connection) could lead to a deadlock. The deadlock would occur when HANDLER OPEN or HANDLER READ would have to wait on a conflicting metadata lock. If the connection that issued HANDLER statement also had other metadata locks (say, acquired in scope of a transaction), a classical deadlock situation of mutual wait could occur. Incompatible change: entering LOCK TABLES mode automatically closes all open HANDLERs in the current connection. Incompatible change: previously an attempt to wait on a lock in a connection that has an open HANDLER statement could wait indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK is produced. The idea of the fix is to merge thd->handler_mdl_context with the main mdl_context of the connection, used for transactional locks. This makes deadlock detection possible, since all waits with locks are "visible" and available to analysis in a single MDL context of the connection. Since HANDLER locks and transactional locks have a different life cycle -- HANDLERs are explicitly open and closed, and so are HANDLER locks, explicitly acquired and released, whereas transactional locks "accumulate" till the end of a transaction and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT, a concept of "sentinel" was introduced to MDL_context. All locks, HANDLER and others, reside in the same linked list. However, a selected element of the list separates locks with different life cycle. HANDLER locks always reside at the end of the list, after the sentinel. Transactional locks are prepended to the beginning of the list, before the sentinel. Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only release those locks that reside before the sentinel. HANDLER locks must be released explicitly as part of HANDLER CLOSE statement, or an implicit close. The same approach with sentinel is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES statement has never worked together, the implementation is made simple and only maintains one sentinel, which is used either for HANDLER locks, or for LOCK TABLES locks.
2009-12-22 17:09:15 +01:00
thd->leave_locked_tables_mode();
DBUG_ASSERT(thd->transaction.stmt.is_empty());
close_thread_tables(thd);
/*
We rely on the caller to implicitly commit the
transaction and release transactional locks.
*/
}
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
/*
After closing tables we can free memory used for storing lock
request for metadata locks and TABLE_LIST elements.
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
*/
free_root(&m_locked_tables_root, MYF(0));
m_locked_tables= NULL;
m_locked_tables_last= &m_locked_tables;
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
m_reopen_array= NULL;
m_locked_tables_count= 0;
2000-07-31 21:29:14 +02:00
}
/**
Unlink a locked table from the locked tables list, either
temporarily or permanently.
@param thd thread handle
@param table_list the element of locked tables list.
The implementation assumes that this argument
points to a TABLE_LIST element linked into
the locked tables list. Passing a TABLE_LIST
instance that is not part of locked tables
list will lead to a crash.
@param remove_from_locked_tables
TRUE if the table is removed from the list
permanently.
This function is a no-op if we're not under LOCK TABLES.
@sa Locked_tables_list::reopen_tables()
*/
2000-07-31 21:29:14 +02:00
void Locked_tables_list::unlink_from_list(THD *thd,
TABLE_LIST *table_list,
bool remove_from_locked_tables)
2000-07-31 21:29:14 +02:00
{
/*
If mode is not LTM_LOCK_TABLES, we needn't do anything. Moreover,
outside this mode pos_in_locked_tables value is not trustworthy.
*/
if (thd->locked_tables_mode != LTM_LOCK_TABLES)
return;
/*
table_list must be set and point to pos_in_locked_tables of some
table.
*/
DBUG_ASSERT(table_list->table->pos_in_locked_tables == table_list);
/* Clear the pointer, the table will be returned to the table cache. */
table_list->table->pos_in_locked_tables= NULL;
/* Mark the table as closed in the locked tables list. */
table_list->table= NULL;
/*
If the table is being dropped or renamed, remove it from
the locked tables list (implicitly drop the LOCK TABLES lock
on it).
*/
if (remove_from_locked_tables)
2000-07-31 21:29:14 +02:00
{
*table_list->prev_global= table_list->next_global;
if (table_list->next_global == NULL)
m_locked_tables_last= table_list->prev_global;
else
table_list->next_global->prev_global= table_list->prev_global;
2000-07-31 21:29:14 +02:00
}
}
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
/**
This is an attempt to recover (somewhat) in case of an error.
If we failed to reopen a closed table, let's unlink it from the
list and forget about it. From a user perspective that would look
as if the server "lost" the lock on one of the locked tables.
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
@note This function is a no-op if we're not under LOCK TABLES.
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
*/
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
void Locked_tables_list::
unlink_all_closed_tables(THD *thd, MYSQL_LOCK *lock, size_t reopen_count)
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
{
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
/* If we managed to take a lock, unlock tables and free the lock. */
if (lock)
mysql_unlock_tables(thd, lock);
/*
If a failure happened in reopen_tables(), we may have succeeded
reopening some tables, but not all.
This works when the connection was killed in mysql_lock_tables().
*/
if (reopen_count)
{
while (reopen_count--)
{
/*
When closing the table, we must remove it
from thd->open_tables list.
We rely on the fact that open_table() that was used
in reopen_tables() always links the opened table
to the beginning of the open_tables list.
*/
DBUG_ASSERT(thd->open_tables == m_reopen_array[reopen_count]);
thd->open_tables->pos_in_locked_tables->table= NULL;
close_thread_table(thd, &thd->open_tables);
}
}
/* Exclude all closed tables from the LOCK TABLES list. */
for (TABLE_LIST *table_list= m_locked_tables; table_list; table_list=
table_list->next_global)
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
{
if (table_list->table == NULL)
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
{
/* Unlink from list. */
*table_list->prev_global= table_list->next_global;
if (table_list->next_global == NULL)
m_locked_tables_last= table_list->prev_global;
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
else
table_list->next_global->prev_global= table_list->prev_global;
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
}
}
}
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
/**
Reopen the tables locked with LOCK TABLES and temporarily closed
by a DDL statement or FLUSH TABLES.
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
@note This function is a no-op if we're not under LOCK TABLES.
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
@return TRUE if an error reopening the tables. May happen in
case of some fatal system error only, e.g. a disk
corruption, out of memory or a serious bug in the
locking.
2000-07-31 21:29:14 +02:00
*/
bool
Locked_tables_list::reopen_tables(THD *thd)
2000-07-31 21:29:14 +02:00
{
Open_table_context ot_ctx(thd, MYSQL_OPEN_REOPEN);
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
size_t reopen_count= 0;
MYSQL_LOCK *lock;
MYSQL_LOCK *merged_lock;
2000-07-31 21:29:14 +02:00
for (TABLE_LIST *table_list= m_locked_tables;
table_list; table_list= table_list->next_global)
2000-07-31 21:29:14 +02:00
{
if (table_list->table) /* The table was not closed */
continue;
/* Links into thd->open_tables upon success */
if (open_table(thd, table_list, thd->mem_root, &ot_ctx))
2000-07-31 21:29:14 +02:00
{
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
unlink_all_closed_tables(thd, 0, reopen_count);
return TRUE;
2000-07-31 21:29:14 +02:00
}
table_list->table->pos_in_locked_tables= table_list;
/* See also the comment on lock type in init_locked_tables(). */
table_list->table->reginfo.lock_type= table_list->lock_type;
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
DBUG_ASSERT(reopen_count < m_locked_tables_count);
m_reopen_array[reopen_count++]= table_list->table;
}
if (reopen_count)
{
thd->in_lock_tables= 1;
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
/*
We re-lock all tables with mysql_lock_tables() at once rather
than locking one table at a time because of the case
reported in Bug#45035: when the same table is present
in the list many times, thr_lock.c fails to grant READ lock
on a table that is already locked by WRITE lock, even if
WRITE lock is taken by the same thread. If READ and WRITE
lock are passed to thr_lock.c in the same list, everything
works fine. Patching legacy code of thr_lock.c is risking to
break something else.
*/
lock= mysql_lock_tables(thd, m_reopen_array, reopen_count,
MYSQL_OPEN_REOPEN);
thd->in_lock_tables= 0;
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
if (lock == NULL || (merged_lock=
mysql_lock_merge(thd->lock, lock)) == NULL)
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
{
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
unlink_all_closed_tables(thd, lock, reopen_count);
if (! thd->killed)
my_error(ER_LOCK_DEADLOCK, MYF(0));
return TRUE;
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
}
Backport of: ---------------------------------------------------------- revno: 2617.69.2 committer: Konstantin Osipov <kostja@sun.com> branch nick: 5.4-azalea-bugfixing timestamp: Mon 2009-08-03 19:26:04 +0400 message: A fix and a test case for Bug#45035 "Altering table under LOCK TABLES results in "Error 1213 Deadlock found...". If a user had a table locked with LOCK TABLES for READ and for WRITE in the same connection, ALTER TABLE could fail. Root cause analysis: If a connection issues LOCK TABLE t1 write, t1 a read, t1 b read; the new LOCK TABLES code in 6.0 (part of WL 3726) will create the following list of TABLE_LIST objects (thd->locked_tables_list->m_locked_tables): {"t1" "b" tl_read_no_insert}, {"t1" "a" tl_read_no_insert}, {"t1" "t1" tl_write } Later on, when we try to ALTER table t1, mysql_alter_table() closes all TABLE instances and releases its thr_lock locks, keeping only an exclusive metadata lock on t1. But when ALTER is finished, Locked_table_list::reopen_tables() tries to restore the original list of open and locked tables. Before this patch, it used to do so one by one: Open t1 b, get TL_READ_NO_INSERT lock, Open t1 a, get TL_READ_NO_INSERT lock Open t1, try to get TL_WRITE lock, deadlock. The cause of the deadlock is that thr_lock.c doesn't resolve the situation when the read list only consists of locks taken by the same thread, followed by this very thread trying to take a WRITE lock. Indeed, since thr_lock_multi always gets a sorted list of locks, WRITE locks always precede READ locks in the list to lock. Don't try to fix thr_lock.c deficiency, keep this code simple. Instead, try to take all thr_lock locks at once in ::reopen_tables().
2009-12-08 09:38:45 +01:00
thd->lock= merged_lock;
2000-07-31 21:29:14 +02:00
}
return FALSE;
2000-07-31 21:29:14 +02:00
}
/*
Function to assign a new table map id to a table share.
PARAMETERS
share - Pointer to table share structure
2006-02-27 17:23:20 +01:00
DESCRIPTION
We are intentionally not checking that share->mutex is locked
since this function should only be called when opening a table
share and before it is entered into the table_def_cache (meaning
that it cannot be fetched by another thread, even accidentally).
PRE-CONDITION(S)
share is non-NULL
The LOCK_open mutex is locked.
POST-CONDITION(S)
share->table_map_id is given a value that with a high certainty is
not used by any other table (the only case where a table id can be
reused is on wrap-around, which means more than 4 billion table
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
share opens have been executed while one table was open all the
time).
share->table_map_id is not ~0UL.
*/
static ulong last_table_id= ~0UL;
void assign_new_table_id(TABLE_SHARE *share)
{
DBUG_ENTER("assign_new_table_id");
/* Preconditions */
DBUG_ASSERT(share != NULL);
mysql_mutex_assert_owner(&LOCK_open);
ulong tid= ++last_table_id; /* get next id */
/*
There is one reserved number that cannot be used. Remember to
change this when 6-byte global table id's are introduced.
*/
if (unlikely(tid == ~0UL))
tid= ++last_table_id;
share->table_map_id= tid;
DBUG_PRINT("info", ("table_id=%lu", tid));
/* Post conditions */
DBUG_ASSERT(share->table_map_id != ~0UL);
DBUG_VOID_RETURN;
}
#ifndef DBUG_OFF
/* Cause a spurious statement reprepare for debug purposes. */
static bool inject_reprepare(THD *thd)
{
if (thd->m_reprepare_observer && thd->stmt_arena->is_reprepared == FALSE)
{
thd->m_reprepare_observer->report_error(thd);
return TRUE;
}
return FALSE;
}
#endif
/**
Compare metadata versions of an element obtained from the table
definition cache and its corresponding node in the parse tree.
@details If the new and the old values mismatch, invoke
Metadata_version_observer.
At prepared statement prepare, all TABLE_LIST version values are
NULL and we always have a mismatch. But there is no observer set
in THD, and therefore no error is reported. Instead, we update
the value in the parse tree, effectively recording the original
version.
At prepared statement execute, an observer may be installed. If
there is a version mismatch, we push an error and return TRUE.
For conventional execution (no prepared statements), the
observer is never installed.
@sa Execute_observer
@sa check_prepared_statement() to see cases when an observer is installed
@sa TABLE_LIST::is_table_ref_id_equal()
@sa TABLE_SHARE::get_table_ref_id()
@param[in] thd used to report errors
@param[in,out] tables TABLE_LIST instance created by the parser
Metadata version information in this object
is updated upon success.
@param[in] table_share an element from the table definition cache
@retval TRUE an error, which has been reported
@retval FALSE success, version in TABLE_LIST has been updated
*/
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
static bool
check_and_update_table_version(THD *thd,
TABLE_LIST *tables, TABLE_SHARE *table_share)
{
if (! tables->is_table_ref_id_equal(table_share))
{
if (thd->m_reprepare_observer &&
thd->m_reprepare_observer->report_error(thd))
{
/*
Version of the table share is different from the
previous execution of the prepared statement, and it is
unacceptable for this SQLCOM. Error has been reported.
*/
DBUG_ASSERT(thd->is_error());
return TRUE;
}
/* Always maintain the latest version and type */
tables->set_table_ref_id(table_share);
}
DBUG_EXECUTE_IF("reprepare_each_statement", return inject_reprepare(thd););
return FALSE;
}
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
/**
Compares versions of a stored routine obtained from the sp cache
and the version used at prepare.
@details If the new and the old values mismatch, invoke
Metadata_version_observer.
At prepared statement prepare, all Sroutine_hash_entry version values
are NULL and we always have a mismatch. But there is no observer set
in THD, and therefore no error is reported. Instead, we update
the value in Sroutine_hash_entry, effectively recording the original
version.
At prepared statement execute, an observer may be installed. If
there is a version mismatch, we push an error and return TRUE.
For conventional execution (no prepared statements), the
observer is never installed.
@param[in] thd used to report errors
@param[in/out] rt pointer to stored routine entry in the
parse tree
@param[in] sp pointer to stored routine cache entry.
Can be NULL if there is no such routine.
@retval TRUE an error, which has been reported
@retval FALSE success, version in Sroutine_hash_entry has been updated
*/
static bool
check_and_update_routine_version(THD *thd, Sroutine_hash_entry *rt,
sp_head *sp)
{
ulong spc_version= sp_cache_version();
/* sp is NULL if there is no such routine. */
ulong version= sp ? sp->sp_cache_version() : spc_version;
/*
If the version in the parse tree is stale,
or the version in the cache is stale and sp is not used,
we need to reprepare.
Sic: version != spc_version <--> sp is not NULL.
*/
if (rt->m_sp_cache_version != version ||
(version != spc_version && !sp->is_invoked()))
{
if (thd->m_reprepare_observer &&
thd->m_reprepare_observer->report_error(thd))
{
/*
Version of the sp cache is different from the
previous execution of the prepared statement, and it is
unacceptable for this SQLCOM. Error has been reported.
*/
DBUG_ASSERT(thd->is_error());
return TRUE;
}
/* Always maintain the latest cache version. */
rt->m_sp_cache_version= version;
}
return FALSE;
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/**
Open view by getting its definition from disk (and table cache in future).
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@param thd Thread handle
@param table_list TABLE_LIST with db, table_name & belong_to_view
@param alias Alias name
@param cache_key Key for table definition cache
@param cache_key_length Length of cache_key
@param mem_root Memory to be used for .frm parsing.
@param flags Flags which modify how we open the view
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@todo This function is needed for special handling of views under
LOCK TABLES. We probably should get rid of it in long term.
@return FALSE if success, TRUE - otherwise.
*/
bool tdc_open_view(THD *thd, TABLE_LIST *table_list, const char *alias,
char *cache_key, uint cache_key_length,
MEM_ROOT *mem_root, uint flags)
{
TABLE not_used;
int error;
2010-02-03 14:43:03 +01:00
my_hash_value_type hash_value;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
TABLE_SHARE *share;
2010-02-03 14:43:03 +01:00
hash_value= my_calc_hash(&table_def_cache, (uchar*) cache_key,
cache_key_length);
2010-02-02 00:22:16 +01:00
mysql_mutex_lock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!(share= get_table_share(thd, table_list, cache_key,
cache_key_length,
OPEN_VIEW, &error,
hash_value)))
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
goto err;
if (share->is_view &&
!open_new_frm(thd, share, alias,
(uint) (HA_OPEN_KEYFILE | HA_OPEN_RNDFILE |
HA_GET_INDEX | HA_TRY_READ_ONLY),
READ_KEYINFO | COMPUTE_TYPES | EXTRA_RECORD |
flags, thd->open_options, &not_used, table_list,
mem_root))
{
release_table_share(share);
2010-02-02 00:22:16 +01:00
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
return FALSE;
}
my_error(ER_WRONG_OBJECT, MYF(0), share->db.str, share->table_name.str, "VIEW");
release_table_share(share);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
err:
2010-02-02 00:22:16 +01:00
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
return TRUE;
}
/**
Finalize the process of TABLE creation by loading table triggers
and taking action if a HEAP table content was emptied implicitly.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
*/
static bool open_table_entry_fini(THD *thd, TABLE_SHARE *share, TABLE *entry)
{
if (Table_triggers_list::check_n_load(thd, share->db.str,
share->table_name.str, entry, 0))
return TRUE;
2004-06-25 15:52:01 +02:00
/*
If we are here, there was no fatal error (but error may be still
unitialized).
*/
if (unlikely(entry->file->implicit_emptied))
{
entry->file->implicit_emptied= 0;
if (mysql_bin_log.is_open())
{
char *query, *end;
uint query_buf_size= 20 + share->db.length + share->table_name.length +1;
if ((query= (char*) my_malloc(query_buf_size,MYF(MY_WME))))
2004-06-25 15:52:01 +02:00
{
/* this DELETE FROM is needed even with row-based binlogging */
2004-06-25 15:52:01 +02:00
end = strxmov(strmov(query, "DELETE FROM `"),
share->db.str,"`.`",share->table_name.str,"`", NullS);
int errcode= query_error_code(thd, TRUE);
if (thd->binlog_query(THD::STMT_QUERY_TYPE,
query, (ulong)(end-query),
merge mysql-5.1-rep+2-delivery1 --> mysql-5.1-rpl-merge Conflicts: Text conflict in .bzr-mysql/default.conf Text conflict in mysql-test/extra/rpl_tests/rpl_loaddata.test Text conflict in mysql-test/r/mysqlbinlog2.result Text conflict in mysql-test/suite/binlog/r/binlog_stm_mix_innodb_myisam.result Text conflict in mysql-test/suite/binlog/r/binlog_unsafe.result Text conflict in mysql-test/suite/rpl/r/rpl_insert_id.result Text conflict in mysql-test/suite/rpl/r/rpl_loaddata.result Text conflict in mysql-test/suite/rpl/r/rpl_stm_auto_increment_bug33029.result Text conflict in mysql-test/suite/rpl/r/rpl_udf.result Text conflict in mysql-test/suite/rpl/t/rpl_slow_query_log.test Text conflict in sql/field.h Text conflict in sql/log.cc Text conflict in sql/log_event.cc Text conflict in sql/log_event_old.cc Text conflict in sql/mysql_priv.h Text conflict in sql/share/errmsg.txt Text conflict in sql/sp.cc Text conflict in sql/sql_acl.cc Text conflict in sql/sql_base.cc Text conflict in sql/sql_class.h Text conflict in sql/sql_db.cc Text conflict in sql/sql_delete.cc Text conflict in sql/sql_insert.cc Text conflict in sql/sql_lex.cc Text conflict in sql/sql_lex.h Text conflict in sql/sql_load.cc Text conflict in sql/sql_table.cc Text conflict in sql/sql_update.cc Text conflict in sql/sql_view.cc Conflict adding files to storage/innobase. Created directory. Conflict because storage/innobase is not versioned, but has versioned children. Versioned directory. Conflict adding file storage/innobase. Moved existing file to storage/innobase.moved. Conflict adding files to storage/innobase/handler. Created directory. Conflict because storage/innobase/handler is not versioned, but has versioned children. Versioned directory. Contents conflict in storage/innobase/handler/ha_innodb.cc
2010-01-07 16:39:11 +01:00
FALSE, FALSE, FALSE, errcode))
{
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(query);
return TRUE;
}
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(query);
2004-06-25 15:52:01 +02:00
}
else
{
/*
As replication is maybe going to be corrupted, we need to warn the
DBA on top of warning the client (which will automatically be done
because of MYF(MY_WME) in my_malloc() above).
*/
sql_print_error("When opening HEAP table, could not allocate memory "
"to write 'DELETE FROM `%s`.`%s`' to the binary log",
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
share->db.str, share->table_name.str);
delete entry->triggers;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
return TRUE;
2004-06-25 15:52:01 +02:00
}
}
2000-07-31 21:29:14 +02:00
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
return FALSE;
}
/**
Auxiliary routine which is used for performing automatical table repair.
*/
static bool auto_repair_table(THD *thd, TABLE_LIST *table_list)
{
char cache_key[MAX_DBKEY_LENGTH];
uint cache_key_length;
TABLE_SHARE *share;
TABLE *entry;
int not_used;
bool result= TRUE;
2010-02-03 14:43:03 +01:00
my_hash_value_type hash_value;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
cache_key_length= create_table_def_key(thd, cache_key, table_list, 0);
thd->clear_error();
2010-02-03 14:43:03 +01:00
hash_value= my_calc_hash(&table_def_cache, (uchar*) cache_key,
cache_key_length);
2010-02-02 00:22:16 +01:00
mysql_mutex_lock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!(share= get_table_share(thd, table_list, cache_key,
cache_key_length,
OPEN_VIEW, &not_used,
hash_value)))
goto end_unlock;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (share->is_view)
{
release_table_share(share);
goto end_unlock;
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (!(entry= (TABLE*)my_malloc(sizeof(TABLE), MYF(MY_WME))))
{
release_table_share(share);
goto end_unlock;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
2010-02-02 00:22:16 +01:00
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (open_table_from_share(thd, share, table_list->alias,
(uint) (HA_OPEN_KEYFILE | HA_OPEN_RNDFILE |
HA_GET_INDEX |
HA_TRY_READ_ONLY),
READ_KEYINFO | COMPUTE_TYPES | EXTRA_RECORD,
ha_open_options | HA_OPEN_FOR_REPAIR,
entry, FALSE) || ! entry->file ||
(entry->file->is_crashed() && entry->file->ha_check_and_repair(thd)))
{
/* Give right error message */
thd->clear_error();
2011-05-21 10:21:08 +02:00
my_error(ER_NOT_KEYFILE, MYF(0), share->table_name.str);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
sql_print_error("Couldn't repair table: %s.%s", share->db.str,
share->table_name.str);
if (entry->file)
closefrm(entry, 0);
}
else
{
thd->clear_error(); // Clear error message
closefrm(entry, 0);
result= FALSE;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(entry);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
2010-02-02 00:22:16 +01:00
mysql_mutex_lock(&LOCK_open);
release_table_share(share);
/* Remove the repaired share from the table cache. */
tdc_remove_table(thd, TDC_RT_REMOVE_ALL,
table_list->db, table_list->table_name,
TRUE);
end_unlock:
2010-02-02 00:22:16 +01:00
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
return result;
}
/** Open_table_context */
Open_table_context::Open_table_context(THD *thd, uint flags)
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
:m_failed_table(NULL),
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
m_start_of_statement_svp(thd->mdl_context.mdl_savepoint()),
m_timeout(flags & MYSQL_LOCK_IGNORE_TIMEOUT ?
LONG_TIMEOUT : thd->variables.lock_wait_timeout),
m_flags(flags),
m_action(OT_NO_ACTION),
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
m_has_locks(thd->mdl_context.has_locks()),
m_has_protection_against_grl(FALSE)
{}
/**
Check if we can back-off and set back off action if we can.
Otherwise report and return error.
@retval TRUE if back-off is impossible.
@retval FALSE if we can back off. Back off action has been set.
*/
bool
Open_table_context::
request_backoff_action(enum_open_table_action action_arg,
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
TABLE_LIST *table)
{
/*
A back off action may be one of three kinds:
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
* We met a broken table that needs repair, or a table that
is not present on this MySQL server and needs re-discovery.
To perform the action, we need an exclusive metadata lock on
the table. Acquiring an X lock while holding other shared
locks is very deadlock-prone. If this is a multi- statement
transaction that holds metadata locks for completed
statements, we don't do it, and report an error instead.
The action type in this case is OT_DISCOVER or OT_REPAIR.
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
* Our attempt to acquire an MDL lock lead to a deadlock,
detected by the MDL deadlock detector. The current
session was chosen a victim. If this is a multi-statement
transaction that holds metadata locks taken by completed
statements, restarting locking for the current statement
may lead to a livelock. Releasing locks of completed
statements can not be done as will lead to violation
of ACID. Thus, again, if m_has_locks is set,
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
we report an error. Otherwise, when there are no metadata
locks other than which belong to this statement, we can
try to recover from error by releasing all locks and
restarting the pre-locking.
Similarly, a deadlock error can occur when the
pre-locking process met a TABLE_SHARE that is being
flushed, and unsuccessfully waited for the flush to
complete. A deadlock in this case can happen, e.g.,
when our session is holding a metadata lock that
is being waited on by a session which is using
the table which is being flushed. The only way
to recover from this error is, again, to close all
open tables, release all locks, and retry pre-locking.
Action type name is OT_REOPEN_TABLES. Re-trying
while holding some locks may lead to a livelock,
and thus we don't do it.
* Finally, this session has open TABLEs from different
"generations" of the table cache. This can happen, e.g.,
when, after this session has successfully opened one
table used for a statement, FLUSH TABLES interfered and
expelled another table used in it. FLUSH TABLES then
blocks and waits on the table already opened by this
statement.
We detect this situation by ensuring that table cache
version of all tables used in a statement is the same.
If it isn't, all tables needs to be reopened.
Note, that we can always perform a reopen in this case,
even if we already have metadata locks, since we don't
keep tables open between statements and a livelock
is not possible.
*/
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
if (action_arg != OT_REOPEN_TABLES && m_has_locks)
{
my_error(ER_LOCK_DEADLOCK, MYF(0));
return TRUE;
}
/*
If auto-repair or discovery are requested, a pointer to table
list element must be provided.
*/
if (table)
{
DBUG_ASSERT(action_arg == OT_DISCOVER || action_arg == OT_REPAIR);
m_failed_table= (TABLE_LIST*) current_thd->alloc(sizeof(TABLE_LIST));
if (m_failed_table == NULL)
return TRUE;
m_failed_table->init_one_table(table->db, table->db_length,
table->table_name,
table->table_name_length,
table->alias, TL_WRITE);
m_failed_table->mdl_request.set_type(MDL_EXCLUSIVE);
}
m_action= action_arg;
return FALSE;
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/**
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
Recover from failed attempt of open table by performing requested action.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@param thd Thread context
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
@pre This function should be called only with "action" != OT_NO_ACTION
and after having called @sa close_tables_for_reopen().
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@retval FALSE - Success. One should try to open tables once again.
@retval TRUE - Error
*/
bool
Open_table_context::
recover_from_failed_open(THD *thd)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
bool result= FALSE;
/* Execute the action. */
switch (m_action)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
case OT_BACKOFF_AND_RETRY:
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
break;
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
case OT_REOPEN_TABLES:
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
break;
case OT_DISCOVER:
{
if ((result= lock_table_names(thd, m_failed_table, NULL,
get_timeout(),
MYSQL_OPEN_SKIP_TEMPORARY)))
break;
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
tdc_remove_table(thd, TDC_RT_REMOVE_ALL, m_failed_table->db,
m_failed_table->table_name, FALSE);
ha_create_table_from_engine(thd, m_failed_table->db,
m_failed_table->table_name);
thd->warning_info->clear_warning_info(thd->query_id);
thd->clear_error(); // Clear error message
thd->mdl_context.release_transactional_locks();
break;
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
case OT_REPAIR:
{
if ((result= lock_table_names(thd, m_failed_table, NULL,
get_timeout(),
MYSQL_OPEN_SKIP_TEMPORARY)))
break;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
tdc_remove_table(thd, TDC_RT_REMOVE_ALL, m_failed_table->db,
m_failed_table->table_name, FALSE);
result= auto_repair_table(thd, m_failed_table);
thd->mdl_context.release_transactional_locks();
break;
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
default:
DBUG_ASSERT(0);
}
/*
Reset the pointers to conflicting MDL request and the
TABLE_LIST element, set when we need auto-discovery or repair,
for safety.
*/
m_failed_table= NULL;
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
/*
Reset flag indicating that we have already acquired protection
against GRL. It is no longer valid as the corresponding lock was
released by close_tables_for_reopen().
*/
m_has_protection_against_grl= FALSE;
/* Prepare for possible another back-off. */
m_action= OT_NO_ACTION;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
return result;
2000-07-31 21:29:14 +02:00
}
/*
Return a appropriate read lock type given a table object.
@param thd Thread context
Committing on behalf or Dmitry Lenev: Fix for bug #46947 "Embedded SELECT without FOR UPDATE is causing a lock", with after-review fixes. SELECT statements with subqueries referencing InnoDB tables were acquiring shared locks on rows in these tables when they were executed in REPEATABLE-READ mode and with statement or mixed mode binary logging turned on. This was a regression which were introduced when fixing bug 39843. The problem was that for tables belonging to subqueries parser set TL_READ_DEFAULT as a lock type. In cases when statement/mixed binary logging at open_tables() time this type of lock was converted to TL_READ_NO_INSERT lock at open_tables() time and caused InnoDB engine to acquire shared locks on reads from these tables. Although in some cases such behavior was correct (e.g. for subqueries in DELETE) in case of SELECT it has caused unnecessary locking. This patch tries to solve this problem by rethinking our approach to how we handle locking for SELECT and subqueries. Now we always set TL_READ_DEFAULT lock type for all cases when we read data. When at open_tables() time this lock is interpreted as TL_READ_NO_INSERT or TL_READ depending on whether this statement as a whole or call to function which uses particular table should be written to the binary log or not (if yes then statement should be properly serialized with concurrent statements and stronger lock should be acquired). Test coverage is added for both InnoDB and MyISAM. This patch introduces an "incompatible" change in locking scheme for subqueries used in SELECT ... FOR UPDATE and SELECT .. IN SHARE MODE. In 4.1 the server would use a snapshot InnoDB read for subqueries in SELECT FOR UPDATE and SELECT .. IN SHARE MODE statements, regardless of whether the binary log is on or off. If the user required a different type of read (i.e. locking read), he/she could request so explicitly by providing FOR UPDATE/IN SHARE MODE clause for each individual subquery. On of the patches for 5.0 broke this behaviour (which was not documented or tested), and started to use locking reads fora all subqueries in SELECT ... FOR UPDATE/IN SHARE MODE. This patch restored 4.1 behaviour.
2010-04-28 12:04:11 +02:00
@param prelocking_ctx Prelocking context.
@param table_list Table list element for table to be locked.
@remark Due to a statement-based replication limitation, statements such as
INSERT INTO .. SELECT FROM .. and CREATE TABLE .. SELECT FROM need
to grab a TL_READ_NO_INSERT lock on the source table in order to
prevent the replication of a concurrent statement that modifies the
source table. If such a statement gets applied on the slave before
the INSERT .. SELECT statement finishes, data on the master could
differ from data on the slave and end-up with a discrepancy between
Committing on behalf or Dmitry Lenev: Fix for bug #46947 "Embedded SELECT without FOR UPDATE is causing a lock", with after-review fixes. SELECT statements with subqueries referencing InnoDB tables were acquiring shared locks on rows in these tables when they were executed in REPEATABLE-READ mode and with statement or mixed mode binary logging turned on. This was a regression which were introduced when fixing bug 39843. The problem was that for tables belonging to subqueries parser set TL_READ_DEFAULT as a lock type. In cases when statement/mixed binary logging at open_tables() time this type of lock was converted to TL_READ_NO_INSERT lock at open_tables() time and caused InnoDB engine to acquire shared locks on reads from these tables. Although in some cases such behavior was correct (e.g. for subqueries in DELETE) in case of SELECT it has caused unnecessary locking. This patch tries to solve this problem by rethinking our approach to how we handle locking for SELECT and subqueries. Now we always set TL_READ_DEFAULT lock type for all cases when we read data. When at open_tables() time this lock is interpreted as TL_READ_NO_INSERT or TL_READ depending on whether this statement as a whole or call to function which uses particular table should be written to the binary log or not (if yes then statement should be properly serialized with concurrent statements and stronger lock should be acquired). Test coverage is added for both InnoDB and MyISAM. This patch introduces an "incompatible" change in locking scheme for subqueries used in SELECT ... FOR UPDATE and SELECT .. IN SHARE MODE. In 4.1 the server would use a snapshot InnoDB read for subqueries in SELECT FOR UPDATE and SELECT .. IN SHARE MODE statements, regardless of whether the binary log is on or off. If the user required a different type of read (i.e. locking read), he/she could request so explicitly by providing FOR UPDATE/IN SHARE MODE clause for each individual subquery. On of the patches for 5.0 broke this behaviour (which was not documented or tested), and started to use locking reads fora all subqueries in SELECT ... FOR UPDATE/IN SHARE MODE. This patch restored 4.1 behaviour.
2010-04-28 12:04:11 +02:00
the binary log and table state.
This also applies to SELECT/SET/DO statements which use stored
functions. Calls to such functions are going to be logged as a
whole and thus should be serialized against concurrent changes
to tables used by those functions. This can be avoided if functions
only read data but doing so requires more complex analysis than it
is done now.
Furthermore, this does not apply to I_S and log tables as it's
always unsafe to replicate such tables under statement-based
replication as the table on the slave might contain other data
(ie: general_log is enabled on the slave). The statement will
be marked as unsafe for SBR in decide_logging_format().
@remark Note that even in prelocked mode it is important to correctly
determine lock type value. In this mode lock type is passed to
handler::start_stmt() method and can be used by storage engine,
for example, to determine what kind of row locks it should acquire
when reading data from the table.
*/
Committing on behalf or Dmitry Lenev: Fix for bug #46947 "Embedded SELECT without FOR UPDATE is causing a lock", with after-review fixes. SELECT statements with subqueries referencing InnoDB tables were acquiring shared locks on rows in these tables when they were executed in REPEATABLE-READ mode and with statement or mixed mode binary logging turned on. This was a regression which were introduced when fixing bug 39843. The problem was that for tables belonging to subqueries parser set TL_READ_DEFAULT as a lock type. In cases when statement/mixed binary logging at open_tables() time this type of lock was converted to TL_READ_NO_INSERT lock at open_tables() time and caused InnoDB engine to acquire shared locks on reads from these tables. Although in some cases such behavior was correct (e.g. for subqueries in DELETE) in case of SELECT it has caused unnecessary locking. This patch tries to solve this problem by rethinking our approach to how we handle locking for SELECT and subqueries. Now we always set TL_READ_DEFAULT lock type for all cases when we read data. When at open_tables() time this lock is interpreted as TL_READ_NO_INSERT or TL_READ depending on whether this statement as a whole or call to function which uses particular table should be written to the binary log or not (if yes then statement should be properly serialized with concurrent statements and stronger lock should be acquired). Test coverage is added for both InnoDB and MyISAM. This patch introduces an "incompatible" change in locking scheme for subqueries used in SELECT ... FOR UPDATE and SELECT .. IN SHARE MODE. In 4.1 the server would use a snapshot InnoDB read for subqueries in SELECT FOR UPDATE and SELECT .. IN SHARE MODE statements, regardless of whether the binary log is on or off. If the user required a different type of read (i.e. locking read), he/she could request so explicitly by providing FOR UPDATE/IN SHARE MODE clause for each individual subquery. On of the patches for 5.0 broke this behaviour (which was not documented or tested), and started to use locking reads fora all subqueries in SELECT ... FOR UPDATE/IN SHARE MODE. This patch restored 4.1 behaviour.
2010-04-28 12:04:11 +02:00
thr_lock_type read_lock_type_for_table(THD *thd,
Query_tables_list *prelocking_ctx,
TABLE_LIST *table_list)
{
/*
In cases when this function is called for a sub-statement executed in
prelocked mode we can't rely on OPTION_BIN_LOG flag in THD::options
bitmap to determine that binary logging is turned on as this bit can
be cleared before executing sub-statement. So instead we have to look
at THD::variables::sql_log_bin member.
*/
bool log_on= mysql_bin_log.is_open() && thd->variables.sql_log_bin;
ulong binlog_format= thd->variables.binlog_format;
if ((log_on == FALSE) || (binlog_format == BINLOG_FORMAT_ROW) ||
Committing on behalf or Dmitry Lenev: Fix for bug #46947 "Embedded SELECT without FOR UPDATE is causing a lock", with after-review fixes. SELECT statements with subqueries referencing InnoDB tables were acquiring shared locks on rows in these tables when they were executed in REPEATABLE-READ mode and with statement or mixed mode binary logging turned on. This was a regression which were introduced when fixing bug 39843. The problem was that for tables belonging to subqueries parser set TL_READ_DEFAULT as a lock type. In cases when statement/mixed binary logging at open_tables() time this type of lock was converted to TL_READ_NO_INSERT lock at open_tables() time and caused InnoDB engine to acquire shared locks on reads from these tables. Although in some cases such behavior was correct (e.g. for subqueries in DELETE) in case of SELECT it has caused unnecessary locking. This patch tries to solve this problem by rethinking our approach to how we handle locking for SELECT and subqueries. Now we always set TL_READ_DEFAULT lock type for all cases when we read data. When at open_tables() time this lock is interpreted as TL_READ_NO_INSERT or TL_READ depending on whether this statement as a whole or call to function which uses particular table should be written to the binary log or not (if yes then statement should be properly serialized with concurrent statements and stronger lock should be acquired). Test coverage is added for both InnoDB and MyISAM. This patch introduces an "incompatible" change in locking scheme for subqueries used in SELECT ... FOR UPDATE and SELECT .. IN SHARE MODE. In 4.1 the server would use a snapshot InnoDB read for subqueries in SELECT FOR UPDATE and SELECT .. IN SHARE MODE statements, regardless of whether the binary log is on or off. If the user required a different type of read (i.e. locking read), he/she could request so explicitly by providing FOR UPDATE/IN SHARE MODE clause for each individual subquery. On of the patches for 5.0 broke this behaviour (which was not documented or tested), and started to use locking reads fora all subqueries in SELECT ... FOR UPDATE/IN SHARE MODE. This patch restored 4.1 behaviour.
2010-04-28 12:04:11 +02:00
(table_list->table->s->table_category == TABLE_CATEGORY_LOG) ||
(table_list->table->s->table_category == TABLE_CATEGORY_PERFORMANCE) ||
!(is_update_query(prelocking_ctx->sql_command) ||
table_list->prelocking_placeholder ||
(thd->locked_tables_mode > LTM_LOCK_TABLES)))
return TL_READ;
else
return TL_READ_NO_INSERT;
}
2004-06-25 15:52:01 +02:00
/*
Handle element of prelocking set other than table. E.g. cache routine
and, if prelocking strategy prescribes so, extend the prelocking set
with tables and routines used by it.
@param[in] thd Thread context.
@param[in] prelocking_ctx Prelocking context.
@param[in] rt Element of prelocking set to be processed.
@param[in] prelocking_strategy Strategy which specifies how the
prelocking set should be extended when
one of its elements is processed.
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
@param[in] has_prelocking_list Indicates that prelocking set/list for
this statement has already been built.
@param[in] ot_ctx Context of open_table used to recover from
locking failures.
@param[out] need_prelocking Set to TRUE if it was detected that this
statement will require prelocked mode for
its execution, not touched otherwise.
@retval FALSE Success.
@retval TRUE Failure (Conflicting metadata lock, OOM, other errors).
*/
static bool
open_and_process_routine(THD *thd, Query_tables_list *prelocking_ctx,
Sroutine_hash_entry *rt,
Prelocking_strategy *prelocking_strategy,
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
bool has_prelocking_list,
Open_table_context *ot_ctx,
bool *need_prelocking)
{
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
MDL_key::enum_mdl_namespace mdl_type= rt->mdl_request.key.mdl_namespace();
DBUG_ENTER("open_and_process_routine");
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
switch (mdl_type)
{
case MDL_key::FUNCTION:
case MDL_key::PROCEDURE:
{
sp_head *sp;
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
/*
Try to get MDL lock on the routine.
Note that we do not take locks on top-level CALLs as this can
lead to a deadlock. Not locking top-level CALLs does not break
the binlog as only the statements in the called procedure show
up there, not the CALL itself.
*/
if (rt != (Sroutine_hash_entry*)prelocking_ctx->sroutines_list.first ||
mdl_type != MDL_key::PROCEDURE)
{
/*
Since we acquire only shared lock on routines we don't
need to care about global intention exclusive locks.
*/
DBUG_ASSERT(rt->mdl_request.type == MDL_SHARED);
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
/*
Waiting for a conflicting metadata lock to go away may
lead to a deadlock, detected by MDL subsystem.
If possible, we try to resolve such deadlocks by releasing all
metadata locks and restarting the pre-locking process.
To prevent the error from polluting the diagnostics area
in case of successful resolution, install a special error
handler for ER_LOCK_DEADLOCK error.
*/
MDL_deadlock_handler mdl_deadlock_handler(ot_ctx);
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
thd->push_internal_handler(&mdl_deadlock_handler);
bool result= thd->mdl_context.acquire_lock(&rt->mdl_request,
ot_ctx->get_timeout());
thd->pop_internal_handler();
if (result)
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
DBUG_RETURN(TRUE);
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
DEBUG_SYNC(thd, "after_shared_lock_pname");
/* Ensures the routine is up-to-date and cached, if exists. */
if (sp_cache_routine(thd, rt, has_prelocking_list, &sp))
DBUG_RETURN(TRUE);
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
/* Remember the version of the routine in the parse tree. */
if (check_and_update_routine_version(thd, rt, sp))
DBUG_RETURN(TRUE);
/* 'sp' is NULL when there is no such routine. */
if (sp && !has_prelocking_list)
{
prelocking_strategy->handle_routine(thd, prelocking_ctx, rt, sp,
need_prelocking);
}
}
else
{
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
/*
If it's a top level call, just make sure we have a recent
version of the routine, if it exists.
Validating routine version is unnecessary, since CALL
does not affect the prepared statement prelocked list.
*/
if (sp_cache_routine(thd, rt, FALSE, &sp))
DBUG_RETURN(TRUE);
}
}
break;
case MDL_key::TRIGGER:
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
/**
We add trigger entries to lex->sroutines_list, but we don't
load them here. The trigger entry is only used when building
a transitive closure of objects used in a statement, to avoid
adding to this closure objects that are used in the trigger more
than once.
E.g. if a trigger trg refers to table t2, and the trigger table t1
is used multiple times in the statement (say, because it's used in
function f1() twice), we will only add t2 once to the list of
tables to prelock.
We don't take metadata locks on triggers either: they are protected
by a respective lock on the table, on which the trigger is defined.
The only two cases which give "trouble" are SHOW CREATE TRIGGER
and DROP TRIGGER statements. For these, statement syntax doesn't
specify the table on which this trigger is defined, so we have
to make a "dirty" read in the data dictionary to find out the
table name. Once we discover the table name, we take a metadata
lock on it, and this protects all trigger operations.
Of course the table, in theory, may disappear between the dirty
read and metadata lock acquisition, but in that case we just return
a run-time error.
Grammar of other trigger DDL statements (CREATE, DROP) requires
the table to be specified explicitly, so we use the table metadata
lock to protect trigger metadata in these statements. Similarly, in
DML we always use triggers together with their tables, and thus don't
need to take separate metadata locks on them.
*/
break;
default:
/* Impossible type value. */
DBUG_ASSERT(0);
}
DBUG_RETURN(FALSE);
}
/**
Handle table list element by obtaining metadata lock, opening table or view
and, if prelocking strategy prescribes so, extending the prelocking set with
tables and routines used by it.
@param[in] thd Thread context.
@param[in] lex LEX structure for statement.
@param[in] tables Table list element to be processed.
@param[in,out] counter Number of tables which are open.
@param[in] flags Bitmap of flags to modify how the tables
will be open, see open_table() description
for details.
@param[in] prelocking_strategy Strategy which specifies how the
prelocking set should be extended
when table or view is processed.
@param[in] has_prelocking_list Indicates that prelocking set/list for
this statement has already been built.
@param[in] ot_ctx Context used to recover from a failed
open_table() attempt.
@param[in] new_frm_mem Temporary MEM_ROOT to be used for
parsing .FRMs for views.
@retval FALSE Success.
@retval TRUE Error, reported unless there is a chance to recover from it.
*/
static bool
open_and_process_table(THD *thd, LEX *lex, TABLE_LIST *tables,
uint *counter, uint flags,
Prelocking_strategy *prelocking_strategy,
bool has_prelocking_list,
Open_table_context *ot_ctx,
MEM_ROOT *new_frm_mem)
{
bool error= FALSE;
bool safe_to_ignore_table= FALSE;
DBUG_ENTER("open_and_process_table");
DEBUG_SYNC(thd, "open_and_process_table");
/*
Ignore placeholders for derived tables. After derived tables
processing, link to created temporary table will be put here.
If this is derived table for view then we still want to process
routines used by this view.
*/
if (tables->derived)
{
if (!tables->view)
goto end;
/*
We restore view's name and database wiped out by derived tables
processing and fall back to standard open process in order to
obtain proper metadata locks and do other necessary steps like
stored routine processing.
*/
tables->db= tables->view_db.str;
tables->db_length= tables->view_db.length;
tables->table_name= tables->view_name.str;
tables->table_name_length= tables->view_name.length;
}
/*
If this TABLE_LIST object is a placeholder for an information_schema
table, create a temporary table to represent the information_schema
table in the query. Do not fill it yet - will be filled during
execution.
*/
if (tables->schema_table)
{
/*
If this information_schema table is merged into a mergeable
view, ignore it for now -- it will be filled when its respective
TABLE_LIST is processed. This code works only during re-execution.
*/
if (tables->view)
{
MDL_ticket *mdl_ticket;
/*
We still need to take a MDL lock on the merged view to protect
it from concurrent changes.
*/
if (!open_table_get_mdl_lock(thd, ot_ctx, &tables->mdl_request,
flags, &mdl_ticket) &&
mdl_ticket != NULL)
goto process_view_routines;
/* Fall-through to return error. */
}
else if (!mysql_schema_table(thd, lex, tables) &&
!check_and_update_table_version(thd, tables, tables->table->s))
{
goto end;
}
error= TRUE;
goto end;
}
DBUG_PRINT("tcache", ("opening table: '%s'.'%s' item: %p",
tables->db, tables->table_name, tables)); //psergey: invalid read of size 1 here
(*counter)++;
/* Not a placeholder: must be a base table or a view. Let us open it. */
DBUG_ASSERT(!tables->table);
if (tables->prelocking_placeholder)
{
/*
For the tables added by the pre-locking code, attempt to open
the table but fail silently if the table does not exist.
The real failure will occur when/if a statement attempts to use
that table.
*/
No_such_table_error_handler no_such_table_handler;
thd->push_internal_handler(&no_such_table_handler);
error= open_table(thd, tables, new_frm_mem, ot_ctx);
thd->pop_internal_handler();
safe_to_ignore_table= no_such_table_handler.safely_trapped_errors();
}
else
error= open_table(thd, tables, new_frm_mem, ot_ctx);
free_root(new_frm_mem, MYF(MY_KEEP_PREALLOC));
if (error)
{
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
if (! ot_ctx->can_recover_from_failed_open() && safe_to_ignore_table)
{
DBUG_PRINT("info", ("open_table: ignoring table '%s'.'%s'",
tables->db, tables->alias));
error= FALSE;
}
goto end;
}
/*
We can't rely on simple check for TABLE_LIST::view to determine
that this is a view since during re-execution we might reopen
ordinary table in place of view and thus have TABLE_LIST::view
set from repvious execution and TABLE_LIST::table set from
current.
*/
if (!tables->table && tables->view)
{
/* VIEW placeholder */
(*counter)--;
/*
tables->next_global list consists of two parts:
1) Query tables and underlying tables of views.
2) Tables used by all stored routines that this statement invokes on
execution.
We need to know where the bound between these two parts is. If we've
just opened a view, which was the last table in part #1, and it
has added its base tables after itself, adjust the boundary pointer
accordingly.
*/
if (lex->query_tables_own_last == &(tables->next_global) &&
tables->view->query_tables)
lex->query_tables_own_last= tables->view->query_tables_last;
/*
Let us free memory used by 'sroutines' hash here since we never
call destructor for this LEX.
*/
my_hash_free(&tables->view->sroutines);
goto process_view_routines;
}
/*
Special types of open can succeed but still don't set
TABLE_LIST::table to anything.
*/
if (tables->open_strategy && !tables->table)
goto end;
/*
If we are not already in prelocked mode and extended table list is not
yet built we might have to build the prelocking set for this statement.
Since currently no prelocking strategy prescribes doing anything for
tables which are only read, we do below checks only if table is going
to be changed.
*/
if (thd->locked_tables_mode <= LTM_LOCK_TABLES &&
! has_prelocking_list &&
tables->lock_type >= TL_WRITE_ALLOW_WRITE)
{
bool need_prelocking= FALSE;
TABLE_LIST **save_query_tables_last= lex->query_tables_last;
/*
Extend statement's table list and the prelocking set with
tables and routines according to the current prelocking
strategy.
For example, for DML statements we need to add tables and routines
used by triggers which are going to be invoked for this element of
table list and also add tables required for handling of foreign keys.
*/
error= prelocking_strategy->handle_table(thd, lex, tables,
&need_prelocking);
if (need_prelocking && ! lex->requires_prelocking())
lex->mark_as_requiring_prelocking(save_query_tables_last);
if (error)
goto end;
}
if (tables->lock_type != TL_UNLOCK && ! thd->locked_tables_mode)
{
if (tables->lock_type == TL_WRITE_DEFAULT)
tables->table->reginfo.lock_type= thd->update_lock_default;
else if (tables->lock_type == TL_READ_DEFAULT)
tables->table->reginfo.lock_type=
Committing on behalf or Dmitry Lenev: Fix for bug #46947 "Embedded SELECT without FOR UPDATE is causing a lock", with after-review fixes. SELECT statements with subqueries referencing InnoDB tables were acquiring shared locks on rows in these tables when they were executed in REPEATABLE-READ mode and with statement or mixed mode binary logging turned on. This was a regression which were introduced when fixing bug 39843. The problem was that for tables belonging to subqueries parser set TL_READ_DEFAULT as a lock type. In cases when statement/mixed binary logging at open_tables() time this type of lock was converted to TL_READ_NO_INSERT lock at open_tables() time and caused InnoDB engine to acquire shared locks on reads from these tables. Although in some cases such behavior was correct (e.g. for subqueries in DELETE) in case of SELECT it has caused unnecessary locking. This patch tries to solve this problem by rethinking our approach to how we handle locking for SELECT and subqueries. Now we always set TL_READ_DEFAULT lock type for all cases when we read data. When at open_tables() time this lock is interpreted as TL_READ_NO_INSERT or TL_READ depending on whether this statement as a whole or call to function which uses particular table should be written to the binary log or not (if yes then statement should be properly serialized with concurrent statements and stronger lock should be acquired). Test coverage is added for both InnoDB and MyISAM. This patch introduces an "incompatible" change in locking scheme for subqueries used in SELECT ... FOR UPDATE and SELECT .. IN SHARE MODE. In 4.1 the server would use a snapshot InnoDB read for subqueries in SELECT FOR UPDATE and SELECT .. IN SHARE MODE statements, regardless of whether the binary log is on or off. If the user required a different type of read (i.e. locking read), he/she could request so explicitly by providing FOR UPDATE/IN SHARE MODE clause for each individual subquery. On of the patches for 5.0 broke this behaviour (which was not documented or tested), and started to use locking reads fora all subqueries in SELECT ... FOR UPDATE/IN SHARE MODE. This patch restored 4.1 behaviour.
2010-04-28 12:04:11 +02:00
read_lock_type_for_table(thd, lex, tables);
else
tables->table->reginfo.lock_type= tables->lock_type;
}
tables->table->grant= tables->grant;
/* Check and update metadata version of a base table. */
error= check_and_update_table_version(thd, tables, tables->table->s);
if (error)
goto end;
/*
After opening a MERGE table add the children to the query list of
tables, so that they are opened too.
Note that placeholders don't have the handler open.
*/
/* MERGE tables need to access parent and child TABLE_LISTs. */
DBUG_ASSERT(tables->table->pos_in_table_list == tables);
/* Non-MERGE tables ignore this call. */
if (tables->table->file->extra(HA_EXTRA_ADD_CHILDREN_LIST))
{
error= TRUE;
goto end;
}
process_view_routines:
/*
Again we may need cache all routines used by this view and add
tables used by them to table list.
*/
if (tables->view &&
thd->locked_tables_mode <= LTM_LOCK_TABLES &&
! has_prelocking_list)
{
bool need_prelocking= FALSE;
TABLE_LIST **save_query_tables_last= lex->query_tables_last;
error= prelocking_strategy->handle_view(thd, lex, tables,
&need_prelocking);
if (need_prelocking && ! lex->requires_prelocking())
lex->mark_as_requiring_prelocking(save_query_tables_last);
if (error)
goto end;
}
end:
DBUG_RETURN(error);
}
extern "C" uchar *schema_set_get_key(const uchar *record, size_t *length,
my_bool not_used __attribute__((unused)))
{
TABLE_LIST *table=(TABLE_LIST*) record;
*length= table->db_length;
return (uchar*) table->db;
}
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
/**
Acquire upgradable (SNW, SNRW) metadata locks on tables used by
LOCK TABLES or by a DDL statement. Under LOCK TABLES, we can't take
new locks, so use open_tables_check_upgradable_mdl() instead.
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
@param thd Thread context.
@param tables_start Start of list of tables on which upgradable locks
should be acquired.
@param tables_end End of list of tables.
@param lock_wait_timeout Seconds to wait before timeout.
@param flags Bitmap of flags to modify how the tables will be
open, see open_table() description for details.
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
@retval FALSE Success.
@retval TRUE Failure (e.g. connection was killed)
*/
bool
lock_table_names(THD *thd,
TABLE_LIST *tables_start, TABLE_LIST *tables_end,
ulong lock_wait_timeout, uint flags)
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
{
MDL_request_list mdl_requests;
TABLE_LIST *table;
MDL_request global_request;
Hash_set<TABLE_LIST, schema_set_get_key> schema_set;
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
DBUG_ASSERT(!thd->locked_tables_mode);
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
for (table= tables_start; table && table != tables_end;
table= table->next_global)
{
if (table->mdl_request.type >= MDL_SHARED_NO_WRITE &&
!(table->open_type == OT_TEMPORARY_ONLY ||
(flags & MYSQL_OPEN_TEMPORARY_ONLY) ||
(table->open_type != OT_BASE_ONLY &&
! (flags & MYSQL_OPEN_SKIP_TEMPORARY) &&
find_temporary_table(thd, table))))
{
if (! (flags & MYSQL_OPEN_SKIP_SCOPED_MDL_LOCK) &&
schema_set.insert(table))
return TRUE;
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
mdl_requests.push_front(&table->mdl_request);
}
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
}
if (! (flags & MYSQL_OPEN_SKIP_SCOPED_MDL_LOCK) &&
! mdl_requests.is_empty())
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
{
/*
Scoped locks: Take intention exclusive locks on all involved
schemas.
*/
Hash_set<TABLE_LIST, schema_set_get_key>::Iterator it(schema_set);
while ((table= it++))
{
MDL_request *schema_request= new (thd->mem_root) MDL_request;
if (schema_request == NULL)
return TRUE;
schema_request->init(MDL_key::SCHEMA, table->db, "",
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
MDL_INTENTION_EXCLUSIVE,
MDL_TRANSACTION);
mdl_requests.push_front(schema_request);
}
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
/*
Protect this statement against concurrent global read lock
by acquiring global intention exclusive lock with statement
duration.
*/
if (thd->global_read_lock.can_acquire_protection())
return TRUE;
global_request.init(MDL_key::GLOBAL, "", "", MDL_INTENTION_EXCLUSIVE,
MDL_STATEMENT);
mdl_requests.push_front(&global_request);
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
}
if (thd->mdl_context.acquire_locks(&mdl_requests, lock_wait_timeout))
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
return TRUE;
return FALSE;
}
/**
Check for upgradable (SNW, SNRW) metadata locks on tables to be opened
for a DDL statement. Under LOCK TABLES, we can't take new locks, so we
must check if appropriate locks were pre-acquired.
@param thd Thread context.
@param tables_start Start of list of tables on which upgradable locks
should be searched for.
@param tables_end End of list of tables.
@param flags Bitmap of flags to modify how the tables will be
open, see open_table() description for details.
@retval FALSE Success.
@retval TRUE Failure (e.g. connection was killed)
*/
static bool
open_tables_check_upgradable_mdl(THD *thd, TABLE_LIST *tables_start,
TABLE_LIST *tables_end, uint flags)
{
TABLE_LIST *table;
DBUG_ASSERT(thd->locked_tables_mode);
for (table= tables_start; table && table != tables_end;
table= table->next_global)
{
if (table->mdl_request.type >= MDL_SHARED_NO_WRITE &&
!(table->open_type == OT_TEMPORARY_ONLY ||
(flags & MYSQL_OPEN_TEMPORARY_ONLY) ||
(table->open_type != OT_BASE_ONLY &&
! (flags & MYSQL_OPEN_SKIP_TEMPORARY) &&
find_temporary_table(thd, table))))
{
/*
We don't need to do anything about the found TABLE instance as it
will be handled later in open_tables(), we only need to check that
an upgradable lock is already acquired. When we enter LOCK TABLES
mode, SNRW locks are acquired before all other locks. So if under
LOCK TABLES we find that there is TABLE instance with upgradeable
lock, all other instances of TABLE for the same table will have the
same ticket.
Note that this works OK even for CREATE TABLE statements which
request X type of metadata lock. This is because under LOCK TABLES
such statements don't create the table but only check if it exists
or, in most complex case, only insert into it.
Thus SNRW lock should be enough.
Note that find_table_for_mdl_upgrade() will report an error if
no suitable ticket is found.
*/
if (!find_table_for_mdl_upgrade(thd, table->db, table->table_name, false))
return TRUE;
}
}
return FALSE;
}
/**
2004-06-25 15:52:01 +02:00
Open all tables in list
@param[in] thd Thread context.
@param[in,out] start List of tables to be open (it can be adjusted for
statement that uses tables only implicitly, e.g.
for "SELECT f1()").
@param[out] counter Number of tables which were open.
@param[in] flags Bitmap of flags to modify how the tables will be
open, see open_table() description for details.
@param[in] prelocking_strategy Strategy which specifies how prelocking
algorithm should work for this statement.
2000-07-31 21:29:14 +02:00
@note
Unless we are already in prelocked mode and prelocking strategy prescribes
so this function will also precache all SP/SFs explicitly or implicitly
(via views and triggers) used by the query and add tables needed for their
execution to table list. Statement that uses SFs, invokes triggers or
requires foreign key checks will be marked as requiring prelocking.
Prelocked mode will be enabled for such query during lock_tables() call.
If query for which we are opening tables is already marked as requiring
prelocking it won't do such precaching and will simply reuse table list
which is already built.
@retval FALSE Success.
@retval TRUE Error, reported.
2004-06-25 15:52:01 +02:00
*/
bool open_tables(THD *thd, TABLE_LIST **start, uint *counter, uint flags,
Prelocking_strategy *prelocking_strategy)
2000-07-31 21:29:14 +02:00
{
/*
We use pointers to "next_global" member in the last processed TABLE_LIST
element and to the "next" member in the last processed Sroutine_hash_entry
element as iterators over, correspondingly, the table list and stored routines
list which stay valid and allow to continue iteration when new elements are
added to the tail of the lists.
*/
TABLE_LIST **table_to_open;
Sroutine_hash_entry **sroutine_to_open;
TABLE_LIST *tables;
Open_table_context ot_ctx(thd, flags);
bool error= FALSE;
2004-07-16 00:15:55 +02:00
MEM_ROOT new_frm_mem;
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
bool has_prelocking_list;
DBUG_ENTER("open_tables");
/* Accessing data in XA_IDLE or XA_PREPARED is not allowed. */
enum xa_states xa_state= thd->transaction.xid_state.xa_state;
if (*start && (xa_state == XA_IDLE || xa_state == XA_PREPARED))
{
my_error(ER_XAER_RMFAIL, MYF(0), xa_state_names[xa_state]);
DBUG_RETURN(true);
}
/*
temporary mem_root for new .frm parsing.
TODO: variables for size
*/
init_sql_alloc(&new_frm_mem, 8024, 8024);
thd->current_tablenr= 0;
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
restart:
Patch that changes approach to how we acquire metadata locks for DML statements and changes the way MDL locks are acquired/granted in contended case. Instead of backing-off when a lock conflict is encountered and waiting for it to go away before restarting open_tables() process we now wait for lock to be released without releasing any previously acquired locks. If conflicting lock goes away we resume opening tables. If waiting leads to a deadlock we try to resolve it by backing-off and restarting open_tables() immediately. As result both waiting for possibility to acquire and acquiring of a metadata lock now always happen within the same MDL API call. This has allowed to make release of a lock and granting it to the most appropriate pending request an atomic operation. Thanks to this it became possible to wake up during release of lock only those waiters which requests can be satisfied at the moment as well as wake up only one waiter in case when granting its request would prevent all other requests from being satisfied. This solves thundering herd problem which occured in cases when we were releasing some lock and woke up many waiters for SNRW or X locks (this was the issue in bug#52289 "performance regression for MyISAM in sysbench OLTP_RW test". This also allowed to implement more fair (FIFO) scheduling among waiters with the same priority. It also opens the door for introducing new types of requests for metadata locks such as low-prio SNRW lock which is necessary in order to support LOCK TABLES LOW_PRIORITY WRITE. Notice that after this sometimes can report ER_LOCK_DEADLOCK error in cases in which it has not happened before. Particularly we will always report this error if waiting for conflicting lock has happened in the middle of transaction and resulted in a deadlock. Before this patch the error was not reported if deadlock could have been resolved by backing off all metadata locks acquired by the current statement.
2010-06-07 09:06:55 +02:00
/*
Close HANDLER tables which are marked for flush or against which there
are pending exclusive metadata locks. This is needed both in order to
avoid deadlocks and to have a point during statement execution at
which such HANDLERs are closed even if they don't create problems for
the current session (i.e. to avoid having a DDL blocked by HANDLERs
opened for a long time).
*/
if (thd->handler_tables_hash.records)
mysql_ha_flush(thd);
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
has_prelocking_list= thd->lex->requires_prelocking();
table_to_open= start;
sroutine_to_open= (Sroutine_hash_entry**) &thd->lex->sroutines_list.first;
*counter= 0;
thd_proc_info(thd, "Opening tables");
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
/*
If we are executing LOCK TABLES statement or a DDL statement
(in non-LOCK TABLES mode) we might have to acquire upgradable
semi-exclusive metadata locks (SNW or SNRW) on some of the
tables to be opened.
When executing CREATE TABLE .. If NOT EXISTS .. SELECT, the
table may not yet exist, in which case we acquire an exclusive
lock.
We acquire all such locks at once here as doing this in one
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
by one fashion may lead to deadlocks or starvation. Later when
we will be opening corresponding table pre-acquired metadata
lock will be reused (thanks to the fact that in recursive case
metadata locks are acquired without waiting).
*/
if (! (flags & (MYSQL_OPEN_HAS_MDL_LOCK |
MYSQL_OPEN_FORCE_SHARED_MDL |
MYSQL_OPEN_FORCE_SHARED_HIGH_PRIO_MDL)))
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
{
if (thd->locked_tables_mode)
{
/*
Under LOCK TABLES, we can't acquire new locks, so we instead
need to check if appropriate locks were pre-acquired.
*/
if (open_tables_check_upgradable_mdl(thd, *start,
thd->lex->first_not_own_table(),
flags))
{
error= TRUE;
goto err;
}
}
else
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
{
TABLE_LIST *table;
if (lock_table_names(thd, *start, thd->lex->first_not_own_table(),
ot_ctx.get_timeout(), flags))
{
error= TRUE;
goto err;
}
for (table= *start; table && table != thd->lex->first_not_own_table();
table= table->next_global)
{
if (table->mdl_request.type >= MDL_SHARED_NO_WRITE)
table->mdl_request.ticket= NULL;
}
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
}
}
/*
Perform steps of prelocking algorithm until there are unprocessed
elements in prelocking list/set.
*/
while (*table_to_open ||
(thd->locked_tables_mode <= LTM_LOCK_TABLES &&
*sroutine_to_open))
2000-07-31 21:29:14 +02:00
{
/*
For every table in the list of tables to open, try to find or open
a table.
*/
for (tables= *table_to_open; tables;
table_to_open= &tables->next_global, tables= tables->next_global)
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
{
error= open_and_process_table(thd, thd->lex, tables, counter,
flags, prelocking_strategy,
has_prelocking_list, &ot_ctx,
&new_frm_mem);
2004-07-16 00:15:55 +02:00
if (error)
2000-07-31 21:29:14 +02:00
{
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
if (ot_ctx.can_recover_from_failed_open())
{
/*
We have met exclusive metadata lock or old version of table.
Now we have to close all tables and release metadata locks.
We also have to throw away set of prelocked tables (and thus
close tables from this set that were open by now) since it
is possible that one of tables which determined its content
was changed.
Instead of implementing complex/non-robust logic mentioned
above we simply close and then reopen all tables.
We have to save pointer to table list element for table which we
have failed to open since closing tables can trigger removal of
elements from the table list (if MERGE tables are involved),
*/
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
close_tables_for_reopen(thd, start, ot_ctx.start_of_statement_svp());
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
/*
Here we rely on the fact that 'tables' still points to the valid
TABLE_LIST element. Altough currently this assumption is valid
it may change in future.
*/
if (ot_ctx.recover_from_failed_open(thd))
goto err;
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
error= FALSE;
goto restart;
}
goto err;
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
}
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
DEBUG_SYNC(thd, "open_tables_after_open_and_process_table");
2000-07-31 21:29:14 +02:00
}
/*
If we are not already in prelocked mode and extended table list is
not yet built for our statement we need to cache routines it uses
and build the prelocking list for it.
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
If we are not in prelocked mode but have built the extended table
list, we still need to call open_and_process_routine() to take
MDL locks on the routines.
*/
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
if (thd->locked_tables_mode <= LTM_LOCK_TABLES)
{
bool need_prelocking= FALSE;
TABLE_LIST **save_query_tables_last= thd->lex->query_tables_last;
/*
Process elements of the prelocking set which are present there
since parsing stage or were added to it by invocations of
Prelocking_strategy methods in the above loop over tables.
For example, if element is a routine, cache it and then,
if prelocking strategy prescribes so, add tables it uses to the
table list and routines it might invoke to the prelocking set.
*/
for (Sroutine_hash_entry *rt= *sroutine_to_open; rt;
sroutine_to_open= &rt->next, rt= rt->next)
{
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
error= open_and_process_routine(thd, thd->lex, rt, prelocking_strategy,
has_prelocking_list, &ot_ctx,
&need_prelocking);
if (error)
{
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
if (ot_ctx.can_recover_from_failed_open())
{
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
close_tables_for_reopen(thd, start,
ot_ctx.start_of_statement_svp());
if (ot_ctx.recover_from_failed_open(thd))
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
goto err;
error= FALSE;
goto restart;
}
/*
Serious error during reading stored routines from mysql.proc table.
Something is wrong with the table or its contents, and an error has
been emitted; we must abort.
*/
goto err;
}
}
if (need_prelocking && ! thd->lex->requires_prelocking())
thd->lex->mark_as_requiring_prelocking(save_query_tables_last);
if (need_prelocking && ! *start)
*start= thd->lex->query_tables;
}
2000-07-31 21:29:14 +02:00
}
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
/*
After successful open of all tables, including MERGE parents and
children, attach the children to their parents. At end of statement,
the children are detached. Attaching and detaching are always done,
even under LOCK TABLES.
*/
for (tables= *start; tables; tables= tables->next_global)
{
TABLE *tbl= tables->table;
/* Schema tables may not have a TABLE object here. */
if (tbl && tbl->file->ht->db_type == DB_TYPE_MRG_MYISAM)
{
/* MERGE tables need to access parent and child TABLE_LISTs. */
DBUG_ASSERT(tbl->pos_in_table_list == tables);
if (tbl->file->extra(HA_EXTRA_ATTACH_CHILDREN))
{
error= TRUE;
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
goto err;
}
}
}
err:
Prevent bugs by making DBUG_* expressions syntactically equivalent to a single statement. --- Bug#24795: SHOW PROFILE Profiling is only partially functional on some architectures. Where there is no getrusage() system call, presently Null values are returned where it would be required. Notably, Windows needs some love applied to make it as useful. Syntax this adds: SHOW PROFILES SHOW PROFILE [types] [FOR QUERY n] [OFFSET n] [LIMIT n] where "n" is an integer and "types" is zero or many (comma-separated) of "CPU" "MEMORY" (not presently supported) "BLOCK IO" "CONTEXT SWITCHES" "PAGE FAULTS" "IPC" "SWAPS" "SOURCE" "ALL" It also adds a session variable (boolean) "profiling", set to "no" by default, and (integer) profiling_history_size, set to 15 by default. This patch abstracts setting THDs' "proc_info" behind a macro that can be used as a hook into the profiling code when profiling support is compiled in. All future code in this line should use that mechanism for setting thd->proc_info. --- Tests are now set to omit the statistics. --- Adds an Information_schema table, "profiling" for access to "show profile" data. --- Merge zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-community-3--bug24795 into zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-community --- Fix merge problems. --- Fixed one bug in the query_source being NULL. Updated test results. --- Include more thorough profiling tests. Improve support for prepared statements. Use session-specific query IDs, starting at zero. --- Selecting from I_S.profiling is no longer quashed in profiling, as requested by Giuseppe. Limit the size of captured query text. No longer log queries that are zero length.
2007-02-22 16:03:08 +01:00
thd_proc_info(thd, 0);
free_root(&new_frm_mem, MYF(0)); // Free pre-alloced block
if (error && *table_to_open)
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
{
(*table_to_open)->table= NULL;
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
}
DBUG_PRINT("open_tables", ("returning: %d", (int) error));
DBUG_RETURN(error);
2000-07-31 21:29:14 +02:00
}
/**
Defines how prelocking algorithm for DML statements should handle routines:
- For CALL statements we do unrolling (i.e. open and lock tables for each
sub-statement individually). So for such statements prelocking is enabled
only if stored functions are used in parameter list and only for period
during which we calculate values of parameters. Thus in this strategy we
ignore procedure which is directly called by such statement and extend
the prelocking set only with tables/functions used by SF called from the
parameter list.
- For any other statement any routine which is directly or indirectly called
by statement is going to be executed in prelocked mode. So in this case we
simply add all tables and routines used by it to the prelocking set.
@param[in] thd Thread context.
@param[in] prelocking_ctx Prelocking context of the statement.
@param[in] rt Prelocking set element describing routine.
@param[in] sp Routine body.
@param[out] need_prelocking Set to TRUE if method detects that prelocking
required, not changed otherwise.
@retval FALSE Success.
@retval TRUE Failure (OOM).
*/
bool DML_prelocking_strategy::
handle_routine(THD *thd, Query_tables_list *prelocking_ctx,
Sroutine_hash_entry *rt, sp_head *sp, bool *need_prelocking)
{
/*
We assume that for any "CALL proc(...)" statement sroutines_list will
have 'proc' as first element (it may have several, consider e.g.
"proc(sp_func(...)))". This property is currently guaranted by the
parser.
*/
if (rt != (Sroutine_hash_entry*)prelocking_ctx->sroutines_list.first ||
rt->mdl_request.key.mdl_namespace() != MDL_key::PROCEDURE)
{
*need_prelocking= TRUE;
sp_update_stmt_used_routines(thd, prelocking_ctx, &sp->m_sroutines,
rt->belong_to_view);
(void)sp->add_used_tables_to_table_list(thd,
&prelocking_ctx->query_tables_last,
rt->belong_to_view);
}
sp->propagate_attributes(prelocking_ctx);
return FALSE;
}
/**
Defines how prelocking algorithm for DML statements should handle table list
elements:
- If table has triggers we should add all tables and routines
used by them to the prelocking set.
We do not need to acquire metadata locks on trigger names
in DML statements, since all DDL statements
that change trigger metadata always lock their
subject tables.
@param[in] thd Thread context.
@param[in] prelocking_ctx Prelocking context of the statement.
@param[in] table_list Table list element for table.
@param[in] sp Routine body.
@param[out] need_prelocking Set to TRUE if method detects that prelocking
required, not changed otherwise.
@retval FALSE Success.
@retval TRUE Failure (OOM).
*/
bool DML_prelocking_strategy::
handle_table(THD *thd, Query_tables_list *prelocking_ctx,
TABLE_LIST *table_list, bool *need_prelocking)
{
/* We rely on a caller to check that table is going to be changed. */
DBUG_ASSERT(table_list->lock_type >= TL_WRITE_ALLOW_WRITE);
if (table_list->trg_event_map)
{
if (table_list->table->triggers)
{
*need_prelocking= TRUE;
if (table_list->table->triggers->
add_tables_and_routines_for_triggers(thd, prelocking_ctx, table_list))
return TRUE;
}
}
return FALSE;
}
/**
Defines how prelocking algorithm for DML statements should handle view -
all view routines should be added to the prelocking set.
@param[in] thd Thread context.
@param[in] prelocking_ctx Prelocking context of the statement.
@param[in] table_list Table list element for view.
@param[in] sp Routine body.
@param[out] need_prelocking Set to TRUE if method detects that prelocking
required, not changed otherwise.
@retval FALSE Success.
@retval TRUE Failure (OOM).
*/
bool DML_prelocking_strategy::
handle_view(THD *thd, Query_tables_list *prelocking_ctx,
TABLE_LIST *table_list, bool *need_prelocking)
{
if (table_list->view->uses_stored_routines())
{
*need_prelocking= TRUE;
sp_update_stmt_used_routines(thd, prelocking_ctx,
&table_list->view->sroutines_list,
table_list->top_table());
}
return FALSE;
}
/**
Defines how prelocking algorithm for LOCK TABLES statement should handle
table list elements.
@param[in] thd Thread context.
@param[in] prelocking_ctx Prelocking context of the statement.
@param[in] table_list Table list element for table.
@param[in] sp Routine body.
@param[out] need_prelocking Set to TRUE if method detects that prelocking
required, not changed otherwise.
@retval FALSE Success.
@retval TRUE Failure (OOM).
*/
bool Lock_tables_prelocking_strategy::
handle_table(THD *thd, Query_tables_list *prelocking_ctx,
TABLE_LIST *table_list, bool *need_prelocking)
{
if (DML_prelocking_strategy::handle_table(thd, prelocking_ctx, table_list,
need_prelocking))
return TRUE;
/* We rely on a caller to check that table is going to be changed. */
DBUG_ASSERT(table_list->lock_type >= TL_WRITE_ALLOW_WRITE);
return FALSE;
}
/**
Defines how prelocking algorithm for ALTER TABLE statement should handle
routines - do nothing as this statement is not supposed to call routines.
We still can end up in this method when someone tries
to define a foreign key referencing a view, and not just
a simple view, but one that uses stored routines.
*/
bool Alter_table_prelocking_strategy::
handle_routine(THD *thd, Query_tables_list *prelocking_ctx,
Sroutine_hash_entry *rt, sp_head *sp, bool *need_prelocking)
{
return FALSE;
}
/**
Defines how prelocking algorithm for ALTER TABLE statement should handle
table list elements.
Unlike in DML, we do not process triggers here.
@param[in] thd Thread context.
@param[in] prelocking_ctx Prelocking context of the statement.
@param[in] table_list Table list element for table.
@param[in] sp Routine body.
@param[out] need_prelocking Set to TRUE if method detects that prelocking
required, not changed otherwise.
@retval FALSE Success.
@retval TRUE Failure (OOM).
*/
bool Alter_table_prelocking_strategy::
handle_table(THD *thd, Query_tables_list *prelocking_ctx,
TABLE_LIST *table_list, bool *need_prelocking)
{
return FALSE;
}
/**
Defines how prelocking algorithm for ALTER TABLE statement
should handle view - do nothing. We don't need to add view
routines to the prelocking set in this case as view is not going
to be materialized.
*/
bool Alter_table_prelocking_strategy::
handle_view(THD *thd, Query_tables_list *prelocking_ctx,
TABLE_LIST *table_list, bool *need_prelocking)
{
return FALSE;
}
/**
Check that lock is ok for tables; Call start stmt if ok
@param thd Thread handle.
@param prelocking_ctx Prelocking context.
@param table_list Table list element for table to be checked.
@retval FALSE - Ok.
@retval TRUE - Error.
*/
static bool check_lock_and_start_stmt(THD *thd,
Query_tables_list *prelocking_ctx,
TABLE_LIST *table_list)
{
int error;
thr_lock_type lock_type;
DBUG_ENTER("check_lock_and_start_stmt");
/*
TL_WRITE_DEFAULT and TL_READ_DEFAULT are supposed to be parser only
types of locks so they should be converted to appropriate other types
to be passed to storage engine. The exact lock type passed to the
engine is important as, for example, InnoDB uses it to determine
what kind of row locks should be acquired when executing statement
in prelocked mode or under LOCK TABLES with @@innodb_table_locks = 0.
*/
if (table_list->lock_type == TL_WRITE_DEFAULT)
lock_type= thd->update_lock_default;
else if (table_list->lock_type == TL_READ_DEFAULT)
lock_type= read_lock_type_for_table(thd, prelocking_ctx, table_list);
else
lock_type= table_list->lock_type;
Fix for bug #51263 "Deadlock between transactional SELECT and ALTER TABLE ... REBUILD PARTITION". ALTER TABLE on InnoDB table (including partitioned tables) acquired exclusive locks on rows of table being altered. In cases when there was concurrent transaction which did locking reads from this table this sometimes led to a deadlock which was not detected by MDL subsystem nor by InnoDB engine (and was reported only after exceeding innodb_lock_wait_timeout). This problem stemmed from the fact that ALTER TABLE acquired TL_WRITE_ALLOW_READ lock on table being altered. This lock was interpreted as a write lock and thus for table being altered handler::external_lock() method was called with F_WRLCK as an argument. As result InnoDB engine treated ALTER TABLE as an operation which is going to change data and acquired LOCK_X locks on rows being read from old version of table. In case when there was a transaction which already acquired SR metadata lock on table and some LOCK_S locks on its rows (e.g. by using it in subquery of DML statement) concurrent ALTER TABLE was blocked at the moment when it tried to acquire LOCK_X lock before reading one of these rows. The transaction's attempt to acquire SW metadata lock on table being altered led to deadlock, since it had to wait for ALTER TABLE to release SNW lock. This deadlock was not detected and got resolved only after timeout expiring because waiting were happening in two different subsystems. Similar deadlocks could have occured in other situations. This patch tries to solve the problem by changing ALTER TABLE implementation to use TL_READ_NO_INSERT lock instead of TL_WRITE_ALLOW_READ. After this step handler::external_lock() is called with F_RDLCK as an argument and InnoDB engine correctly interprets ALTER TABLE as operation which only reads data from original version of table. Thanks to this ALTER TABLE acquires only LOCK_S locks on rows it reads. This, in its turn, causes inter-subsystem deadlocks to go away, as all potential lock conflicts and thus deadlocks will be limited to metadata locking subsystem: - When ALTER TABLE reads rows from table being altered it can't encounter any locks which conflict with LOCK_S row locks. There should be no concurrent transactions holding LOCK_X row locks. Such a transaction should have been acquired SW metadata lock on table first which would have conflicted with ALTER's SNW lock. - Vice versa, when DML which runs concurrently with ALTER TABLE tries to lock row it should be requesting only LOCK_S lock which is compatible with locks acquired by ALTER, as otherwise such DML must own an SW metadata lock on table which would be incompatible with ALTER's SNW lock.
2010-05-26 14:18:08 +02:00
if ((int) lock_type > (int) TL_WRITE_ALLOW_WRITE &&
(int) table_list->table->reginfo.lock_type <= (int) TL_WRITE_ALLOW_WRITE)
{
my_error(ER_TABLE_NOT_LOCKED_FOR_WRITE, MYF(0), table_list->alias);
DBUG_RETURN(1);
}
if ((error= table_list->table->file->start_stmt(thd, lock_type)))
{
table_list->table->file->print_error(error, MYF(0));
DBUG_RETURN(1);
}
DBUG_RETURN(0);
}
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
/**
@brief Open and lock one table
@param[in] thd thread handle
@param[in] table_l table to open is first table in this list
@param[in] lock_type lock to use for table
@param[in] flags options to be used while opening and locking
table (see open_table(), mysql_lock_tables())
@param[in] prelocking_strategy Strategy which specifies how prelocking
algorithm should work for this statement.
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
@return table
@retval != NULL OK, opened table returned
@retval NULL Error
@note
If ok, the following are also set:
table_list->lock_type lock_type
table_list->table table
@note
If table_l is a list, not a single table, the list is temporarily
broken.
@detail
This function is meant as a replacement for open_ltable() when
MERGE tables can be opened. open_ltable() cannot open MERGE tables.
There may be more differences between open_n_lock_single_table() and
open_ltable(). One known difference is that open_ltable() does
BUG#39934: Slave stops for engine that only support row-based logging General overview: The logic for switching to row format when binlog_format=MIXED had numerous flaws. The underlying problem was the lack of a consistent architecture. General purpose of this changeset: This changeset introduces an architecture for switching to row format when binlog_format=MIXED. It enforces the architecture where it has to. It leaves some bugs to be fixed later. It adds extensive tests to verify that unsafe statements work as expected and that appropriate errors are produced by problems with the selection of binlog format. It was not practical to split this into smaller pieces of work. Problem 1: To determine the logging mode, the code has to take several parameters into account (namely: (1) the value of binlog_format; (2) the capabilities of the engines; (3) the type of the current statement: normal, unsafe, or row injection). These parameters may conflict in several ways, namely: - binlog_format=STATEMENT for a row injection - binlog_format=STATEMENT for an unsafe statement - binlog_format=STATEMENT for an engine only supporting row logging - binlog_format=ROW for an engine only supporting statement logging - statement is unsafe and engine does not support row logging - row injection in a table that does not support statement logging - statement modifies one table that does not support row logging and one that does not support statement logging Several of these conflicts were not detected, or were detected with an inappropriate error message. The problem of BUG#39934 was that no appropriate error message was written for the case when an engine only supporting row logging executed a row injection with binlog_format=ROW. However, all above cases must be handled. Fix 1: Introduce new error codes (sql/share/errmsg.txt). Ensure that all conditions are detected and handled in decide_logging_format() Problem 2: The binlog format shall be determined once per statement, in decide_logging_format(). It shall not be changed before or after that. Before decide_logging_format() is called, all information necessary to determine the logging format must be available. This principle ensures that all unsafe statements are handled in a consistent way. However, this principle is not followed: thd->set_current_stmt_binlog_row_based_if_mixed() is called in several places, including from code executing UPDATE..LIMIT, INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and SET @@binlog_format. After Problem 1 was fixed, that caused inconsistencies where these unsafe statements would not print the appropriate warnings or errors for some of the conflicts. Fix 2: Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from code executed after decide_logging_format(). Compensate by calling the set_current_stmt_unsafe() at parse time. This way, all unsafe statements are detected by decide_logging_format(). Problem 3: INSERT DELAYED is not unsafe: it is logged in statement format even if binlog_format=MIXED, and no warning is printed even if binlog_format=STATEMENT. This is BUG#45825. Fix 3: Made INSERT DELAYED set itself to unsafe at parse time. This allows decide_logging_format() to detect that a warning should be printed or the binlog_format changed. Problem 4: LIMIT clause were not marked as unsafe when executed inside stored functions/triggers/views/prepared statements. This is BUG#45785. Fix 4: Make statements containing the LIMIT clause marked as unsafe at parse time, instead of at execution time. This allows propagating unsafe-ness to the view.
2009-07-14 21:31:19 +02:00
neither call thd->decide_logging_format() nor handle some other logging
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
and locking issues because it does not call lock_tables().
*/
TABLE *open_n_lock_single_table(THD *thd, TABLE_LIST *table_l,
thr_lock_type lock_type, uint flags,
Prelocking_strategy *prelocking_strategy)
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
{
TABLE_LIST *save_next_global;
DBUG_ENTER("open_n_lock_single_table");
/* Remember old 'next' pointer. */
save_next_global= table_l->next_global;
/* Break list. */
table_l->next_global= NULL;
/* Set requested lock type. */
table_l->lock_type= lock_type;
/* Allow to open real tables only. */
table_l->required_type= FRMTYPE_TABLE;
/* Open the table. */
if (open_and_lock_tables(thd, table_l, FALSE, flags,
prelocking_strategy))
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
table_l->table= NULL; /* Just to be sure. */
/* Restore list. */
table_l->next_global= save_next_global;
DBUG_RETURN(table_l->table);
}
/*
Open and lock one table
SYNOPSIS
open_ltable()
thd Thread handler
table_list Table to open is first table in this list
lock_type Lock to use for open
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
lock_flags Flags passed to mysql_lock_table
NOTE
This function doesn't do anything like SP/SF/views/triggers analysis done
2009-03-27 13:55:14 +01:00
in open_table()/lock_tables(). It is intended for opening of only one
concrete table. And used only in special contexts.
RETURN VALUES
table Opened table
0 Error
If ok, the following are also set:
table_list->lock_type lock_type
table_list->table table
*/
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
TABLE *open_ltable(THD *thd, TABLE_LIST *table_list, thr_lock_type lock_type,
uint lock_flags)
2000-07-31 21:29:14 +02:00
{
TABLE *table;
Open_table_context ot_ctx(thd, lock_flags);
bool error;
2000-07-31 21:29:14 +02:00
DBUG_ENTER("open_ltable");
2009-03-27 13:55:14 +01:00
/* should not be used in a prelocked_mode context, see NOTE above */
DBUG_ASSERT(thd->locked_tables_mode < LTM_PRELOCKED);
2009-03-27 13:55:14 +01:00
Prevent bugs by making DBUG_* expressions syntactically equivalent to a single statement. --- Bug#24795: SHOW PROFILE Profiling is only partially functional on some architectures. Where there is no getrusage() system call, presently Null values are returned where it would be required. Notably, Windows needs some love applied to make it as useful. Syntax this adds: SHOW PROFILES SHOW PROFILE [types] [FOR QUERY n] [OFFSET n] [LIMIT n] where "n" is an integer and "types" is zero or many (comma-separated) of "CPU" "MEMORY" (not presently supported) "BLOCK IO" "CONTEXT SWITCHES" "PAGE FAULTS" "IPC" "SWAPS" "SOURCE" "ALL" It also adds a session variable (boolean) "profiling", set to "no" by default, and (integer) profiling_history_size, set to 15 by default. This patch abstracts setting THDs' "proc_info" behind a macro that can be used as a hook into the profiling code when profiling support is compiled in. All future code in this line should use that mechanism for setting thd->proc_info. --- Tests are now set to omit the statistics. --- Adds an Information_schema table, "profiling" for access to "show profile" data. --- Merge zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-community-3--bug24795 into zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-community --- Fix merge problems. --- Fixed one bug in the query_source being NULL. Updated test results. --- Include more thorough profiling tests. Improve support for prepared statements. Use session-specific query IDs, starting at zero. --- Selecting from I_S.profiling is no longer quashed in profiling, as requested by Giuseppe. Limit the size of captured query text. No longer log queries that are zero length.
2007-02-22 16:03:08 +01:00
thd_proc_info(thd, "Opening table");
thd->current_tablenr= 0;
/* open_ltable can be used only for BASIC TABLEs */
table_list->required_type= FRMTYPE_TABLE;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/* This function can't properly handle requests for such metadata locks. */
DBUG_ASSERT(table_list->mdl_request.type < MDL_SHARED_NO_WRITE);
while ((error= open_table(thd, table_list, thd->mem_root, &ot_ctx)) &&
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
ot_ctx.can_recover_from_failed_open())
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
/*
Even though we have failed to open table we still need to
A prerequisite patch for the fix for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". Introduce a notion of a sentinel to MDL_context. A sentinel is a ticket that separates all tickets in the context into two groups: before and after it. Currently we can have (and need) only one designated sentinel -- it separates all locks taken by LOCK TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK and all other locks, which must be released at COMMIT or ROLLBACK. The tricky part is maintaining the sentinel up to date when someone release its corresponding ticket. This can happen, e.g. if someone issues DROP TABLE under LOCK TABLES (generally, see all calls to release_all_locks_for_name()). MDL_context::release_ticket() is modified to take care of it. ****** A fix and a test case for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". An attempt to mix HANDLER SQL statements, which are transaction- agnostic, an open multi-statement transaction, and DDL against the involved tables (in a concurrent connection) could lead to a deadlock. The deadlock would occur when HANDLER OPEN or HANDLER READ would have to wait on a conflicting metadata lock. If the connection that issued HANDLER statement also had other metadata locks (say, acquired in scope of a transaction), a classical deadlock situation of mutual wait could occur. Incompatible change: entering LOCK TABLES mode automatically closes all open HANDLERs in the current connection. Incompatible change: previously an attempt to wait on a lock in a connection that has an open HANDLER statement could wait indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK is produced. The idea of the fix is to merge thd->handler_mdl_context with the main mdl_context of the connection, used for transactional locks. This makes deadlock detection possible, since all waits with locks are "visible" and available to analysis in a single MDL context of the connection. Since HANDLER locks and transactional locks have a different life cycle -- HANDLERs are explicitly open and closed, and so are HANDLER locks, explicitly acquired and released, whereas transactional locks "accumulate" till the end of a transaction and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT, a concept of "sentinel" was introduced to MDL_context. All locks, HANDLER and others, reside in the same linked list. However, a selected element of the list separates locks with different life cycle. HANDLER locks always reside at the end of the list, after the sentinel. Transactional locks are prepended to the beginning of the list, before the sentinel. Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only release those locks that reside before the sentinel. HANDLER locks must be released explicitly as part of HANDLER CLOSE statement, or an implicit close. The same approach with sentinel is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES statement has never worked together, the implementation is made simple and only maintains one sentinel, which is used either for HANDLER locks, or for LOCK TABLES locks.
2009-12-22 17:09:15 +01:00
call release_transactional_locks() to release metadata locks which
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
might have been acquired successfully.
*/
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
thd->mdl_context.rollback_to_savepoint(ot_ctx.start_of_statement_svp());
table_list->mdl_request.ticket= 0;
if (ot_ctx.recover_from_failed_open(thd))
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
break;
}
2002-11-23 22:21:49 +01:00
if (!error)
2000-07-31 21:29:14 +02:00
{
/*
Backport of revno ## 2617.31.1, 2617.31.3, 2617.31.4, 2617.31.5, 2617.31.12, 2617.31.15, 2617.31.15, 2617.31.16, 2617.43.1 - initial changeset that introduced the fix for Bug#989 and follow up fixes for all test suite failures introduced in the initial changeset. ------------------------------------------------------------ revno: 2617.31.1 committer: Davi Arnaut <Davi.Arnaut@Sun.COM> branch nick: 4284-6.0 timestamp: Fri 2009-03-06 19:17:00 -0300 message: Bug#989: If DROP TABLE while there's an active transaction, wrong binlog order WL#4284: Transactional DDL locking Currently the MySQL server does not keep metadata locks on schema objects for the duration of a transaction, thus failing to guarantee the integrity of the schema objects being used during the transaction and to protect then from concurrent DDL operations. This also poses a problem for replication as a DDL operation might be replicated even thought there are active transactions using the object being modified. The solution is to defer the release of metadata locks until a active transaction is either committed or rolled back. This prevents other statements from modifying the table for the entire duration of the transaction. This provides commitment ordering for guaranteeing serializability across multiple transactions. - Incompatible change: If MySQL's metadata locking system encounters a lock conflict, the usual schema is to use the try and back-off technique to avoid deadlocks -- this schema consists in releasing all locks and trying to acquire them all in one go. But in a transactional context this algorithm can't be utilized as its not possible to release locks acquired during the course of the transaction without breaking the transaction commitments. To avoid deadlocks in this case, the ER_LOCK_DEADLOCK will be returned if a lock conflict is encountered during a transaction. Let's consider an example: A transaction has two statements that modify table t1, then table t2, and then commits. The first statement of the transaction will acquire a shared metadata lock on table t1, and it will be kept utill COMMIT to ensure serializability. At the moment when the second statement attempts to acquire a shared metadata lock on t2, a concurrent ALTER or DROP statement might have locked t2 exclusively. The prescription of the current locking protocol is that the acquirer of the shared lock backs off -- gives up all his current locks and retries. This implies that the entire multi-statement transaction has to be rolled back. - Incompatible change: FLUSH commands such as FLUSH PRIVILEGES and FLUSH TABLES WITH READ LOCK won't cause locked tables to be implicitly unlocked anymore.
2009-12-05 00:02:48 +01:00
We can't have a view or some special "open_strategy" in this function
so there should be a TABLE instance.
*/
DBUG_ASSERT(table_list->table);
table= table_list->table;
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
if (table->file->ht->db_type == DB_TYPE_MRG_MYISAM)
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
{
/* A MERGE table must not come here. */
/* purecov: begin tested */
my_error(ER_WRONG_OBJECT, MYF(0), table->s->db.str,
table->s->table_name.str, "BASE TABLE");
table= 0;
goto end;
/* purecov: end */
}
table_list->lock_type= lock_type;
2000-07-31 21:29:14 +02:00
table->grant= table_list->grant;
if (thd->locked_tables_mode)
2000-07-31 21:29:14 +02:00
{
if (check_lock_and_start_stmt(thd, thd->lex, table_list))
table= 0;
}
else
{
DBUG_ASSERT(thd->lock == 0); // You must lock everything at once
if ((table->reginfo.lock_type= lock_type) != TL_UNLOCK)
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
if (! (thd->lock= mysql_lock_tables(thd, &table_list->table, 1,
lock_flags)))
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
table= 0;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
2000-07-31 21:29:14 +02:00
}
}
else
table= 0;
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
end:
if (table == NULL)
{
if (!thd->in_sub_stmt)
trans_rollback_stmt(thd);
close_thread_tables(thd);
}
Prevent bugs by making DBUG_* expressions syntactically equivalent to a single statement. --- Bug#24795: SHOW PROFILE Profiling is only partially functional on some architectures. Where there is no getrusage() system call, presently Null values are returned where it would be required. Notably, Windows needs some love applied to make it as useful. Syntax this adds: SHOW PROFILES SHOW PROFILE [types] [FOR QUERY n] [OFFSET n] [LIMIT n] where "n" is an integer and "types" is zero or many (comma-separated) of "CPU" "MEMORY" (not presently supported) "BLOCK IO" "CONTEXT SWITCHES" "PAGE FAULTS" "IPC" "SWAPS" "SOURCE" "ALL" It also adds a session variable (boolean) "profiling", set to "no" by default, and (integer) profiling_history_size, set to 15 by default. This patch abstracts setting THDs' "proc_info" behind a macro that can be used as a hook into the profiling code when profiling support is compiled in. All future code in this line should use that mechanism for setting thd->proc_info. --- Tests are now set to omit the statistics. --- Adds an Information_schema table, "profiling" for access to "show profile" data. --- Merge zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-community-3--bug24795 into zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-community --- Fix merge problems. --- Fixed one bug in the query_source being NULL. Updated test results. --- Include more thorough profiling tests. Improve support for prepared statements. Use session-specific query IDs, starting at zero. --- Selecting from I_S.profiling is no longer quashed in profiling, as requested by Giuseppe. Limit the size of captured query text. No longer log queries that are zero length.
2007-02-22 16:03:08 +01:00
thd_proc_info(thd, 0);
2000-07-31 21:29:14 +02:00
DBUG_RETURN(table);
}
/**
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
Open all tables in list, locks them and optionally process derived tables.
2004-06-25 15:52:01 +02:00
@param thd Thread context.
@param tables List of tables for open and locking.
@param derived If to handle derived tables.
@param flags Bitmap of options to be used to open and lock
tables (see open_tables() and mysql_lock_tables()
for details).
@param prelocking_strategy Strategy which specifies how prelocking algorithm
should work for this statement.
2004-06-25 15:52:01 +02:00
@note
The thr_lock locks will automatically be freed by
close_thread_tables().
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
@retval FALSE OK.
@retval TRUE Error
2004-06-25 15:52:01 +02:00
*/
bool open_and_lock_tables(THD *thd, TABLE_LIST *tables,
bool derived, uint flags,
Prelocking_strategy *prelocking_strategy)
2004-06-25 15:52:01 +02:00
{
uint counter;
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
MDL_savepoint mdl_savepoint= thd->mdl_context.mdl_savepoint();
DBUG_ENTER("open_and_lock_tables");
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
DBUG_PRINT("enter", ("derived handling: %d", derived));
if (open_tables(thd, &tables, &counter, flags, prelocking_strategy))
goto err;
DBUG_EXECUTE_IF("sleep_open_and_lock_after_open", {
const char *old_proc_info= thd->proc_info;
thd->proc_info= "DBUG sleep";
my_sleep(6000000);
thd->proc_info= old_proc_info;});
if (lock_tables(thd, tables, counter, flags))
goto err;
Bug #58730 Assertion failed: table->key_read == 0 in close_thread_table, temptable views The TABLE::key_read field indicates if the optimizer has found that row retrieval only should access the index tree. The triggered assert inside close_thread_table() checks that this field has been reset when the table is about to be closed. During normal execution, these fields are reset right before tables are closed at the end of mysql_execute_command(). But in the case of errors, tables are closed earlier. The patch for Bug#52044 refactored the open tables code so that close_thread_tables() is called immediately if opening of tables fails. At this point in the execution, it could happend that all TABLE::key_read fields had not been properly reset, therefore triggering the assert. The problematic statement in this case was EXPLAIN where the query accessed two derived tables and where the first derived table was processed successfully while the second derived table was not. Since it was an EXPLAIN, TABLE::key_read fields were not reset after successful derived table processing since the state needs to be accessible afterwards. When processing of the second derived table failed, it's corresponding SELECT_LEX_UNIT was cleaned, which caused it's TABLE::key_read fields to be reset. Since processing failed, the error path of open_and_lock_tables() was entered and close_thread_tables() was called. The assert was then triggered due to the TABLE::key_read fields set during processing of the first derived table. This patch fixes the problem by adding a new derived table processor, mysql_derived_cleanup() that is called after mysql_derived_filling(). It causes cleanup of all SELECT_LEX_UNITs to be called, resetting all relevant TABLE::key_read fields. Test case added to derived.test.
2010-12-16 10:55:23 +01:00
if (derived)
{
if (mysql_handle_derived(thd->lex, &mysql_derived_prepare))
goto err;
if (thd->fill_derived_tables() &&
mysql_handle_derived(thd->lex, &mysql_derived_filling))
{
mysql_handle_derived(thd->lex, &mysql_derived_cleanup);
goto err;
}
if (!thd->lex->describe)
mysql_handle_derived(thd->lex, &mysql_derived_cleanup);
}
DBUG_RETURN(FALSE);
err:
if (! thd->in_sub_stmt)
trans_rollback_stmt(thd); /* Necessary if derived handling failed. */
close_thread_tables(thd);
/* Don't keep locks for a failed statement. */
thd->mdl_context.rollback_to_savepoint(mdl_savepoint);
DBUG_RETURN(TRUE);
}
/*
Open all tables in list and process derived tables
SYNOPSIS
open_normal_and_derived_tables
thd - thread handler
tables - list of tables for open
flags - bitmap of flags to modify how the tables will be open:
MYSQL_LOCK_IGNORE_FLUSH - open table even if someone has
done a flush on it.
RETURN
FALSE - ok
TRUE - error
NOTE
This is to be used on prepare stage when you don't read any
data from the tables.
*/
bool open_normal_and_derived_tables(THD *thd, TABLE_LIST *tables, uint flags)
{
DML_prelocking_strategy prelocking_strategy;
uint counter;
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
MDL_savepoint mdl_savepoint= thd->mdl_context.mdl_savepoint();
DBUG_ENTER("open_normal_and_derived_tables");
DBUG_ASSERT(!thd->fill_derived_tables());
if (open_tables(thd, &tables, &counter, flags, &prelocking_strategy) ||
mysql_handle_derived(thd->lex, &mysql_derived_prepare))
goto end;
DBUG_RETURN(0);
end:
/*
No need to commit/rollback the statement transaction: it's
either not started or we're filling in an INFORMATION_SCHEMA
table on the fly, and thus mustn't manipulate with the
transaction of the enclosing statement.
*/
DBUG_ASSERT(thd->transaction.stmt.is_empty() ||
(thd->state_flags & Open_tables_state::BACKUPS_AVAIL));
close_thread_tables(thd);
/* Don't keep locks for a failed statement. */
thd->mdl_context.rollback_to_savepoint(mdl_savepoint);
DBUG_RETURN(TRUE); /* purecov: inspected */
}
/*
Mark all real tables in the list as free for reuse.
SYNOPSIS
mark_real_tables_as_free_for_reuse()
thd - thread context
table - head of the list of tables
DESCRIPTION
Marks all real tables in the list (i.e. not views, derived
or schema tables) as free for reuse.
*/
static void mark_real_tables_as_free_for_reuse(TABLE_LIST *table_list)
{
TABLE_LIST *table;
for (table= table_list; table; table= table->next_global)
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
if (!table->placeholder())
{
table->table->query_id= 0;
}
for (table= table_list; table; table= table->next_global)
if (!table->placeholder())
{
/*
Detach children of MyISAMMRG tables used in
sub-statements, they will be reattached at open.
This has to be done in a separate loop to make sure
that children have had their query_id cleared.
*/
table->table->file->extra(HA_EXTRA_DETACH_CHILDREN);
}
}
/**
Lock all tables in a list.
@param thd Thread handler
@param tables Tables to lock
@param count Number of opened tables
@param flags Options (see mysql_lock_tables() for details)
You can't call lock_tables() while holding thr_lock locks, as
this would break the dead-lock-free handling thr_lock gives us.
You must always get all needed locks at once.
If the query for which we are calling this function is marked as
requiring prelocking, this function will change
locked_tables_mode to LTM_PRELOCKED.
@retval FALSE Success.
@retval TRUE A lock wait timeout, deadlock or out of memory.
*/
bool lock_tables(THD *thd, TABLE_LIST *tables, uint count,
uint flags)
2000-07-31 21:29:14 +02:00
{
2001-04-19 19:41:19 +02:00
TABLE_LIST *table;
DBUG_ENTER("lock_tables");
/*
We can't meet statement requiring prelocking if we already
in prelocked mode.
*/
DBUG_ASSERT(thd->locked_tables_mode <= LTM_LOCK_TABLES ||
!thd->lex->requires_prelocking());
if (!tables && !thd->lex->requires_prelocking())
BUG#39934: Slave stops for engine that only support row-based logging General overview: The logic for switching to row format when binlog_format=MIXED had numerous flaws. The underlying problem was the lack of a consistent architecture. General purpose of this changeset: This changeset introduces an architecture for switching to row format when binlog_format=MIXED. It enforces the architecture where it has to. It leaves some bugs to be fixed later. It adds extensive tests to verify that unsafe statements work as expected and that appropriate errors are produced by problems with the selection of binlog format. It was not practical to split this into smaller pieces of work. Problem 1: To determine the logging mode, the code has to take several parameters into account (namely: (1) the value of binlog_format; (2) the capabilities of the engines; (3) the type of the current statement: normal, unsafe, or row injection). These parameters may conflict in several ways, namely: - binlog_format=STATEMENT for a row injection - binlog_format=STATEMENT for an unsafe statement - binlog_format=STATEMENT for an engine only supporting row logging - binlog_format=ROW for an engine only supporting statement logging - statement is unsafe and engine does not support row logging - row injection in a table that does not support statement logging - statement modifies one table that does not support row logging and one that does not support statement logging Several of these conflicts were not detected, or were detected with an inappropriate error message. The problem of BUG#39934 was that no appropriate error message was written for the case when an engine only supporting row logging executed a row injection with binlog_format=ROW. However, all above cases must be handled. Fix 1: Introduce new error codes (sql/share/errmsg.txt). Ensure that all conditions are detected and handled in decide_logging_format() Problem 2: The binlog format shall be determined once per statement, in decide_logging_format(). It shall not be changed before or after that. Before decide_logging_format() is called, all information necessary to determine the logging format must be available. This principle ensures that all unsafe statements are handled in a consistent way. However, this principle is not followed: thd->set_current_stmt_binlog_row_based_if_mixed() is called in several places, including from code executing UPDATE..LIMIT, INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and SET @@binlog_format. After Problem 1 was fixed, that caused inconsistencies where these unsafe statements would not print the appropriate warnings or errors for some of the conflicts. Fix 2: Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from code executed after decide_logging_format(). Compensate by calling the set_current_stmt_unsafe() at parse time. This way, all unsafe statements are detected by decide_logging_format(). Problem 3: INSERT DELAYED is not unsafe: it is logged in statement format even if binlog_format=MIXED, and no warning is printed even if binlog_format=STATEMENT. This is BUG#45825. Fix 3: Made INSERT DELAYED set itself to unsafe at parse time. This allows decide_logging_format() to detect that a warning should be printed or the binlog_format changed. Problem 4: LIMIT clause were not marked as unsafe when executed inside stored functions/triggers/views/prepared statements. This is BUG#45785. Fix 4: Make statements containing the LIMIT clause marked as unsafe at parse time, instead of at execution time. This allows propagating unsafe-ness to the view.
2009-07-14 21:31:19 +02:00
DBUG_RETURN(thd->decide_logging_format(tables));
/*
Check for thd->locked_tables_mode to avoid a redundant
and harmful attempt to lock the already locked tables again.
Checking for thd->lock is not enough in some situations. For example,
if a stored function contains
"drop table t3; create temporary t3 ..; insert into t3 ...;"
thd->lock may be 0 after drop tables, whereas locked_tables_mode
is still on. In this situation an attempt to lock temporary
table t3 will lead to a memory leak.
*/
if (! thd->locked_tables_mode)
2000-07-31 21:29:14 +02:00
{
DBUG_ASSERT(thd->lock == 0); // You must lock everything at once
2000-07-31 21:29:14 +02:00
TABLE **start,**ptr;
if (!(ptr=start=(TABLE**) thd->alloc(sizeof(TABLE*)*count)))
DBUG_RETURN(TRUE);
2004-07-16 00:15:55 +02:00
for (table= tables; table; table= table->next_global)
2004-06-25 15:52:01 +02:00
{
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
if (!table->placeholder())
2004-06-25 15:52:01 +02:00
*(ptr++)= table->table;
}
/* We have to emulate LOCK TABLES if we are statement needs prelocking. */
if (thd->lex->requires_prelocking())
{
* Mixed replication mode * : 1) Fix for BUG#19630 "stored function inserting into two auto_increment breaks statement-based binlog": a stored function inserting into two such tables may fail to replicate (inserting wrong data in the slave's copy of the second table) if the slave's second table had an internal auto_increment counter different from master's. Because the auto_increment value autogenerated by master for the 2nd table does not go into binlog, only the first does, so the slave lacks information. To fix this, if running in mixed binlogging mode, if the stored function or trigger plans to update two different tables both having auto_increment columns, we switch to row-based for the whole function. We don't have a simple solution for statement-based binlogging mode, there the bug remains and will be documented as a known problem. Re-enabling rpl_switch_stm_row_mixed. 2) Fix for BUG#20630 "Mixed binlogging mode does not work with stored functions, triggers, views", which was a documented limitation (in mixed mode, we didn't detect that a stored function's execution needed row-based binlogging (due to some UUID() call for example); same for triggers, same for views (a view created from a SELECT UUID(), and doing INSERT INTO sometable SELECT theview; would not replicate row-based). This is implemented by, after parsing a routine's body, remembering in sp_head that this routine needs row-based binlogging. Then when this routine is used, the caller is marked to require row-based binlogging too. Same for views: when we parse a view and detect that its SELECT needs row-based binary logging, we mark the calling LEX as such. 3) Fix for BUG#20499 "mixed mode with temporary table breaks binlog": a temporary table containing e.g. UUID has its changes not binlogged, so any query updating a permanent table with data from the temporary table will run wrongly on slave. Solution: in mixed mode we don't switch back from row-based to statement-based when there exists temporary tables. 4) Attempt to test mysqlbinlog on a binlog generated by mysqlbinlog; impossible due to BUG#11312 and BUG#20329, but test is in place for when they are fixed.
2006-07-09 17:00:47 +02:00
/*
A query that modifies autoinc column in sub-statement can make the
master and slave inconsistent.
We can solve these problems in mixed mode by switching to binlogging
if at least one updated table is used by sub-statement
* Mixed replication mode * : 1) Fix for BUG#19630 "stored function inserting into two auto_increment breaks statement-based binlog": a stored function inserting into two such tables may fail to replicate (inserting wrong data in the slave's copy of the second table) if the slave's second table had an internal auto_increment counter different from master's. Because the auto_increment value autogenerated by master for the 2nd table does not go into binlog, only the first does, so the slave lacks information. To fix this, if running in mixed binlogging mode, if the stored function or trigger plans to update two different tables both having auto_increment columns, we switch to row-based for the whole function. We don't have a simple solution for statement-based binlogging mode, there the bug remains and will be documented as a known problem. Re-enabling rpl_switch_stm_row_mixed. 2) Fix for BUG#20630 "Mixed binlogging mode does not work with stored functions, triggers, views", which was a documented limitation (in mixed mode, we didn't detect that a stored function's execution needed row-based binlogging (due to some UUID() call for example); same for triggers, same for views (a view created from a SELECT UUID(), and doing INSERT INTO sometable SELECT theview; would not replicate row-based). This is implemented by, after parsing a routine's body, remembering in sp_head that this routine needs row-based binlogging. Then when this routine is used, the caller is marked to require row-based binlogging too. Same for views: when we parse a view and detect that its SELECT needs row-based binary logging, we mark the calling LEX as such. 3) Fix for BUG#20499 "mixed mode with temporary table breaks binlog": a temporary table containing e.g. UUID has its changes not binlogged, so any query updating a permanent table with data from the temporary table will run wrongly on slave. Solution: in mixed mode we don't switch back from row-based to statement-based when there exists temporary tables. 4) Attempt to test mysqlbinlog on a binlog generated by mysqlbinlog; impossible due to BUG#11312 and BUG#20329, but test is in place for when they are fixed.
2006-07-09 17:00:47 +02:00
*/
if (thd->variables.binlog_format != BINLOG_FORMAT_ROW && tables &&
has_write_table_with_auto_increment(thd->lex->first_not_own_table()))
thd->lex->set_stmt_unsafe(LEX::BINLOG_STMT_UNSAFE_AUTOINC_COLUMNS);
}
DEBUG_SYNC(thd, "before_lock_tables_takes_lock");
if (! (thd->lock= mysql_lock_tables(thd, start, (uint) (ptr - start),
flags)))
DBUG_RETURN(TRUE);
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
DEBUG_SYNC(thd, "after_lock_tables_takes_lock");
if (thd->lex->requires_prelocking() &&
thd->lex->sql_command != SQLCOM_LOCK_TABLES)
{
TABLE_LIST *first_not_own= thd->lex->first_not_own_table();
/*
We just have done implicit LOCK TABLES, and now we have
to emulate first open_and_lock_tables() after it.
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
When open_and_lock_tables() is called for a single table out of
a table list, the 'next_global' chain is temporarily broken. We
may not find 'first_not_own' before the end of the "list".
Look for example at those places where open_n_lock_single_table()
is called. That function implements the temporary breaking of
a table list for opening a single table.
*/
for (table= tables;
table && table != first_not_own;
table= table->next_global)
{
Bug#8407 (Stored functions/triggers ignore exception handler) Bug 18914 (Calling certain SPs from triggers fail) Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02') Bug 21825 (Incorrect message error deleting records in a table with a trigger for inserting) Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency error) Bug 25345 (Cursors from Functions) This fix resolves a long standing issue originally reported with bug 8407, which affect the behavior of Stored Procedures, Stored Functions and Trigger in many different ways, causing symptoms reported by all the bugs listed. In all cases, the root cause of the problem traces back to 8407 and how the server locks tables involved with sub statements. Prior to this fix, the implementation of stored routines would: - compute the transitive closure of all the tables referenced by a top level statement - open and lock all the tables involved - execute the top level statement "transitive closure of tables" means collecting: - all the tables, - all the stored functions, - all the views, - all the table triggers - all the stored procedures involved, and recursively inspect these objects definition to find more references to more objects, until the list of every object referenced does not grow any more. This mechanism is known as "pre-locking" tables before execution. The motivation for locking all the tables (possibly) used at once is to prevent dead locks. One problem with this approach is that, if the execution path the code really takes during runtime does not use a given table, and if the table is missing, the server would not execute the statement. This in particular has a major impact on triggers, since a missing table referenced by an update/delete trigger would prevent an insert trigger to run. Another problem is that stored routines might define SQL exception handlers to deal with missing tables, but the server implementation would never give user code a chance to execute this logic, since the routine is never executed when a missing table cause the pre-locking code to fail. With this fix, the internal implementation of the pre-locking code has been relaxed of some constraints, so that failure to open a table does not necessarily prevent execution of a stored routine. In particular, the pre-locking mechanism is now behaving as follows: 1) the first step, to compute the transitive closure of all the tables possibly referenced by a statement, is unchanged. 2) the next step, which is to open all the tables involved, only attempts to open the tables added by the pre-locking code, but silently fails without reporting any error or invoking any exception handler is the table is not present. This is achieved by trapping internal errors with Prelock_error_handler 3) the locking step only locks tables that were successfully opened. 4) when executing sub statements, the list of tables used by each statements is evaluated as before. The tables needed by the sub statement are expected to be already opened and locked. Statement referencing tables that were not opened in step 2) will fail to find the table in the open list, and only at this point will execution of the user code fail. 5) when a runtime exception is raised at 4), the instruction continuation destination (the next instruction to execute in case of SQL continue handlers) is evaluated. This is achieved with sp_instr::exec_open_and_lock_tables() 6) if a user exception handler is present in the stored routine, that handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be trapped by stored routines. If no handler exists, then the runtime execution will fail as expected. With all these changes, a side effect is that view security is impacted, in two different ways. First, a view defined as "select stored_function()", where the stored function references a table that may not exist, is considered valid. The rationale is that, because the stored function might trap exceptions during execution and still return a valid result, there is no way to decide when the view is created if a missing table really cause the view to be invalid. Secondly, testing for existence of tables is now done later during execution. View security, which consist of trapping errors and return a generic ER_VIEW_INVALID (to prevent disclosing information) was only implemented at very specific phases covering *opening* tables, but not covering the runtime execution. Because of this existing limitation, errors that were previously trapped and converted into ER_VIEW_INVALID are not trapped, causing table names to be reported to the user. This change is exposing an existing problem, which is independent and will be resolved separately.
2007-03-06 03:42:07 +01:00
if (!table->placeholder())
{
table->table->query_id= thd->query_id;
if (check_lock_and_start_stmt(thd, thd->lex, table))
{
mysql_unlock_tables(thd, thd->lock);
thd->lock= 0;
DBUG_RETURN(TRUE);
}
}
}
/*
Let us mark all tables which don't belong to the statement itself,
and was marked as occupied during open_tables() as free for reuse.
*/
mark_real_tables_as_free_for_reuse(first_not_own);
Backport of: ---------------------------------------------------------- revno: 2630.4.38 committer: Konstantin Osipov <konstantin@mysql.com> branch nick: mysql-6.0-4144 timestamp: Wed 2008-06-25 22:07:06 +0400 message: WL#4144 - Lock MERGE engine children. Committing a version of the patch merged with WL#3726 on behalf of Ingo. Step #1: Move locking from parent to children. MERGE children are now left in the query list of tables after inserted there in open_tables(). So they are locked by lock_tables() as all other tables are. The MERGE parent does not store locks any more. It appears in a MYSQL_LOCK with zero lock data. This is kind of a "dummy" lock. All other lock handling is also done directly on the children. To protect against parent or child modifications during LOCK TABLES, the children are detached after every statement and attached before every statement, even under LOCK TABLES. The children table list is removed from the query list of tables on every detach and on close of the parent. Step #2: Move MERGE specific functionality from SQL layer into table handler. Functionality moved from SQL layer (mainly sql_base.cc) to the table handler (ha_myisammrg.cc). Unnecessary code is removed from the SQL layer. Step #3: Moved all MERGE specific members from TABLE to ha_myisammrg. Moved members from TABLE to ha_myisammrg. Renamed some mebers. Fixed comments. Step #4: Valgrind and coverage testing Valgrind did not uncover new problems. Added purecov comments. Added a new test for DATA/INDEX DIRECTORY options. Changed handling of ::reset() for non-attached children. Fixed the merge-big test. Step #5: Fixed crashes detected during review Changed detection when to attach/detach. Added new tests. Backport also the fix for Bug#44040 "MySQL allows creating a MERGE table upon VIEWs but crashes when using it"
2009-12-03 00:09:22 +01:00
DBUG_PRINT("info",("locked_tables_mode= LTM_PRELOCKED"));
A prerequisite patch for the fix for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". Introduce a notion of a sentinel to MDL_context. A sentinel is a ticket that separates all tickets in the context into two groups: before and after it. Currently we can have (and need) only one designated sentinel -- it separates all locks taken by LOCK TABLE or HANDLER statement, which must survive COMMIT and ROLLBACK and all other locks, which must be released at COMMIT or ROLLBACK. The tricky part is maintaining the sentinel up to date when someone release its corresponding ticket. This can happen, e.g. if someone issues DROP TABLE under LOCK TABLES (generally, see all calls to release_all_locks_for_name()). MDL_context::release_ticket() is modified to take care of it. ****** A fix and a test case for Bug#46224 "HANDLER statements within a transaction might lead to deadlocks". An attempt to mix HANDLER SQL statements, which are transaction- agnostic, an open multi-statement transaction, and DDL against the involved tables (in a concurrent connection) could lead to a deadlock. The deadlock would occur when HANDLER OPEN or HANDLER READ would have to wait on a conflicting metadata lock. If the connection that issued HANDLER statement also had other metadata locks (say, acquired in scope of a transaction), a classical deadlock situation of mutual wait could occur. Incompatible change: entering LOCK TABLES mode automatically closes all open HANDLERs in the current connection. Incompatible change: previously an attempt to wait on a lock in a connection that has an open HANDLER statement could wait indefinitely/deadlock. After this patch, an error ER_LOCK_DEADLOCK is produced. The idea of the fix is to merge thd->handler_mdl_context with the main mdl_context of the connection, used for transactional locks. This makes deadlock detection possible, since all waits with locks are "visible" and available to analysis in a single MDL context of the connection. Since HANDLER locks and transactional locks have a different life cycle -- HANDLERs are explicitly open and closed, and so are HANDLER locks, explicitly acquired and released, whereas transactional locks "accumulate" till the end of a transaction and are released only with COMMIT, ROLLBACK and ROLLBACK TO SAVEPOINT, a concept of "sentinel" was introduced to MDL_context. All locks, HANDLER and others, reside in the same linked list. However, a selected element of the list separates locks with different life cycle. HANDLER locks always reside at the end of the list, after the sentinel. Transactional locks are prepended to the beginning of the list, before the sentinel. Thus, ROLLBACK, COMMIT or ROLLBACK TO SAVEPOINT, only release those locks that reside before the sentinel. HANDLER locks must be released explicitly as part of HANDLER CLOSE statement, or an implicit close. The same approach with sentinel is also employed for LOCK TABLES locks. Since HANDLER and LOCK TABLES statement has never worked together, the implementation is made simple and only maintains one sentinel, which is used either for HANDLER locks, or for LOCK TABLES locks.
2009-12-22 17:09:15 +01:00
thd->enter_locked_tables_mode(LTM_PRELOCKED);
}
2000-07-31 21:29:14 +02:00
}
2001-04-19 19:41:19 +02:00
else
{
TABLE_LIST *first_not_own= thd->lex->first_not_own_table();
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
/*
When open_and_lock_tables() is called for a single table out of
a table list, the 'next_global' chain is temporarily broken. We
may not find 'first_not_own' before the end of the "list".
Look for example at those places where open_n_lock_single_table()
is called. That function implements the temporary breaking of
a table list for opening a single table.
*/
for (table= tables;
table && table != first_not_own;
table= table->next_global)
2001-04-19 19:41:19 +02:00
{
if (table->placeholder())
continue;
/*
In a stored function or trigger we should ensure that we won't change
a table that is already used by the calling statement.
*/
if (thd->locked_tables_mode >= LTM_PRELOCKED &&
table->lock_type >= TL_WRITE_ALLOW_WRITE)
{
for (TABLE* opentab= thd->open_tables; opentab; opentab= opentab->next)
{
2009-03-27 13:55:14 +01:00
if (table->table->s == opentab->s && opentab->query_id &&
table->table->query_id != opentab->query_id)
{
my_error(ER_CANT_UPDATE_USED_TABLE_IN_SF_OR_TRG, MYF(0),
2009-03-27 13:55:14 +01:00
table->table->s->table_name.str);
DBUG_RETURN(TRUE);
}
}
}
if (check_lock_and_start_stmt(thd, thd->lex, table))
2001-04-19 19:41:19 +02:00
{
DBUG_RETURN(TRUE);
2001-04-19 19:41:19 +02:00
}
}
/*
If we are under explicit LOCK TABLES and our statement requires
prelocking, we should mark all "additional" tables as free for use
and enter prelocked mode.
*/
if (thd->lex->requires_prelocking())
{
mark_real_tables_as_free_for_reuse(first_not_own);
DBUG_PRINT("info",
("thd->locked_tables_mode= LTM_PRELOCKED_UNDER_LOCK_TABLES"));
thd->locked_tables_mode= LTM_PRELOCKED_UNDER_LOCK_TABLES;
}
2001-04-19 19:41:19 +02:00
}
BUG#39934: Slave stops for engine that only support row-based logging General overview: The logic for switching to row format when binlog_format=MIXED had numerous flaws. The underlying problem was the lack of a consistent architecture. General purpose of this changeset: This changeset introduces an architecture for switching to row format when binlog_format=MIXED. It enforces the architecture where it has to. It leaves some bugs to be fixed later. It adds extensive tests to verify that unsafe statements work as expected and that appropriate errors are produced by problems with the selection of binlog format. It was not practical to split this into smaller pieces of work. Problem 1: To determine the logging mode, the code has to take several parameters into account (namely: (1) the value of binlog_format; (2) the capabilities of the engines; (3) the type of the current statement: normal, unsafe, or row injection). These parameters may conflict in several ways, namely: - binlog_format=STATEMENT for a row injection - binlog_format=STATEMENT for an unsafe statement - binlog_format=STATEMENT for an engine only supporting row logging - binlog_format=ROW for an engine only supporting statement logging - statement is unsafe and engine does not support row logging - row injection in a table that does not support statement logging - statement modifies one table that does not support row logging and one that does not support statement logging Several of these conflicts were not detected, or were detected with an inappropriate error message. The problem of BUG#39934 was that no appropriate error message was written for the case when an engine only supporting row logging executed a row injection with binlog_format=ROW. However, all above cases must be handled. Fix 1: Introduce new error codes (sql/share/errmsg.txt). Ensure that all conditions are detected and handled in decide_logging_format() Problem 2: The binlog format shall be determined once per statement, in decide_logging_format(). It shall not be changed before or after that. Before decide_logging_format() is called, all information necessary to determine the logging format must be available. This principle ensures that all unsafe statements are handled in a consistent way. However, this principle is not followed: thd->set_current_stmt_binlog_row_based_if_mixed() is called in several places, including from code executing UPDATE..LIMIT, INSERT..SELECT..LIMIT, DELETE..LIMIT, INSERT DELAYED, and SET @@binlog_format. After Problem 1 was fixed, that caused inconsistencies where these unsafe statements would not print the appropriate warnings or errors for some of the conflicts. Fix 2: Remove calls to THD::set_current_stmt_binlog_row_based_if_mixed() from code executed after decide_logging_format(). Compensate by calling the set_current_stmt_unsafe() at parse time. This way, all unsafe statements are detected by decide_logging_format(). Problem 3: INSERT DELAYED is not unsafe: it is logged in statement format even if binlog_format=MIXED, and no warning is printed even if binlog_format=STATEMENT. This is BUG#45825. Fix 3: Made INSERT DELAYED set itself to unsafe at parse time. This allows decide_logging_format() to detect that a warning should be printed or the binlog_format changed. Problem 4: LIMIT clause were not marked as unsafe when executed inside stored functions/triggers/views/prepared statements. This is BUG#45785. Fix 4: Make statements containing the LIMIT clause marked as unsafe at parse time, instead of at execution time. This allows propagating unsafe-ness to the view.
2009-07-14 21:31:19 +02:00
DBUG_RETURN(thd->decide_logging_format(tables));
2000-07-31 21:29:14 +02:00
}
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
/**
Prepare statement for reopening of tables and recalculation of set of
prelocked tables.
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
@param[in] thd Thread context.
@param[in,out] tables List of tables which we were trying to open
and lock.
@param[in] start_of_statement_svp MDL savepoint which represents the set
of metadata locks which the current transaction
managed to acquire before execution of the current
statement and to which we should revert before
trying to reopen tables. NULL if no metadata locks
were held and thus all metadata locks should be
released.
*/
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
void close_tables_for_reopen(THD *thd, TABLE_LIST **tables,
Patch that refactors global read lock implementation and fixes bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ LOCK" and bug #54673 "It takes too long to get readlock for 'FLUSH TABLES WITH READ LOCK'". The first bug manifested itself as a deadlock which occurred when a connection, which had some table open through HANDLER statement, tried to update some data through DML statement while another connection tried to execute FLUSH TABLES WITH READ LOCK concurrently. What happened was that FTWRL in the second connection managed to perform first step of GRL acquisition and thus blocked all upcoming DML. After that it started to wait for table open through HANDLER statement to be flushed. When the first connection tried to execute DML it has started to wait for GRL/the second connection creating deadlock. The second bug manifested itself as starvation of FLUSH TABLES WITH READ LOCK statements in cases when there was a constant stream of concurrent DML statements (in two or more connections). This has happened because requests for protection against GRL which were acquired by DML statements were ignoring presence of pending GRL and thus the latter was starved. This patch solves both these problems by re-implementing GRL using metadata locks. Similar to the old implementation acquisition of GRL in new implementation is two-step. During the first step we block all concurrent DML and DDL statements by acquiring global S metadata lock (each DML and DDL statement acquires global IX lock for its duration). During the second step we block commits by acquiring global S lock in COMMIT namespace (commit code acquires global IX lock in this namespace). Note that unlike in old implementation acquisition of protection against GRL in DML and DDL is semi-automatic. We assume that any statement which should be blocked by GRL will either open and acquires write-lock on tables or acquires metadata locks on objects it is going to modify. For any such statement global IX metadata lock is automatically acquired for its duration. The first problem is solved because waits for GRL become visible to deadlock detector in metadata locking subsystem and thus deadlocks like one in the first bug become impossible. The second problem is solved because global S locks which are used for GRL implementation are given preference over IX locks which are acquired by concurrent DML (and we can switch to fair scheduling in future if needed). Important change: FTWRL/GRL no longer blocks DML and DDL on temporary tables. Before this patch behavior was not consistent in this respect: in some cases DML/DDL statements on temporary tables were blocked while in others they were not. Since the main use cases for FTWRL are various forms of backups and temporary tables are not preserved during backups we have opted for consistently allowing DML/DDL on temporary tables during FTWRL/GRL. Important change: This patch changes thread state names which are used when DML/DDL of FTWRL is waiting for global read lock. It is now either "Waiting for global read lock" or "Waiting for commit lock" depending on the stage on which FTWRL is. Incompatible change: To solve deadlock in events code which was exposed by this patch we have to replace LOCK_event_metadata mutex with metadata locks on events. As result we have to prohibit DDL on events under LOCK TABLES. This patch also adds extensive test coverage for interaction of DML/DDL and FTWRL. Performance of new and old global read lock implementations in sysbench tests were compared. There were no significant difference between new and old implementations.
2010-11-11 18:11:05 +01:00
const MDL_savepoint &start_of_statement_svp)
{
TABLE_LIST *first_not_own_table= thd->lex->first_not_own_table();
TABLE_LIST *tmp;
/*
If table list consists only from tables from prelocking set, table list
for new attempt should be empty, so we have to update list's root pointer.
*/
if (first_not_own_table == *tables)
*tables= 0;
thd->lex->chop_off_not_own_tables();
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
/* Reset MDL tickets for procedures/functions */
for (Sroutine_hash_entry *rt=
(Sroutine_hash_entry*)thd->lex->sroutines_list.first;
rt; rt= rt->next)
rt->mdl_request.ticket= NULL;
sp_remove_not_own_routines(thd->lex);
for (tmp= *tables; tmp; tmp= tmp->next_global)
{
tmp->table= 0;
tmp->mdl_request.ticket= NULL;
Apply and review: 3655 Jon Olav Hauglid 2009-10-19 Bug #30977 Concurrent statement using stored function and DROP FUNCTION breaks SBR Bug #48246 assert in close_thread_table Implement a fix for: Bug #41804 purge stored procedure cache causes mysterious hang for many minutes Bug #49972 Crash in prepared statements The problem was that concurrent execution of DML statements that use stored functions and DDL statements that drop/modify the same function might result in incorrect binary log in statement (and mixed) mode and therefore break replication. This patch fixes the problem by introducing metadata locking for stored procedures and functions. This is similar to what is done in Bug#25144 for views. Procedures and functions now are locked using metadata locks until the transaction is either committed or rolled back. This prevents other statements from modifying the procedure/function while it is being executed. This provides commit ordering - guaranteeing serializability across multiple transactions and thus fixes the reported binlog problem. Note that we do not take locks for top-level CALLs. This means that procedures called directly are not protected from changes by simultaneous DDL operations so they are executed at the state they had at the time of the CALL. By not taking locks for top-level CALLs, we still allow transactions to be started inside procedures. This patch also changes stored procedure cache invalidation. Upon a change of cache version, we no longer invalidate the entire cache, but only those routines which we use, only when a statement is executed that uses them. This patch also changes the logic of prepared statement validation. A stored procedure used by a prepared statement is now validated only once a metadata lock has been acquired. A version mismatch causes a flush of the obsolete routine from the cache and statement reprepare. Incompatible changes: 1) ER_LOCK_DEADLOCK is reported for a transaction trying to access a procedure/function that is locked by a DDL operation in another connection. 2) Procedure/function DDL operations are now prohibited in LOCK TABLES mode as exclusive locks must be taken all at once and LOCK TABLES provides no way to specifiy procedures/functions to be locked. Test cases have been added to sp-lock.test and rpl_sp.test. Work on this bug has very much been a team effort and this patch includes and is based on contributions from Davi Arnaut, Dmitry Lenev, Magne Mæhre and Konstantin Osipov.
2009-12-29 13:19:05 +01:00
/* We have to cleanup translation tables of views. */
tmp->cleanup_items();
}
/*
No need to commit/rollback the statement transaction: it's
either not started or we're filling in an INFORMATION_SCHEMA
table on the fly, and thus mustn't manipulate with the
transaction of the enclosing statement.
*/
DBUG_ASSERT(thd->transaction.stmt.is_empty() ||
(thd->state_flags & Open_tables_state::BACKUPS_AVAIL));
close_thread_tables(thd);
Implementation of simple deadlock detection for metadata locks. This change is supposed to reduce number of ER_LOCK_DEADLOCK errors which occur when multi-statement transaction encounters conflicting metadata lock in cases when waiting is possible. The idea is not to fail ER_LOCK_DEADLOCK error immediately when we encounter conflicting metadata lock. Instead we release all metadata locks acquired by current statement and start to wait until conflicting lock go away. To avoid deadlocks we use simple empiric which aborts waiting with ER_LOCK_DEADLOCK error if it turns out that somebody is waiting for metadata locks owned by this transaction. This patch also fixes bug #46273 "MySQL 5.4.4 new MDL: Bug#989 is not fully fixed in case of ALTER". The bug was that concurrent execution of UPDATE or MULTI-UPDATE statement as a part of multi-statement transaction that already has used table being updated and ALTER TABLE statement might have resulted of loss of isolation between this transaction and ALTER TABLE statement, which manifested itself as changes performed by ALTER TABLE becoming visible in transaction and wrong binary log order as a consequence. This problem occurred when UPDATE or MULTI-UPDATE's wait in mysql_lock_tables() call was aborted due to metadata lock upgrade performed by concurrent ALTER TABLE. After such abort all metadata locks held by transaction were released but transaction silently continued to be executed as if nothing has happened. We solve this problem by changing our code not to release all locks in such case. Instead we release only locks which were acquired by current statement and then try to reacquire them by restarting open/lock tables process. We piggyback on simple deadlock detector implementation since this change has to be done anyway for it.
2009-12-30 18:53:30 +01:00
thd->mdl_context.rollback_to_savepoint(start_of_statement_svp);
}
/**
Open a single table without table caching and don't add it to
THD::open_tables. Depending on the 'add_to_temporary_tables_list' value,
the opened TABLE instance will be addded to THD::temporary_tables list.
@param thd Thread context.
@param path Path (without .frm)
@param db Database name.
@param table_name Table name.
@param add_to_temporary_tables_list Specifies if the opened TABLE
instance should be linked into
THD::temporary_tables list.
@note This function is used:
- by alter_table() to open a temporary table;
- when creating a temporary table with CREATE TEMPORARY TABLE.
@return TABLE instance for opened table.
@retval NULL on error.
2000-07-31 21:29:14 +02:00
*/
TABLE *open_table_uncached(THD *thd, const char *path, const char *db,
const char *table_name,
bool add_to_temporary_tables_list)
2000-07-31 21:29:14 +02:00
{
TABLE *tmp_table;
TABLE_SHARE *share;
char cache_key[MAX_DBKEY_LENGTH], *saved_cache_key, *tmp_path;
uint key_length;
TABLE_LIST table_list;
DBUG_ENTER("open_table_uncached");
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
DBUG_PRINT("enter",
("table: '%s'.'%s' path: '%s' server_id: %u "
"pseudo_thread_id: %lu",
db, table_name, path,
(uint) thd->server_id, (ulong) thd->variables.pseudo_thread_id));
table_list.db= (char*) db;
table_list.table_name= (char*) table_name;
/* Create the cache_key for temporary tables */
key_length= create_table_def_key(thd, cache_key, &table_list, 1);
if (!(tmp_table= (TABLE*) my_malloc(sizeof(*tmp_table) + sizeof(*share) +
strlen(path)+1 + key_length,
MYF(MY_WME))))
2000-07-31 21:29:14 +02:00
DBUG_RETURN(0); /* purecov: inspected */
share= (TABLE_SHARE*) (tmp_table+1);
tmp_path= (char*) (share+1);
saved_cache_key= strmov(tmp_path, path)+1;
memcpy(saved_cache_key, cache_key, key_length);
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
init_tmp_table_share(thd, share, saved_cache_key, key_length,
strend(saved_cache_key)+1, tmp_path);
if (open_table_def(thd, share, 0) ||
open_table_from_share(thd, share, table_name,
(uint) (HA_OPEN_KEYFILE | HA_OPEN_RNDFILE |
HA_GET_INDEX),
READ_KEYINFO | COMPUTE_TYPES | EXTRA_RECORD,
ha_open_options,
tmp_table, FALSE))
2000-07-31 21:29:14 +02:00
{
/* No need to lock share->mutex as this is not needed for tmp tables */
free_table_share(share);
Bug#34043: Server loops excessively in _checkchunk() when safemalloc is enabled Essentially, the problem is that safemalloc is excruciatingly slow as it checks all allocated blocks for overrun at each memory management primitive, yielding a almost exponential slowdown for the memory management functions (malloc, realloc, free). The overrun check basically consists of verifying some bytes of a block for certain magic keys, which catches some simple forms of overrun. Another minor problem is violation of aliasing rules and that its own internal list of blocks is prone to corruption. Another issue with safemalloc is rather the maintenance cost as the tool has a significant impact on the server code. Given the magnitude of memory debuggers available nowadays, especially those that are provided with the platform malloc implementation, maintenance of a in-house and largely obsolete memory debugger becomes a burden that is not worth the effort due to its slowness and lack of support for detecting more common forms of heap corruption. Since there are third-party tools that can provide the same functionality at a lower or comparable performance cost, the solution is to simply remove safemalloc. Third-party tools can provide the same functionality at a lower or comparable performance cost. The removal of safemalloc also allows a simplification of the malloc wrappers, removing quite a bit of kludge: redefinition of my_malloc, my_free and the removal of the unused second argument of my_free. Since free() always check whether the supplied pointer is null, redudant checks are also removed. Also, this patch adds unit testing for my_malloc and moves my_realloc implementation into the same file as the other memory allocation primitives.
2010-07-08 23:20:08 +02:00
my_free(tmp_table);
2000-07-31 21:29:14 +02:00
DBUG_RETURN(0);
}
tmp_table->reginfo.lock_type= TL_WRITE; // Simulate locked
share->tmp_table= (tmp_table->file->has_transactions() ?
TRANSACTIONAL_TMP_TABLE : NON_TRANSACTIONAL_TMP_TABLE);
if (add_to_temporary_tables_list)
2000-07-31 21:29:14 +02:00
{
/* growing temp list at the head */
tmp_table->next= thd->temporary_tables;
if (tmp_table->next)
tmp_table->next->prev= tmp_table;
thd->temporary_tables= tmp_table;
thd->temporary_tables->prev= 0;
2001-11-25 12:50:48 +01:00
if (thd->slave_thread)
slave_open_temp_tables++;
2000-07-31 21:29:14 +02:00
}
tmp_table->pos_in_table_list= 0;
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
DBUG_PRINT("tmptable", ("opened table: '%s'.'%s' 0x%lx", tmp_table->s->db.str,
tmp_table->s->table_name.str, (long) tmp_table));
2000-07-31 21:29:14 +02:00
DBUG_RETURN(tmp_table);
}
bool rm_temporary_table(handlerton *base, char *path)
2000-07-31 21:29:14 +02:00
{
bool error=0;
handler *file;
char *ext;
DBUG_ENTER("rm_temporary_table");
strmov(ext= strend(path), reg_ext);
if (mysql_file_delete(key_file_frm, path, MYF(0)))
2000-07-31 21:29:14 +02:00
error=1; /* purecov: inspected */
*ext= 0; // remove extension
file= get_new_handler((TABLE_SHARE*) 0, current_thd->mem_root, base);
if (file && file->ha_delete_table(path))
2004-06-25 15:52:01 +02:00
{
2000-07-31 21:29:14 +02:00
error=1;
sql_print_warning("Could not remove temporary table: '%s', error: %d",
2004-09-04 20:17:09 +02:00
path, my_errno);
2004-06-25 15:52:01 +02:00
}
2000-07-31 21:29:14 +02:00
delete file;
DBUG_RETURN(error);
2000-07-31 21:29:14 +02:00
}
/*****************************************************************************
* The following find_field_in_XXX procedures implement the core of the
* name resolution functionality. The entry point to resolve a column name in a
* list of tables is 'find_field_in_tables'. It calls 'find_field_in_table_ref'
* for each table reference. In turn, depending on the type of table reference,
* 'find_field_in_table_ref' calls one of the 'find_field_in_XXX' procedures
* below specific for the type of table reference.
2000-07-31 21:29:14 +02:00
******************************************************************************/
/* Special Field pointers as return values of find_field_in_XXX functions. */
Field *not_found_field= (Field*) 0x1;
Field *view_ref_found= (Field*) 0x2;
2004-07-16 00:15:55 +02:00
2000-07-31 21:29:14 +02:00
#define WRONG_GRANT (Field*) -1
static void update_field_dependencies(THD *thd, Field *field, TABLE *table)
{
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
DBUG_ENTER("update_field_dependencies");
if (thd->mark_used_columns != MARK_COLUMNS_NONE)
{
MY_BITMAP *bitmap;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
/*
We always want to register the used keys, as the column bitmap may have
been set for all fields (for example for view).
*/
table->covering_keys.intersect(field->part_of_key);
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
table->merge_keys.merge(field->part_of_key);
if (thd->mark_used_columns == MARK_COLUMNS_READ)
bitmap= table->read_set;
else
bitmap= table->write_set;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
/*
The test-and-set mechanism in the bitmap is not reliable during
multi-UPDATE statements under MARK_COLUMNS_READ mode
(thd->mark_used_columns == MARK_COLUMNS_READ), as this bitmap contains
only those columns that are used in the SET clause. I.e they are being
set here. See multi_update::prepare()
*/
if (bitmap_fast_test_and_set(bitmap, field->field_index))
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
{
if (thd->mark_used_columns == MARK_COLUMNS_WRITE)
{
DBUG_PRINT("warning", ("Found duplicated field"));
thd->dup_field= field;
}
else
{
DBUG_PRINT("note", ("Field found before"));
}
DBUG_VOID_RETURN;
}
if (table->get_fields_in_item_tree)
field->flags|= GET_FIXED_FIELDS_FLAG;
table->used_fields++;
}
else if (table->get_fields_in_item_tree)
2005-08-25 19:50:11 +02:00
field->flags|= GET_FIXED_FIELDS_FLAG;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
DBUG_VOID_RETURN;
}
2004-07-16 00:15:55 +02:00
/*
Find a field by name in a view that uses merge algorithm.
2004-07-16 00:15:55 +02:00
SYNOPSIS
find_field_in_view()
2004-07-16 00:15:55 +02:00
thd thread handler
table_list view to search for 'name'
2004-07-16 00:15:55 +02:00
name name of field
length length of name
item_name name of item if it will be created (VIEW)
ref expression substituted in VIEW should be passed
using this reference (return view_ref_found)
2004-11-21 18:33:49 +01:00
register_tree_change TRUE if ref is not stack variable and we
need register changes in item tree
2004-07-16 00:15:55 +02:00
RETURN
0 field is not found
view_ref_found found value in VIEW (real result is in *ref)
# pointer to field - only for schema table fields
2004-07-16 00:15:55 +02:00
*/
static Field *
find_field_in_view(THD *thd, TABLE_LIST *table_list,
const char *name, uint length,
const char *item_name, Item **ref,
bool register_tree_change)
2004-07-16 00:15:55 +02:00
{
DBUG_ENTER("find_field_in_view");
DBUG_PRINT("enter",
("view: '%s', field name: '%s', item name: '%s', ref 0x%lx",
table_list->alias, name, item_name, (ulong) ref));
Field_iterator_view field_it;
field_it.set(table_list);
Query_arena *arena= 0, backup;
DBUG_ASSERT(table_list->schema_table_reformed ||
(ref != 0 && table_list->view != 0));
for (; !field_it.end_of_fields(); field_it.next())
2004-07-16 00:15:55 +02:00
{
if (!my_strcasecmp(system_charset_info, field_it.name(), name))
{
// in PS use own arena or data will be freed after prepare
if (register_tree_change &&
thd->stmt_arena->is_stmt_prepare_or_first_stmt_execute())
arena= thd->activate_stmt_arena_if_needed(&backup);
/*
create_item() may, or may not create a new Item, depending on
the column reference. See create_view_field() for details.
*/
Item *item= field_it.create_item(thd);
if (arena)
thd->restore_active_arena(arena, &backup);
if (!item)
DBUG_RETURN(0);
/*
*ref != NULL means that *ref contains the item that we need to
replace. If the item was aliased by the user, set the alias to
the replacing item.
We need to set alias on both ref itself and on ref real item.
*/
if (*ref && !(*ref)->is_autogenerated_name)
{
item->set_name((*ref)->name, (*ref)->name_length,
system_charset_info);
item->real_item()->set_name((*ref)->name, (*ref)->name_length,
system_charset_info);
}
if (register_tree_change)
thd->change_item_tree(ref, item);
else
*ref= item;
DBUG_RETURN((Field*) view_ref_found);
}
}
DBUG_RETURN(0);
}
/*
Find field by name in a NATURAL/USING join table reference.
SYNOPSIS
find_field_in_natural_join()
thd [in] thread handler
table_ref [in] table reference to search
name [in] name of field
length [in] length of name
ref [in/out] if 'name' is resolved to a view field, ref is
set to point to the found view field
register_tree_change [in] TRUE if ref is not stack variable and we
need register changes in item tree
actual_table [out] the original table reference where the field
belongs - differs from 'table_list' only for
NATURAL/USING joins
DESCRIPTION
Search for a field among the result fields of a NATURAL/USING join.
Notice that this procedure is called only for non-qualified field
names. In the case of qualified fields, we search directly the base
tables of a natural join.
RETURN
NULL if the field was not found
WRONG_GRANT if no access rights to the found field
# Pointer to the found Field
*/
static Field *
find_field_in_natural_join(THD *thd, TABLE_LIST *table_ref, const char *name,
uint length, Item **ref, bool register_tree_change,
TABLE_LIST **actual_table)
{
List_iterator_fast<Natural_join_column>
field_it(*(table_ref->join_columns));
Natural_join_column *nj_col, *curr_nj_col;
Field *found_field;
Query_arena *arena, backup;
DBUG_ENTER("find_field_in_natural_join");
DBUG_PRINT("enter", ("field name: '%s', ref 0x%lx",
name, (ulong) ref));
DBUG_ASSERT(table_ref->is_natural_join && table_ref->join_columns);
DBUG_ASSERT(*actual_table == NULL);
LINT_INIT(arena);
LINT_INIT(found_field);
for (nj_col= NULL, curr_nj_col= field_it++; curr_nj_col;
curr_nj_col= field_it++)
{
if (!my_strcasecmp(system_charset_info, curr_nj_col->name(), name))
{
if (nj_col)
{
my_error(ER_NON_UNIQ_ERROR, MYF(0), name, thd->where);
DBUG_RETURN(NULL);
}
nj_col= curr_nj_col;
}
2004-07-16 00:15:55 +02:00
}
if (!nj_col)
DBUG_RETURN(NULL);
if (nj_col->view_field)
{
Item *item;
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
LINT_INIT(arena);
if (register_tree_change)
arena= thd->activate_stmt_arena_if_needed(&backup);
/*
create_item() may, or may not create a new Item, depending on the
column reference. See create_view_field() for details.
*/
item= nj_col->create_item(thd);
/*
*ref != NULL means that *ref contains the item that we need to
replace. If the item was aliased by the user, set the alias to
the replacing item.
We need to set alias on both ref itself and on ref real item.
*/
if (*ref && !(*ref)->is_autogenerated_name)
{
item->set_name((*ref)->name, (*ref)->name_length,
system_charset_info);
item->real_item()->set_name((*ref)->name, (*ref)->name_length,
system_charset_info);
}
if (register_tree_change && arena)
thd->restore_active_arena(arena, &backup);
if (!item)
DBUG_RETURN(NULL);
DBUG_ASSERT(nj_col->table_field == NULL);
if (nj_col->table_ref->schema_table_reformed)
{
/*
Translation table items are always Item_fields and fixed
already('mysql_schema_table' function). So we can return
->field. It is used only for 'show & where' commands.
*/
DBUG_RETURN(((Item_field*) (nj_col->view_field->item))->field);
}
if (register_tree_change)
thd->change_item_tree(ref, item);
else
*ref= item;
found_field= (Field*) view_ref_found;
}
else
{
/* This is a base table. */
DBUG_ASSERT(nj_col->view_field == NULL);
/*
This fix_fields is not necessary (initially this item is fixed by
the Item_field constructor; after reopen_tables the Item_func_eq
calls fix_fields on that item), it's just a check during table
reopening for columns that was dropped by the concurrent connection.
*/
if (!nj_col->table_field->fixed &&
nj_col->table_field->fix_fields(thd, (Item **)&nj_col->table_field))
{
DBUG_PRINT("info", ("column '%s' was dropped by the concurrent connection",
nj_col->table_field->name));
DBUG_RETURN(NULL);
}
DBUG_ASSERT(nj_col->table_ref->table == nj_col->table_field->field->table);
found_field= nj_col->table_field->field;
update_field_dependencies(thd, found_field, nj_col->table_ref->table);
}
*actual_table= nj_col->table_ref;
DBUG_RETURN(found_field);
2004-07-16 00:15:55 +02:00
}
2004-07-16 00:15:55 +02:00
/*
Find field by name in a base table or a view with temp table algorithm.
2004-07-16 00:15:55 +02:00
The caller is expected to check column-level privileges.
2004-07-16 00:15:55 +02:00
SYNOPSIS
find_field_in_table()
2004-07-16 00:15:55 +02:00
thd thread handler
table table where to search for the field
2004-07-16 00:15:55 +02:00
name name of field
length length of name
allow_rowid do allow finding of "_rowid" field?
cached_field_index_ptr cached position in field list (used to speedup
lookup for fields in prepared tables)
2004-07-16 00:15:55 +02:00
RETURN
0 field is not found
# pointer to field
2004-07-16 00:15:55 +02:00
*/
Field *
find_field_in_table(THD *thd, TABLE *table, const char *name, uint length,
bool allow_rowid, uint *cached_field_index_ptr)
2000-07-31 21:29:14 +02:00
{
2004-06-25 15:52:01 +02:00
Field **field_ptr, *field;
uint cached_field_index= *cached_field_index_ptr;
DBUG_ENTER("find_field_in_table");
DBUG_PRINT("enter", ("table: '%s', field name: '%s'", table->alias, name));
2004-06-25 15:52:01 +02:00
/* We assume here that table->field < NO_CACHED_FIELD_INDEX = UINT_MAX */
if (cached_field_index < table->s->fields &&
!my_strcasecmp(system_charset_info,
2004-06-25 15:52:01 +02:00
table->field[cached_field_index]->field_name, name))
field_ptr= table->field + cached_field_index;
else if (table->s->name_hash.records)
{
field_ptr= (Field**) my_hash_search(&table->s->name_hash, (uchar*) name,
length);
if (field_ptr)
{
/*
field_ptr points to field in TABLE_SHARE. Convert it to the matching
field in table
*/
field_ptr= (table->field + (field_ptr - table->s->field));
}
}
2004-06-25 15:52:01 +02:00
else
{
if (!(field_ptr= table->field))
DBUG_RETURN((Field *)0);
2004-06-25 15:52:01 +02:00
for (; *field_ptr; ++field_ptr)
if (!my_strcasecmp(system_charset_info, (*field_ptr)->field_name, name))
break;
}
if (field_ptr && *field_ptr)
2000-07-31 21:29:14 +02:00
{
2004-06-25 15:52:01 +02:00
*cached_field_index_ptr= field_ptr - table->field;
field= *field_ptr;
2000-07-31 21:29:14 +02:00
}
else
{
2004-06-25 15:52:01 +02:00
if (!allow_rowid ||
my_strcasecmp(system_charset_info, name, "_rowid") ||
table->s->rowid_field_offset == 0)
DBUG_RETURN((Field*) 0);
field= table->field[table->s->rowid_field_offset-1];
2000-07-31 21:29:14 +02:00
}
update_field_dependencies(thd, field, table);
DBUG_RETURN(field);
}
/*
Find field in a table reference.
SYNOPSIS
find_field_in_table_ref()
thd [in] thread handler
table_list [in] table reference to search
name [in] name of field
length [in] field length of name
item_name [in] name of item if it will be created (VIEW)
db_name [in] optional database name that qualifies the
table_name [in] optional table name that qualifies the field
ref [in/out] if 'name' is resolved to a view field, ref
is set to point to the found view field
check_privileges [in] check privileges
allow_rowid [in] do allow finding of "_rowid" field?
cached_field_index_ptr [in] cached position in field list (used to
speedup lookup for fields in prepared tables)
register_tree_change [in] TRUE if ref is not stack variable and we
need register changes in item tree
actual_table [out] the original table reference where the field
belongs - differs from 'table_list' only for
NATURAL_USING joins.
DESCRIPTION
Find a field in a table reference depending on the type of table
reference. There are three types of table references with respect
to the representation of their result columns:
- an array of Field_translator objects for MERGE views and some
information_schema tables,
- an array of Field objects (and possibly a name hash) for stored
tables,
- a list of Natural_join_column objects for NATURAL/USING joins.
This procedure detects the type of the table reference 'table_list'
and calls the corresponding search routine.
The routine checks column-level privieleges for the found field.
RETURN
0 field is not found
view_ref_found found value in VIEW (real result is in *ref)
# pointer to field
*/
Field *
find_field_in_table_ref(THD *thd, TABLE_LIST *table_list,
const char *name, uint length,
const char *item_name, const char *db_name,
const char *table_name, Item **ref,
bool check_privileges, bool allow_rowid,
uint *cached_field_index_ptr,
bool register_tree_change, TABLE_LIST **actual_table)
{
Field *fld;
DBUG_ENTER("find_field_in_table_ref");
Fixed compiler warnings Fixed compile-pentium64 scripts Fixed wrong estimate of update_with_key_prefix in sql-bench Merge bk-internal.mysql.com:/home/bk/mysql-5.1 into mysql.com:/home/my/mysql-5.1 Fixed unsafe define of uint4korr() Fixed that --extern works with mysql-test-run.pl Small trivial cleanups This also fixes a bug in counting number of rows that are updated when we have many simultanous queries Move all connection handling and command exectuion main loop from sql_parse.cc to sql_connection.cc Split handle_one_connection() into reusable sub functions. Split create_new_thread() into reusable sub functions. Added thread_scheduler; Preliminary interface code for future thread_handling code. Use 'my_thread_id' for internal thread id's Make thr_alarm_kill() to depend on thread_id instead of thread Make thr_abort_locks_for_thread() depend on thread_id instead of thread In store_globals(), set my_thread_var->id to be thd->thread_id. Use my_thread_var->id as basis for my_thread_name() The above changes makes the connection we have between THD and threads more soft. Added a lot of DBUG_PRINT() and DBUG_ASSERT() functions Fixed compiler warnings Fixed core dumps when running with --debug Removed setting of signal masks (was never used) Made event code call pthread_exit() (portability fix) Fixed that event code doesn't call DBUG_xxx functions before my_thread_init() is called. Made handling of thread_id and thd->variables.pseudo_thread_id uniform. Removed one common 'not freed memory' warning from mysqltest Fixed a couple of usage of not initialized warnings (unlikely cases) Suppress compiler warnings from bdb and (for the moment) warnings from ndb
2007-02-23 12:13:55 +01:00
DBUG_ASSERT(table_list->alias);
DBUG_ASSERT(name);
DBUG_ASSERT(item_name);
DBUG_PRINT("enter",
("table: '%s' field name: '%s' item name: '%s' ref 0x%lx",
table_list->alias, name, item_name, (ulong) ref));
/*
Check that the table and database that qualify the current field name
are the same as the table reference we are going to search for the field.
2005-09-30 09:37:58 +02:00
Exclude from the test below nested joins because the columns in a
nested join generally originate from different tables. Nested joins
also have no table name, except when a nested join is a merge view
or an information schema table.
We include explicitly table references with a 'field_translation' table,
because if there are views over natural joins we don't want to search
inside the view, but we want to search directly in the view columns
which are represented as a 'field_translation'.
TODO: Ensure that table_name, db_name and tables->db always points to
something !
*/
if (/* Exclude nested joins. */
(!table_list->nested_join ||
/* Include merge views and information schema tables. */
table_list->field_translation) &&
/*
Test if the field qualifiers match the table reference we plan
to search.
*/
table_name && table_name[0] &&
(my_strcasecmp(table_alias_charset, table_list->alias, table_name) ||
(db_name && db_name[0] && table_list->db && table_list->db[0] &&
(table_list->schema_table ?
my_strcasecmp(system_charset_info, db_name, table_list->db) :
strcmp(db_name, table_list->db)))))
DBUG_RETURN(0);
*actual_table= NULL;
if (table_list->field_translation)
{
/* 'table_list' is a view or an information schema table. */
if ((fld= find_field_in_view(thd, table_list, name, length, item_name, ref,
register_tree_change)))
*actual_table= table_list;
}
else if (!table_list->nested_join)
{
/* 'table_list' is a stored table. */
DBUG_ASSERT(table_list->table);
if ((fld= find_field_in_table(thd, table_list->table, name, length,
allow_rowid,
cached_field_index_ptr)))
*actual_table= table_list;
}
else
{
2005-09-08 10:29:52 +02:00
/*
'table_list' is a NATURAL/USING join, or an operand of such join that
is a nested join itself.
If the field name we search for is qualified, then search for the field
in the table references used by NATURAL/USING the join.
2005-09-08 10:29:52 +02:00
*/
if (table_name && table_name[0])
{
List_iterator<TABLE_LIST> it(table_list->nested_join->join_list);
TABLE_LIST *table;
while ((table= it++))
{
if ((fld= find_field_in_table_ref(thd, table, name, length, item_name,
db_name, table_name, ref,
check_privileges, allow_rowid,
cached_field_index_ptr,
register_tree_change, actual_table)))
DBUG_RETURN(fld);
}
DBUG_RETURN(0);
}
/*
Non-qualified field, search directly in the result columns of the
2005-09-08 10:29:52 +02:00
natural join. The condition of the outer IF is true for the top-most
natural join, thus if the field is not qualified, we will search
directly the top-most NATURAL/USING join.
*/
fld= find_field_in_natural_join(thd, table_list, name, length, ref,
register_tree_change, actual_table);
}
if (fld)
{
#ifndef NO_EMBEDDED_ACCESS_CHECKS
/* Check if there are sufficient access rights to the found field. */
if (check_privileges &&
check_column_grant_in_table_ref(thd, *actual_table, name, length))
fld= WRONG_GRANT;
else
#endif
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (thd->mark_used_columns != MARK_COLUMNS_NONE)
{
/*
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
Get rw_set correct for this field so that the handler
knows that this field is involved in the query and gets
retrieved/updated
*/
Field *field_to_set= NULL;
if (fld == view_ref_found)
{
Item *it= (*ref)->real_item();
if (it->type() == Item::FIELD_ITEM)
field_to_set= ((Item_field*)it)->field;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
else
{
if (thd->mark_used_columns == MARK_COLUMNS_READ)
WL#3817: Simplify string / memory area types and make things more consistent (first part) The following type conversions was done: - Changed byte to uchar - Changed gptr to uchar* - Change my_string to char * - Change my_size_t to size_t - Change size_s to size_t Removed declaration of byte, gptr, my_string, my_size_t and size_s. Following function parameter changes was done: - All string functions in mysys/strings was changed to use size_t instead of uint for string lengths. - All read()/write() functions changed to use size_t (including vio). - All protocoll functions changed to use size_t instead of uint - Functions that used a pointer to a string length was changed to use size_t* - Changed malloc(), free() and related functions from using gptr to use void * as this requires fewer casts in the code and is more in line with how the standard functions work. - Added extra length argument to dirname_part() to return the length of the created string. - Changed (at least) following functions to take uchar* as argument: - db_dump() - my_net_write() - net_write_command() - net_store_data() - DBUG_DUMP() - decimal2bin() & bin2decimal() - Changed my_compress() and my_uncompress() to use size_t. Changed one argument to my_uncompress() from a pointer to a value as we only return one value (makes function easier to use). - Changed type of 'pack_data' argument to packfrm() to avoid casts. - Changed in readfrm() and writefrom(), ha_discover and handler::discover() the type for argument 'frmdata' to uchar** to avoid casts. - Changed most Field functions to use uchar* instead of char* (reduced a lot of casts). - Changed field->val_xxx(xxx, new_ptr) to take const pointers. Other changes: - Removed a lot of not needed casts - Added a few new cast required by other changes - Added some cast to my_multi_malloc() arguments for safety (as string lengths needs to be uint, not size_t). - Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done explicitely as this conflict was often hided by casting the function to hash_get_key). - Changed some buffers to memory regions to uchar* to avoid casts. - Changed some string lengths from uint to size_t. - Changed field->ptr to be uchar* instead of char*. This allowed us to get rid of a lot of casts. - Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar - Include zlib.h in some files as we needed declaration of crc32() - Changed MY_FILE_ERROR to be (size_t) -1. - Changed many variables to hold the result of my_read() / my_write() to be size_t. This was needed to properly detect errors (which are returned as (size_t) -1). - Removed some very old VMS code - Changed packfrm()/unpackfrm() to not be depending on uint size (portability fix) - Removed windows specific code to restore cursor position as this causes slowdown on windows and we should not mix read() and pread() calls anyway as this is not thread safe. Updated function comment to reflect this. Changed function that depended on original behavior of my_pwrite() to itself restore the cursor position (one such case). - Added some missing checking of return value of malloc(). - Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow. - Changed type of table_def::m_size from my_size_t to ulong to reflect that m_size is the number of elements in the array, not a string/memory length. - Moved THD::max_row_length() to table.cc (as it's not depending on THD). Inlined max_row_length_blob() into this function. - More function comments - Fixed some compiler warnings when compiled without partitions. - Removed setting of LEX_STRING() arguments in declaration (portability fix). - Some trivial indentation/variable name changes. - Some trivial code simplifications: - Replaced some calls to alloc_root + memcpy to use strmake_root()/strdup_root(). - Changed some calls from memdup() to strmake() (Safety fix) - Simpler loops in client-simple.c
2007-05-10 11:59:39 +02:00
it->walk(&Item::register_field_in_read_map, 1, (uchar *) 0);
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
}
}
else
field_to_set= fld;
if (field_to_set)
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
{
TABLE *table= field_to_set->table;
if (thd->mark_used_columns == MARK_COLUMNS_READ)
bitmap_set_bit(table->read_set, field_to_set->field_index);
else
bitmap_set_bit(table->write_set, field_to_set->field_index);
}
}
}
DBUG_RETURN(fld);
2000-07-31 21:29:14 +02:00
}
2005-08-25 19:50:11 +02:00
/*
Find field in table, no side effects, only purpose is to check for field
in table object and get reference to the field if found.
SYNOPSIS
find_field_in_table_sef()
table table where to find
name Name of field searched for
RETURN
0 field is not found
# pointer to field
*/
Field *find_field_in_table_sef(TABLE *table, const char *name)
{
Field **field_ptr;
if (table->s->name_hash.records)
{
field_ptr= (Field**)my_hash_search(&table->s->name_hash,(uchar*) name,
strlen(name));
if (field_ptr)
{
/*
field_ptr points to field in TABLE_SHARE. Convert it to the matching
field in table
*/
field_ptr= (table->field + (field_ptr - table->s->field));
}
}
2005-08-25 19:50:11 +02:00
else
{
if (!(field_ptr= table->field))
return (Field *)0;
for (; *field_ptr; ++field_ptr)
if (!my_strcasecmp(system_charset_info, (*field_ptr)->field_name, name))
break;
}
if (field_ptr)
return *field_ptr;
else
return (Field *)0;
}
/*
Find field in table list.
SYNOPSIS
find_field_in_tables()
thd pointer to current thread structure
item field item that should be found
first_table list of tables to be searched for item
last_table end of the list of tables to search for item. If NULL
then search to the end of the list 'first_table'.
ref if 'item' is resolved to a view field, ref is set to
point to the found view field
report_error Degree of error reporting:
- IGNORE_ERRORS then do not report any error
- IGNORE_EXCEPT_NON_UNIQUE report only non-unique
fields, suppress all other errors
- REPORT_EXCEPT_NON_UNIQUE report all other errors
except when non-unique fields were found
- REPORT_ALL_ERRORS
check_privileges need to check privileges
register_tree_change TRUE if ref is not a stack variable and we
to need register changes in item tree
RETURN VALUES
0 If error: the found field is not unique, or there are
no sufficient access priviliges for the found field,
or the field is qualified with non-existing table.
not_found_field The function was called with report_error ==
(IGNORE_ERRORS || IGNORE_EXCEPT_NON_UNIQUE) and a
field was not found.
view_ref_found View field is found, item passed through ref parameter
found field If a item was resolved to some field
*/
2000-07-31 21:29:14 +02:00
Field *
find_field_in_tables(THD *thd, Item_ident *item,
TABLE_LIST *first_table, TABLE_LIST *last_table,
Item **ref, find_item_error_report_type report_error,
bool check_privileges, bool register_tree_change)
2000-07-31 21:29:14 +02:00
{
Field *found=0;
const char *db= item->db_name;
const char *table_name= item->table_name;
const char *name= item->field_name;
uint length=(uint) strlen(name);
char name_buff[NAME_LEN+1];
TABLE_LIST *cur_table= first_table;
TABLE_LIST *actual_table;
bool allow_rowid;
if (!table_name || !table_name[0])
{
table_name= 0; // For easier test
db= 0;
}
allow_rowid= table_name || (cur_table && !cur_table->next_local);
2004-06-25 15:52:01 +02:00
if (item->cached_table)
{
/*
This shortcut is used by prepared statements. We assume that
TABLE_LIST *first_table is not changed during query execution (which
is true for all queries except RENAME but luckily RENAME doesn't
2004-06-25 15:52:01 +02:00
use fields...) so we can rely on reusing pointer to its member.
2004-11-02 19:11:00 +01:00
With this optimization we also miss case when addition of one more
field makes some prepared query ambiguous and so erroneous, but we
2004-06-25 15:52:01 +02:00
accept this trade off.
*/
TABLE_LIST *table_ref= item->cached_table;
/*
The condition (table_ref->view == NULL) ensures that we will call
find_field_in_table even in the case of information schema tables
when table_ref->field_translation != NULL.
*/
if (table_ref->table && !table_ref->view)
{
found= find_field_in_table(thd, table_ref->table, name, length,
TRUE, &(item->cached_field_index));
#ifndef NO_EMBEDDED_ACCESS_CHECKS
/* Check if there are sufficient access rights to the found field. */
if (found && check_privileges &&
check_column_grant_in_table_ref(thd, table_ref, name, length))
found= WRONG_GRANT;
#endif
}
else
found= find_field_in_table_ref(thd, table_ref, name, length, item->name,
NULL, NULL, ref, check_privileges,
TRUE, &(item->cached_field_index),
register_tree_change,
&actual_table);
2004-06-25 15:52:01 +02:00
if (found)
{
if (found == WRONG_GRANT)
return (Field*) 0;
/*
Only views fields should be marked as dependent, not an underlying
fields.
*/
if (!table_ref->belong_to_view)
{
SELECT_LEX *current_sel= thd->lex->current_select;
SELECT_LEX *last_select= table_ref->select_lex;
2005-04-30 18:27:22 +02:00
/*
If the field was an outer referencee, mark all selects using this
sub query as dependent on the outer query
2005-04-30 18:27:22 +02:00
*/
if (current_sel != last_select)
2005-04-30 18:27:22 +02:00
mark_select_range_as_dependent(thd, last_select, current_sel,
found, *ref, item);
}
2004-06-25 15:52:01 +02:00
return found;
}
}
if (db && lower_case_table_names)
{
/*
2004-11-02 19:11:00 +01:00
convert database to lower case for comparison.
We can't do this in Item_field as this would change the
'name' of the item which may be used in the select list
*/
strmake(name_buff, db, sizeof(name_buff)-1);
my_casedn_str(files_charset_info, name_buff);
db= name_buff;
}
2000-07-31 21:29:14 +02:00
if (last_table)
last_table= last_table->next_name_resolution_table;
for (; cur_table != last_table ;
cur_table= cur_table->next_name_resolution_table)
2000-07-31 21:29:14 +02:00
{
Field *cur_field= find_field_in_table_ref(thd, cur_table, name, length,
item->name, db, table_name, ref,
(thd->lex->sql_command ==
SQLCOM_SHOW_FIELDS)
? false : check_privileges,
allow_rowid,
&(item->cached_field_index),
register_tree_change,
&actual_table);
if (cur_field)
{
if (cur_field == WRONG_GRANT)
{
if (thd->lex->sql_command != SQLCOM_SHOW_FIELDS)
return (Field*) 0;
thd->clear_error();
cur_field= find_field_in_table_ref(thd, cur_table, name, length,
item->name, db, table_name, ref,
false,
allow_rowid,
&(item->cached_field_index),
register_tree_change,
&actual_table);
if (cur_field)
{
Field *nf=new Field_null(NULL,0,Field::NONE,
cur_field->field_name,
&my_charset_bin);
nf->init(cur_table->table);
cur_field= nf;
}
}
/*
Store the original table of the field, which may be different from
cur_table in the case of NATURAL/USING join.
*/
item->cached_table= (!actual_table->cacheable_table || found) ?
0 : actual_table;
DBUG_ASSERT(thd->where);
/*
If we found a fully qualified field we return it directly as it can't
have duplicates.
*/
if (db)
return cur_field;
2000-07-31 21:29:14 +02:00
if (found)
{
if (report_error == REPORT_ALL_ERRORS ||
report_error == IGNORE_EXCEPT_NON_UNIQUE)
my_error(ER_NON_UNIQ_ERROR, MYF(0),
table_name ? item->full_name() : name, thd->where);
return (Field*) 0;
2000-07-31 21:29:14 +02:00
}
found= cur_field;
2000-07-31 21:29:14 +02:00
}
}
2000-07-31 21:29:14 +02:00
if (found)
return found;
/*
If the field was qualified and there were no tables to search, issue
an error that an unknown table was given. The situation is detected
as follows: if there were no tables we wouldn't go through the loop
and cur_table wouldn't be updated by the loop increment part, so it
will be equal to the first table.
*/
if (table_name && (cur_table == first_table) &&
(report_error == REPORT_ALL_ERRORS ||
report_error == REPORT_EXCEPT_NON_UNIQUE))
{
char buff[NAME_LEN*2 + 2];
if (db && db[0])
{
strxnmov(buff,sizeof(buff)-1,db,".",table_name,NullS);
table_name=buff;
}
my_error(ER_UNKNOWN_TABLE, MYF(0), table_name, thd->where);
}
else
{
if (report_error == REPORT_ALL_ERRORS ||
report_error == REPORT_EXCEPT_NON_UNIQUE)
my_error(ER_BAD_FIELD_ERROR, MYF(0), item->full_name(), thd->where);
else
found= not_found_field;
}
return found;
2000-07-31 21:29:14 +02:00
}
/*
Find Item in list of items (find_field_in_tables analog)
TODO
is it better return only counter?
SYNOPSIS
find_item_in_list()
find Item to find
items List of items
counter To return number of found item
report_error
REPORT_ALL_ERRORS report errors, return 0 if error
REPORT_EXCEPT_NOT_FOUND Do not report 'not found' error and
return not_found_item, report other errors,
return 0
IGNORE_ERRORS Do not report errors, return 0 if error
resolution Set to the resolution type if the item is found
(it says whether the item is resolved
against an alias name,
or as a field name without alias,
or as a field hidden by alias,
or ignoring alias)
RETURN VALUES
0 Item is not found or item is not unique,
error message is reported
not_found_item Function was called with
report_error == REPORT_EXCEPT_NOT_FOUND and
item was not found. No error message was reported
found field
*/
/* Special Item pointer to serve as a return value from find_item_in_list(). */
Item **not_found_item= (Item**) 0x1;
2000-07-31 21:29:14 +02:00
Item **
find_item_in_list(Item *find, List<Item> &items, uint *counter,
find_item_error_report_type report_error,
enum_resolution_type *resolution)
2000-07-31 21:29:14 +02:00
{
List_iterator<Item> li(items);
Item **found=0, **found_unaliased= 0, *item;
const char *db_name=0;
2000-07-31 21:29:14 +02:00
const char *field_name=0;
const char *table_name=0;
bool found_unaliased_non_uniq= 0;
/*
true if the item that we search for is a valid name reference
(and not an item that happens to have a name).
*/
bool is_ref_by_name= 0;
2009-06-17 16:56:44 +02:00
uint unaliased_counter= 0;
*resolution= NOT_RESOLVED;
is_ref_by_name= (find->type() == Item::FIELD_ITEM ||
find->type() == Item::REF_ITEM);
if (is_ref_by_name)
2000-07-31 21:29:14 +02:00
{
field_name= ((Item_ident*) find)->field_name;
table_name= ((Item_ident*) find)->table_name;
db_name= ((Item_ident*) find)->db_name;
2000-07-31 21:29:14 +02:00
}
for (uint i= 0; (item=li++); i++)
2000-07-31 21:29:14 +02:00
{
if (field_name && item->real_item()->type() == Item::FIELD_ITEM)
2000-07-31 21:29:14 +02:00
{
Item_ident *item_field= (Item_ident*) item;
2004-06-25 15:52:01 +02:00
/*
In case of group_concat() with ORDER BY condition in the QUERY
item_field can be field of temporary table without item name
(if this field created from expression argument of group_concat()),
=> we have to check presence of name before compare
*/
if (!item_field->name)
continue;
if (table_name)
2000-07-31 21:29:14 +02:00
{
/*
If table name is specified we should find field 'field_name' in
table 'table_name'. According to SQL-standard we should ignore
aliases in this case.
Since we should NOT prefer fields from the select list over
other fields from the tables participating in this select in
case of ambiguity we have to do extra check outside this function.
We use strcmp for table names and database names as these may be
case sensitive. In cases where they are not case sensitive, they
are always in lower case.
item_field->field_name and item_field->table_name can be 0x0 if
item is not fix_field()'ed yet.
*/
if (item_field->field_name && item_field->table_name &&
!my_strcasecmp(system_charset_info, item_field->field_name,
field_name) &&
!my_strcasecmp(table_alias_charset, item_field->table_name,
table_name) &&
(!db_name || (item_field->db_name &&
!strcmp(item_field->db_name, db_name))))
{
if (found_unaliased)
{
if ((*found_unaliased)->eq(item, 0))
continue;
/*
Two matching fields in select list.
We already can bail out because we are searching through
unaliased names only and will have duplicate error anyway.
*/
if (report_error != IGNORE_ERRORS)
my_error(ER_NON_UNIQ_ERROR, MYF(0),
find->full_name(), current_thd->where);
return (Item**) 0;
}
found_unaliased= li.ref();
unaliased_counter= i;
*resolution= RESOLVED_IGNORING_ALIAS;
if (db_name)
break; // Perfect match
}
}
else
{
int fname_cmp= my_strcasecmp(system_charset_info,
item_field->field_name,
field_name);
if (!my_strcasecmp(system_charset_info,
item_field->name,field_name))
{
/*
If table name was not given we should scan through aliases
and non-aliased fields first. We are also checking unaliased
name of the field in then next else-if, to be able to find
instantly field (hidden by alias) if no suitable alias or
non-aliased field was found.
*/
if (found)
{
if ((*found)->eq(item, 0))
continue; // Same field twice
if (report_error != IGNORE_ERRORS)
my_error(ER_NON_UNIQ_ERROR, MYF(0),
find->full_name(), current_thd->where);
return (Item**) 0;
}
found= li.ref();
*counter= i;
*resolution= fname_cmp ? RESOLVED_AGAINST_ALIAS:
RESOLVED_WITH_NO_ALIAS;
}
else if (!fname_cmp)
{
/*
We will use non-aliased field or react on such ambiguities only if
we won't be able to find aliased field.
Again if we have ambiguity with field outside of select list
we should prefer fields from select list.
*/
if (found_unaliased)
{
if ((*found_unaliased)->eq(item, 0))
continue; // Same field twice
found_unaliased_non_uniq= 1;
}
found_unaliased= li.ref();
unaliased_counter= i;
}
2000-07-31 21:29:14 +02:00
}
}
else if (!table_name)
{
if (is_ref_by_name && find->name && item->name &&
!my_strcasecmp(system_charset_info,item->name,find->name))
{
found= li.ref();
*counter= i;
*resolution= RESOLVED_AGAINST_ALIAS;
break;
}
else if (find->eq(item,0))
{
found= li.ref();
*counter= i;
*resolution= RESOLVED_IGNORING_ALIAS;
break;
}
}
else if (table_name && item->type() == Item::REF_ITEM &&
((Item_ref *)item)->ref_type() == Item_ref::VIEW_REF)
{
/*
TODO:Here we process prefixed view references only. What we should
really do is process all types of Item_refs. But this will currently
lead to a clash with the way references to outer SELECTs (from the
HAVING clause) are handled in e.g. :
SELECT 1 FROM t1 AS t1_o GROUP BY a
HAVING (SELECT t1_o.a FROM t1 AS t1_i GROUP BY t1_i.a LIMIT 1).
Processing all Item_refs here will cause t1_o.a to resolve to itself.
We still need to process the special case of Item_direct_view_ref
because in the context of views they have the same meaning as
Item_field for tables.
*/
Item_ident *item_ref= (Item_ident *) item;
if (item_ref->name && item_ref->table_name &&
!my_strcasecmp(system_charset_info, item_ref->name, field_name) &&
!my_strcasecmp(table_alias_charset, item_ref->table_name,
table_name) &&
(!db_name || (item_ref->db_name &&
!strcmp (item_ref->db_name, db_name))))
{
found= li.ref();
*counter= i;
*resolution= RESOLVED_IGNORING_ALIAS;
break;
}
}
2000-07-31 21:29:14 +02:00
}
if (!found)
{
if (found_unaliased_non_uniq)
{
if (report_error != IGNORE_ERRORS)
my_error(ER_NON_UNIQ_ERROR, MYF(0),
find->full_name(), current_thd->where);
return (Item **) 0;
}
if (found_unaliased)
{
found= found_unaliased;
*counter= unaliased_counter;
*resolution= RESOLVED_BEHIND_ALIAS;
}
}
if (found)
return found;
if (report_error != REPORT_EXCEPT_NOT_FOUND)
{
if (report_error == REPORT_ALL_ERRORS)
my_error(ER_BAD_FIELD_ERROR, MYF(0),
find->full_name(), current_thd->where);
return (Item **) 0;
}
else
return (Item **) not_found_item;
2000-07-31 21:29:14 +02:00
}
/*
Test if a string is a member of a list of strings.
SYNOPSIS
test_if_string_in_list()
find the string to look for
str_list a list of strings to be searched
DESCRIPTION
Sequentially search a list of strings for a string, and test whether
the list contains the same string.
RETURN
TRUE if find is in str_list
FALSE otherwise
*/
static bool
test_if_string_in_list(const char *find, List<String> *str_list)
{
List_iterator<String> str_list_it(*str_list);
String *curr_str;
size_t find_length= strlen(find);
while ((curr_str= str_list_it++))
{
if (find_length != curr_str->length())
continue;
if (!my_strcasecmp(system_charset_info, find, curr_str->ptr()))
return TRUE;
}
return FALSE;
}
/*
Create a new name resolution context for an item so that it is
being resolved in a specific table reference.
SYNOPSIS
set_new_item_local_context()
thd pointer to current thread
item item for which new context is created and set
table_ref table ref where an item showld be resolved
DESCRIPTION
Create a new name resolution context for an item, so that the item
is resolved only the supplied 'table_ref'.
RETURN
FALSE if all OK
TRUE otherwise
*/
static bool
set_new_item_local_context(THD *thd, Item_ident *item, TABLE_LIST *table_ref)
{
Name_resolution_context *context;
if (!(context= new (thd->mem_root) Name_resolution_context))
return TRUE;
context->init();
context->first_name_resolution_table=
context->last_name_resolution_table= table_ref;
item->context= context;
return FALSE;
}
/*
Find and mark the common columns of two table references.
SYNOPSIS
mark_common_columns()
thd [in] current thread
table_ref_1 [in] the first (left) join operand
table_ref_2 [in] the second (right) join operand
using_fields [in] if the join is JOIN...USING - the join columns,
if NATURAL join, then NULL
found_using_fields [out] number of fields from the USING clause that were
found among the common fields
DESCRIPTION
The procedure finds the common columns of two relations (either
tables or intermediate join results), and adds an equi-join condition
to the ON clause of 'table_ref_2' for each pair of matching columns.
If some of table_ref_XXX represents a base table or view, then we
create new 'Natural_join_column' instances for each column
reference and store them in the 'join_columns' of the table
reference.
IMPLEMENTATION
The procedure assumes that store_natural_using_join_columns() was
called for the previous level of NATURAL/USING joins.
RETURN
TRUE error when some common column is non-unique, or out of memory
FALSE OK
*/
static bool
mark_common_columns(THD *thd, TABLE_LIST *table_ref_1, TABLE_LIST *table_ref_2,
List<String> *using_fields, uint *found_using_fields)
{
Field_iterator_table_ref it_1, it_2;
Natural_join_column *nj_col_1, *nj_col_2;
Query_arena *arena, backup;
bool result= TRUE;
bool first_outer_loop= TRUE;
/*
Leaf table references to which new natural join columns are added
if the leaves are != NULL.
*/
TABLE_LIST *leaf_1= (table_ref_1->nested_join &&
!table_ref_1->is_natural_join) ?
NULL : table_ref_1;
TABLE_LIST *leaf_2= (table_ref_2->nested_join &&
!table_ref_2->is_natural_join) ?
NULL : table_ref_2;
DBUG_ENTER("mark_common_columns");
DBUG_PRINT("info", ("operand_1: %s operand_2: %s",
table_ref_1->alias, table_ref_2->alias));
*found_using_fields= 0;
arena= thd->activate_stmt_arena_if_needed(&backup);
for (it_1.set(table_ref_1); !it_1.end_of_fields(); it_1.next())
{
bool found= FALSE;
2006-02-23 14:21:43 +01:00
const char *field_name_1;
/* true if field_name_1 is a member of using_fields */
bool is_using_column_1;
if (!(nj_col_1= it_1.get_or_create_column_ref(thd, leaf_1)))
goto err;
field_name_1= nj_col_1->name();
is_using_column_1= using_fields &&
test_if_string_in_list(field_name_1, using_fields);
DBUG_PRINT ("info", ("field_name_1=%s.%s",
nj_col_1->table_name() ? nj_col_1->table_name() : "",
field_name_1));
/*
Find a field with the same name in table_ref_2.
Note that for the second loop, it_2.set() will iterate over
table_ref_2->join_columns and not generate any new elements or
lists.
*/
nj_col_2= NULL;
for (it_2.set(table_ref_2); !it_2.end_of_fields(); it_2.next())
{
Natural_join_column *cur_nj_col_2;
const char *cur_field_name_2;
if (!(cur_nj_col_2= it_2.get_or_create_column_ref(thd, leaf_2)))
goto err;
cur_field_name_2= cur_nj_col_2->name();
DBUG_PRINT ("info", ("cur_field_name_2=%s.%s",
cur_nj_col_2->table_name() ?
cur_nj_col_2->table_name() : "",
cur_field_name_2));
2005-09-08 10:29:52 +02:00
/*
Compare the two columns and check for duplicate common fields.
A common field is duplicate either if it was already found in
table_ref_2 (then found == TRUE), or if a field in table_ref_2
was already matched by some previous field in table_ref_1
(then cur_nj_col_2->is_common == TRUE).
Note that it is too early to check the columns outside of the
USING list for ambiguity because they are not actually "referenced"
here. These columns must be checked only on unqualified reference
by name (e.g. in SELECT list).
2005-09-08 10:29:52 +02:00
*/
if (!my_strcasecmp(system_charset_info, field_name_1, cur_field_name_2))
{
DBUG_PRINT ("info", ("match c1.is_common=%d", nj_col_1->is_common));
if (cur_nj_col_2->is_common ||
(found && (!using_fields || is_using_column_1)))
{
my_error(ER_NON_UNIQ_ERROR, MYF(0), field_name_1, thd->where);
goto err;
}
nj_col_2= cur_nj_col_2;
found= TRUE;
}
}
if (first_outer_loop && leaf_2)
{
/*
Make sure that the next inner loop "knows" that all columns
are materialized already.
*/
leaf_2->is_join_columns_complete= TRUE;
first_outer_loop= FALSE;
}
if (!found)
continue; // No matching field
/*
field_1 and field_2 have the same names. Check if they are in the USING
clause (if present), mark them as common fields, and add a new
equi-join condition to the ON clause.
*/
if (nj_col_2 && (!using_fields ||is_using_column_1))
{
Item *item_1= nj_col_1->create_item(thd);
Item *item_2= nj_col_2->create_item(thd);
Field *field_1= nj_col_1->field();
Field *field_2= nj_col_2->field();
Item_ident *item_ident_1, *item_ident_2;
Item_func_eq *eq_cond;
if (!item_1 || !item_2)
goto err; // out of memory
/*
The following assert checks that the two created items are of
type Item_ident.
*/
DBUG_ASSERT(!thd->lex->current_select->no_wrap_view_item);
/*
In the case of no_wrap_view_item == 0, the created items must be
of sub-classes of Item_ident.
*/
DBUG_ASSERT(item_1->type() == Item::FIELD_ITEM ||
item_1->type() == Item::REF_ITEM);
DBUG_ASSERT(item_2->type() == Item::FIELD_ITEM ||
item_2->type() == Item::REF_ITEM);
/*
We need to cast item_1,2 to Item_ident, because we need to hook name
resolution contexts specific to each item.
*/
item_ident_1= (Item_ident*) item_1;
item_ident_2= (Item_ident*) item_2;
/*
Create and hook special name resolution contexts to each item in the
new join condition . We need this to both speed-up subsequent name
resolution of these items, and to enable proper name resolution of
the items during the execute phase of PS.
*/
if (set_new_item_local_context(thd, item_ident_1, nj_col_1->table_ref) ||
set_new_item_local_context(thd, item_ident_2, nj_col_2->table_ref))
goto err;
if (!(eq_cond= new Item_func_eq(item_ident_1, item_ident_2)))
goto err; /* Out of memory. */
/*
Add the new equi-join condition to the ON clause. Notice that
fix_fields() is applied to all ON conditions in setup_conds()
so we don't do it here.
*/
add_join_on((table_ref_1->outer_join & JOIN_TYPE_RIGHT ?
table_ref_1 : table_ref_2),
eq_cond);
nj_col_1->is_common= nj_col_2->is_common= TRUE;
DBUG_PRINT ("info", ("%s.%s and %s.%s are common",
nj_col_1->table_name() ?
nj_col_1->table_name() : "",
nj_col_1->name(),
nj_col_2->table_name() ?
nj_col_2->table_name() : "",
nj_col_2->name()));
if (field_1)
{
2005-08-25 21:02:40 +02:00
TABLE *table_1= nj_col_1->table_ref->table;
/* Mark field_1 used for table cache. */
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
bitmap_set_bit(table_1->read_set, field_1->field_index);
table_1->covering_keys.intersect(field_1->part_of_key);
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
table_1->merge_keys.merge(field_1->part_of_key);
}
if (field_2)
{
2005-08-25 21:02:40 +02:00
TABLE *table_2= nj_col_2->table_ref->table;
/* Mark field_2 used for table cache. */
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
bitmap_set_bit(table_2->read_set, field_2->field_index);
table_2->covering_keys.intersect(field_2->part_of_key);
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
table_2->merge_keys.merge(field_2->part_of_key);
}
if (using_fields != NULL)
++(*found_using_fields);
}
}
if (leaf_1)
leaf_1->is_join_columns_complete= TRUE;
/*
Everything is OK.
Notice that at this point there may be some column names in the USING
clause that are not among the common columns. This is an SQL error and
we check for this error in store_natural_using_join_columns() when
(found_using_fields < length(join_using_fields)).
*/
result= FALSE;
err:
if (arena)
thd->restore_active_arena(arena, &backup);
DBUG_RETURN(result);
}
/*
Materialize and store the row type of NATURAL/USING join.
SYNOPSIS
store_natural_using_join_columns()
thd current thread
natural_using_join the table reference of the NATURAL/USING join
table_ref_1 the first (left) operand (of a NATURAL/USING join).
table_ref_2 the second (right) operand (of a NATURAL/USING join).
using_fields if the join is JOIN...USING - the join columns,
if NATURAL join, then NULL
found_using_fields number of fields from the USING clause that were
found among the common fields
DESCRIPTION
Iterate over the columns of both join operands and sort and store
all columns into the 'join_columns' list of natural_using_join
where the list is formed by three parts:
part1: The coalesced columns of table_ref_1 and table_ref_2,
sorted according to the column order of the first table.
part2: The other columns of the first table, in the order in
which they were defined in CREATE TABLE.
part3: The other columns of the second table, in the order in
which they were defined in CREATE TABLE.
Time complexity - O(N1+N2), where Ni = length(table_ref_i).
IMPLEMENTATION
The procedure assumes that mark_common_columns() has been called
for the join that is being processed.
RETURN
TRUE error: Some common column is ambiguous
FALSE OK
*/
static bool
store_natural_using_join_columns(THD *thd, TABLE_LIST *natural_using_join,
TABLE_LIST *table_ref_1,
TABLE_LIST *table_ref_2,
List<String> *using_fields,
uint found_using_fields)
{
Field_iterator_table_ref it_1, it_2;
Natural_join_column *nj_col_1, *nj_col_2;
Query_arena *arena, backup;
bool result= TRUE;
List<Natural_join_column> *non_join_columns;
DBUG_ENTER("store_natural_using_join_columns");
DBUG_ASSERT(!natural_using_join->join_columns);
arena= thd->activate_stmt_arena_if_needed(&backup);
if (!(non_join_columns= new List<Natural_join_column>) ||
!(natural_using_join->join_columns= new List<Natural_join_column>))
goto err;
/* Append the columns of the first join operand. */
for (it_1.set(table_ref_1); !it_1.end_of_fields(); it_1.next())
{
nj_col_1= it_1.get_natural_column_ref();
if (nj_col_1->is_common)
{
natural_using_join->join_columns->push_back(nj_col_1);
/* Reset the common columns for the next call to mark_common_columns. */
nj_col_1->is_common= FALSE;
}
else
non_join_columns->push_back(nj_col_1);
}
/*
Check that all columns in the USING clause are among the common
columns. If this is not the case, report the first one that was
not found in an error.
*/
if (using_fields && found_using_fields < using_fields->elements)
{
String *using_field_name;
List_iterator_fast<String> using_fields_it(*using_fields);
while ((using_field_name= using_fields_it++))
{
const char *using_field_name_ptr= using_field_name->c_ptr();
List_iterator_fast<Natural_join_column>
it(*(natural_using_join->join_columns));
Natural_join_column *common_field;
for (;;)
{
/* If reached the end of fields, and none was found, report error. */
if (!(common_field= it++))
{
my_error(ER_BAD_FIELD_ERROR, MYF(0), using_field_name_ptr,
current_thd->where);
goto err;
}
if (!my_strcasecmp(system_charset_info,
common_field->name(), using_field_name_ptr))
break; // Found match
}
}
}
/* Append the non-equi-join columns of the second join operand. */
for (it_2.set(table_ref_2); !it_2.end_of_fields(); it_2.next())
{
nj_col_2= it_2.get_natural_column_ref();
if (!nj_col_2->is_common)
non_join_columns->push_back(nj_col_2);
else
{
/* Reset the common columns for the next call to mark_common_columns. */
nj_col_2->is_common= FALSE;
}
}
if (non_join_columns->elements > 0)
natural_using_join->join_columns->concat(non_join_columns);
natural_using_join->is_join_columns_complete= TRUE;
result= FALSE;
err:
if (arena)
thd->restore_active_arena(arena, &backup);
DBUG_RETURN(result);
}
/*
Precompute and store the row types of the top-most NATURAL/USING joins.
SYNOPSIS
store_top_level_join_columns()
thd current thread
table_ref nested join or table in a FROM clause
left_neighbor neighbor table reference to the left of table_ref at the
same level in the join tree
right_neighbor neighbor table reference to the right of table_ref at the
same level in the join tree
DESCRIPTION
The procedure performs a post-order traversal of a nested join tree
and materializes the row types of NATURAL/USING joins in a
bottom-up manner until it reaches the TABLE_LIST elements that
represent the top-most NATURAL/USING joins. The procedure should be
applied to each element of SELECT_LEX::top_join_list (i.e. to each
top-level element of the FROM clause).
IMPLEMENTATION
Notice that the table references in the list nested_join->join_list
are in reverse order, thus when we iterate over it, we are moving
from the right to the left in the FROM clause.
RETURN
TRUE Error
FALSE OK
*/
static bool
store_top_level_join_columns(THD *thd, TABLE_LIST *table_ref,
TABLE_LIST *left_neighbor,
TABLE_LIST *right_neighbor)
{
2005-08-24 09:51:28 +02:00
Query_arena *arena, backup;
bool result= TRUE;
DBUG_ENTER("store_top_level_join_columns");
arena= thd->activate_stmt_arena_if_needed(&backup);
2005-08-24 09:51:28 +02:00
/* Call the procedure recursively for each nested table reference. */
if (table_ref->nested_join)
{
List_iterator_fast<TABLE_LIST> nested_it(table_ref->nested_join->join_list);
TABLE_LIST *same_level_left_neighbor= nested_it++;
TABLE_LIST *same_level_right_neighbor= NULL;
/* Left/right-most neighbors, possibly at higher levels in the join tree. */
TABLE_LIST *real_left_neighbor, *real_right_neighbor;
while (same_level_left_neighbor)
{
TABLE_LIST *cur_table_ref= same_level_left_neighbor;
same_level_left_neighbor= nested_it++;
/*
The order of RIGHT JOIN operands is reversed in 'join list' to
transform it into a LEFT JOIN. However, in this procedure we need
the join operands in their lexical order, so below we reverse the
join operands. Notice that this happens only in the first loop,
and not in the second one, as in the second loop
same_level_left_neighbor == NULL.
This is the correct behavior, because the second loop sets
cur_table_ref reference correctly after the join operands are
swapped in the first loop.
*/
if (same_level_left_neighbor &&
cur_table_ref->outer_join & JOIN_TYPE_RIGHT)
{
/* This can happen only for JOIN ... ON. */
DBUG_ASSERT(table_ref->nested_join->join_list.elements == 2);
swap_variables(TABLE_LIST*, same_level_left_neighbor, cur_table_ref);
}
/*
Pick the parent's left and right neighbors if there are no immediate
neighbors at the same level.
*/
real_left_neighbor= (same_level_left_neighbor) ?
same_level_left_neighbor : left_neighbor;
real_right_neighbor= (same_level_right_neighbor) ?
same_level_right_neighbor : right_neighbor;
if (cur_table_ref->nested_join &&
store_top_level_join_columns(thd, cur_table_ref,
real_left_neighbor, real_right_neighbor))
2005-08-24 09:51:28 +02:00
goto err;
same_level_right_neighbor= cur_table_ref;
}
}
/*
If this is a NATURAL/USING join, materialize its result columns and
convert to a JOIN ... ON.
*/
if (table_ref->is_natural_join)
{
DBUG_ASSERT(table_ref->nested_join &&
table_ref->nested_join->join_list.elements == 2);
List_iterator_fast<TABLE_LIST> operand_it(table_ref->nested_join->join_list);
/*
Notice that the order of join operands depends on whether table_ref
represents a LEFT or a RIGHT join. In a RIGHT join, the operands are
in inverted order.
*/
TABLE_LIST *table_ref_2= operand_it++; /* Second NATURAL join operand.*/
TABLE_LIST *table_ref_1= operand_it++; /* First NATURAL join operand. */
List<String> *using_fields= table_ref->join_using_fields;
uint found_using_fields;
/*
The two join operands were interchanged in the parser, change the order
back for 'mark_common_columns'.
*/
if (table_ref_2->outer_join & JOIN_TYPE_RIGHT)
swap_variables(TABLE_LIST*, table_ref_1, table_ref_2);
if (mark_common_columns(thd, table_ref_1, table_ref_2,
using_fields, &found_using_fields))
2005-08-24 09:51:28 +02:00
goto err;
/*
Swap the join operands back, so that we pick the columns of the second
one as the coalesced columns. In this way the coalesced columns are the
same as of an equivalent LEFT JOIN.
*/
if (table_ref_1->outer_join & JOIN_TYPE_RIGHT)
swap_variables(TABLE_LIST*, table_ref_1, table_ref_2);
if (store_natural_using_join_columns(thd, table_ref, table_ref_1,
table_ref_2, using_fields,
found_using_fields))
2005-08-24 09:51:28 +02:00
goto err;
/*
Change NATURAL JOIN to JOIN ... ON. We do this for both operands
because either one of them or the other is the one with the
natural join flag because RIGHT joins are transformed into LEFT,
and the two tables may be reordered.
*/
table_ref_1->natural_join= table_ref_2->natural_join= NULL;
2005-08-23 19:03:32 +02:00
/* Add a TRUE condition to outer joins that have no common columns. */
if (table_ref_2->outer_join &&
!table_ref_1->on_expr && !table_ref_2->on_expr)
table_ref_2->on_expr= new Item_int((longlong) 1,1); /* Always true. */
/* Change this table reference to become a leaf for name resolution. */
if (left_neighbor)
{
TABLE_LIST *last_leaf_on_the_left;
last_leaf_on_the_left= left_neighbor->last_leaf_for_name_resolution();
last_leaf_on_the_left->next_name_resolution_table= table_ref;
}
if (right_neighbor)
{
TABLE_LIST *first_leaf_on_the_right;
first_leaf_on_the_right= right_neighbor->first_leaf_for_name_resolution();
table_ref->next_name_resolution_table= first_leaf_on_the_right;
}
else
table_ref->next_name_resolution_table= NULL;
}
2005-08-24 09:51:28 +02:00
result= FALSE; /* All is OK. */
err:
if (arena)
thd->restore_active_arena(arena, &backup);
2005-08-24 09:51:28 +02:00
DBUG_RETURN(result);
}
/*
Compute and store the row types of the top-most NATURAL/USING joins
in a FROM clause.
SYNOPSIS
setup_natural_join_row_types()
thd current thread
from_clause list of top-level table references in a FROM clause
DESCRIPTION
Apply the procedure 'store_top_level_join_columns' to each of the
top-level table referencs of the FROM clause. Adjust the list of tables
for name resolution - context->first_name_resolution_table to the
top-most, lef-most NATURAL/USING join.
IMPLEMENTATION
Notice that the table references in 'from_clause' are in reverse
order, thus when we iterate over it, we are moving from the right
to the left in the FROM clause.
RETURN
TRUE Error
FALSE OK
*/
static bool setup_natural_join_row_types(THD *thd,
List<TABLE_LIST> *from_clause,
Name_resolution_context *context)
{
DBUG_ENTER("setup_natural_join_row_types");
thd->where= "from clause";
if (from_clause->elements == 0)
DBUG_RETURN(false); /* We come here in the case of UNIONs. */
List_iterator_fast<TABLE_LIST> table_ref_it(*from_clause);
TABLE_LIST *table_ref; /* Current table reference. */
/* Table reference to the left of the current. */
TABLE_LIST *left_neighbor;
/* Table reference to the right of the current. */
TABLE_LIST *right_neighbor= NULL;
/* Note that tables in the list are in reversed order */
for (left_neighbor= table_ref_it++; left_neighbor ; )
{
table_ref= left_neighbor;
left_neighbor= table_ref_it++;
/*
Do not redo work if already done:
1) for stored procedures,
2) for multitable update after lock failure and table reopening.
*/
if (context->select_lex->first_natural_join_processing)
{
if (store_top_level_join_columns(thd, table_ref,
left_neighbor, right_neighbor))
DBUG_RETURN(true);
if (left_neighbor)
{
TABLE_LIST *first_leaf_on_the_right;
first_leaf_on_the_right= table_ref->first_leaf_for_name_resolution();
left_neighbor->next_name_resolution_table= first_leaf_on_the_right;
}
}
right_neighbor= table_ref;
}
/*
Store the top-most, left-most NATURAL/USING join, so that we start
the search from that one instead of context->table_list. At this point
right_neighbor points to the left-most top-level table reference in the
FROM clause.
*/
DBUG_ASSERT(right_neighbor);
context->first_name_resolution_table=
right_neighbor->first_leaf_for_name_resolution();
context->select_lex->first_natural_join_processing= false;
DBUG_RETURN (false);
}
2000-07-31 21:29:14 +02:00
/****************************************************************************
** Expand all '*' in given fields
2000-07-31 21:29:14 +02:00
****************************************************************************/
int setup_wild(THD *thd, TABLE_LIST *tables, List<Item> &fields,
List<Item> *sum_func_list,
uint wild_num)
2000-07-31 21:29:14 +02:00
{
if (!wild_num)
return(0);
Item *item;
List_iterator<Item> it(fields);
Query_arena *arena, backup;
DBUG_ENTER("setup_wild");
2004-06-25 15:52:01 +02:00
/*
Don't use arena if we are not in prepared statements or stored procedures
For PS/SP we have to use arena to remember the changes
2004-06-25 15:52:01 +02:00
*/
arena= thd->activate_stmt_arena_if_needed(&backup);
thd->lex->current_select->cur_pos_in_select_list= 0;
while (wild_num && (item= it++))
2004-07-16 00:15:55 +02:00
{
if (item->type() == Item::FIELD_ITEM &&
((Item_field*) item)->field_name &&
((Item_field*) item)->field_name[0] == '*' &&
!((Item_field*) item)->field)
{
uint elem= fields.elements;
2004-07-16 00:15:55 +02:00
bool any_privileges= ((Item_field *) item)->any_privileges;
Item_subselect *subsel= thd->lex->current_select->master_unit()->item;
if (subsel &&
subsel->substype() == Item_subselect::EXISTS_SUBS)
{
/*
It is EXISTS(SELECT * ...) and we can replace * by any constant.
Item_int do not need fix_fields() because it is basic constant.
*/
it.replace(new Item_int("Not_used", (longlong) 1,
MY_INT64_NUM_DECIMAL_DIGITS));
}
else if (insert_fields(thd, ((Item_field*) item)->context,
((Item_field*) item)->db_name,
2004-09-10 01:22:44 +02:00
((Item_field*) item)->table_name, &it,
any_privileges))
2004-06-25 15:52:01 +02:00
{
if (arena)
thd->restore_active_arena(arena, &backup);
DBUG_RETURN(-1);
2004-06-25 15:52:01 +02:00
}
if (sum_func_list)
{
/*
sum_func_list is a list that has the fields list as a tail.
Because of this we have to update the element count also for this
list after expanding the '*' entry.
*/
sum_func_list->elements+= fields.elements - elem;
}
wild_num--;
}
else
thd->lex->current_select->cur_pos_in_select_list++;
}
thd->lex->current_select->cur_pos_in_select_list= UNDEF_POS;
2004-06-25 15:52:01 +02:00
if (arena)
{
/* make * substituting permanent */
SELECT_LEX *select_lex= thd->lex->current_select;
select_lex->with_wild= 0;
/*
The assignment below is translated to memcpy() call (at least on some
platforms). memcpy() expects that source and destination areas do not
overlap. That problem was detected by valgrind.
*/
if (&select_lex->item_list != &fields)
select_lex->item_list= fields;
2004-06-25 15:52:01 +02:00
thd->restore_active_arena(arena, &backup);
2004-06-25 15:52:01 +02:00
}
DBUG_RETURN(0);
}
/****************************************************************************
** Check that all given fields exists and fill struct with current data
****************************************************************************/
bool setup_fields(THD *thd, Item **ref_pointer_array,
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
List<Item> &fields, enum_mark_columns mark_used_columns,
List<Item> *sum_func_list, bool allow_sum_func)
{
reg2 Item *item;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
enum_mark_columns save_mark_used_columns= thd->mark_used_columns;
nesting_map save_allow_sum_func= thd->lex->allow_sum_func;
List_iterator<Item> it(fields);
bool save_is_item_list_lookup;
DBUG_ENTER("setup_fields");
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
thd->mark_used_columns= mark_used_columns;
DBUG_PRINT("info", ("thd->mark_used_columns: %d", thd->mark_used_columns));
if (allow_sum_func)
thd->lex->allow_sum_func|= 1 << thd->lex->current_select->nest_level;
thd->where= THD::DEFAULT_WHERE;
save_is_item_list_lookup= thd->lex->current_select->is_item_list_lookup;
thd->lex->current_select->is_item_list_lookup= 0;
/*
To prevent fail on forward lookup we fill it with zerows,
then if we got pointer on zero after find_item_in_list we will know
that it is forward lookup.
There is other way to solve problem: fill array with pointers to list,
but it will be slower.
TODO: remove it when (if) we made one list for allfields and
ref_pointer_array
*/
if (ref_pointer_array)
bzero(ref_pointer_array, sizeof(Item *) * fields.elements);
/*
We call set_entry() there (before fix_fields() of the whole list of field
items) because:
1) the list of field items has same order as in the query, and the
Item_func_get_user_var item may go before the Item_func_set_user_var:
SELECT @a, @a := 10 FROM t;
2) The entry->update_query_id value controls constantness of
Item_func_get_user_var items, so in presence of Item_func_set_user_var
items we have to refresh their entries before fixing of
Item_func_get_user_var items.
*/
List_iterator<Item_func_set_user_var> li(thd->lex->set_var_list);
Item_func_set_user_var *var;
while ((var= li++))
var->set_entry(thd, FALSE);
2003-02-02 22:30:01 +01:00
Item **ref= ref_pointer_array;
thd->lex->current_select->cur_pos_in_select_list= 0;
2003-02-02 22:30:01 +01:00
while ((item= it++))
{
if ((!item->fixed && item->fix_fields(thd, it.ref())) ||
2004-06-25 15:52:01 +02:00
(item= *(it.ref()))->check_cols(1))
2004-07-23 08:20:58 +02:00
{
thd->lex->current_select->is_item_list_lookup= save_is_item_list_lookup;
thd->lex->allow_sum_func= save_allow_sum_func;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
thd->mark_used_columns= save_mark_used_columns;
DBUG_PRINT("info", ("thd->mark_used_columns: %d", thd->mark_used_columns));
DBUG_RETURN(TRUE); /* purecov: inspected */
2004-07-23 08:20:58 +02:00
}
2003-02-02 22:30:01 +01:00
if (ref)
*(ref++)= item;
if (item->with_sum_func && item->type() != Item::SUM_FUNC_ITEM &&
sum_func_list)
item->split_sum_func(thd, ref_pointer_array, *sum_func_list);
thd->used_tables|= item->used_tables();
thd->lex->current_select->cur_pos_in_select_list++;
}
thd->lex->current_select->is_item_list_lookup= save_is_item_list_lookup;
thd->lex->current_select->cur_pos_in_select_list= UNDEF_POS;
thd->lex->allow_sum_func= save_allow_sum_func;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
thd->mark_used_columns= save_mark_used_columns;
DBUG_PRINT("info", ("thd->mark_used_columns: %d", thd->mark_used_columns));
DBUG_RETURN(test(thd->is_error()));
}
/*
make list of leaves of join table tree
SYNOPSIS
make_leaves_list()
list pointer to pointer on list first element
tables table list
RETURN pointer on pointer to next_leaf of last element
*/
TABLE_LIST **make_leaves_list(TABLE_LIST **list, TABLE_LIST *tables)
{
for (TABLE_LIST *table= tables; table; table= table->next_local)
{
if (table->merge_underlying_list)
{
DBUG_ASSERT(table->view &&
table->effective_algorithm == VIEW_ALGORITHM_MERGE);
list= make_leaves_list(list, table->merge_underlying_list);
}
else
{
*list= table;
list= &table->next_leaf;
}
}
return list;
}
/*
2004-06-25 15:52:01 +02:00
prepare tables
SYNOPSIS
setup_tables()
2004-11-25 01:23:13 +01:00
thd Thread handler
context name resolution contest to setup table list there
from_clause Top-level list of table references in the FROM clause
tables Table list (select_lex->table_list)
leaves List of join table leaves list (select_lex->leaf_tables)
2004-11-25 01:23:13 +01:00
refresh It is onle refresh for subquery
select_insert It is SELECT ... INSERT command
2004-06-25 15:52:01 +02:00
2004-10-29 18:26:52 +02:00
NOTE
Check also that the 'used keys' and 'ignored keys' exists and set up the
table structure accordingly.
Create a list of leaf tables. For queries with NATURAL/USING JOINs,
compute the row types of the top most natural/using join table references
and link these into a list of table references for name resolution.
2004-10-29 18:26:52 +02:00
This has to be called for all tables that are used by items, as otherwise
table->map is not set and all Item_field will be regarded as const items.
2004-07-16 00:15:55 +02:00
2004-10-29 18:26:52 +02:00
RETURN
FALSE ok; In this case *map will includes the chosen index
2004-11-25 01:23:13 +01:00
TRUE error
*/
bool setup_tables(THD *thd, Name_resolution_context *context,
List<TABLE_LIST> *from_clause, TABLE_LIST *tables,
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
TABLE_LIST **leaves, bool select_insert)
{
2004-12-30 23:44:00 +01:00
uint tablenr= 0;
DBUG_ENTER("setup_tables");
DBUG_ASSERT ((select_insert && !tables->next_name_resolution_table) || !tables ||
(context->table_list && context->first_name_resolution_table));
2004-11-25 01:23:13 +01:00
/*
this is used for INSERT ... SELECT.
For select we setup tables except first (and its underlying tables)
2004-11-25 01:23:13 +01:00
*/
TABLE_LIST *first_select_table= (select_insert ?
tables->next_local:
0);
if (!(*leaves))
make_leaves_list(leaves, tables);
2005-06-02 15:32:02 +02:00
TABLE_LIST *table_list;
for (table_list= *leaves;
2004-07-16 00:15:55 +02:00
table_list;
table_list= table_list->next_leaf, tablenr++)
2000-07-31 21:29:14 +02:00
{
TABLE *table= table_list->table;
2006-06-01 07:55:45 +02:00
table->pos_in_table_list= table_list;
2004-11-25 01:23:13 +01:00
if (first_select_table &&
table_list->top_table() == first_select_table)
2004-11-25 01:23:13 +01:00
{
/* new counting for SELECT of INSERT ... SELECT command */
first_select_table= 0;
tablenr= 0;
}
setup_table_map(table, table_list, tablenr);
if (table_list->process_index_hints(table))
DBUG_RETURN(1);
2000-07-31 21:29:14 +02:00
}
if (tablenr > MAX_TABLES)
{
2011-05-21 10:21:08 +02:00
my_error(ER_TOO_MANY_TABLES,MYF(0), static_cast<int>(MAX_TABLES));
DBUG_RETURN(1);
2000-07-31 21:29:14 +02:00
}
2005-06-02 15:32:02 +02:00
for (table_list= tables;
table_list;
table_list= table_list->next_local)
{
if (table_list->merge_underlying_list)
{
DBUG_ASSERT(table_list->view &&
table_list->effective_algorithm == VIEW_ALGORITHM_MERGE);
Query_arena *arena= thd->stmt_arena, backup;
bool res;
if (arena->is_conventional())
arena= 0; // For easier test
else
thd->set_n_backup_active_arena(arena, &backup);
res= table_list->setup_underlying(thd);
if (arena)
thd->restore_active_arena(arena, &backup);
if (res)
DBUG_RETURN(1);
}
}
/* Precompute and store the row types of NATURAL/USING joins. */
if (setup_natural_join_row_types(thd, from_clause, context))
DBUG_RETURN(1);
DBUG_RETURN(0);
2000-07-31 21:29:14 +02:00
}
2000-07-31 21:29:14 +02:00
/*
prepare tables and check access for the view tables
SYNOPSIS
setup_tables_and_check_view_access()
thd Thread handler
context name resolution contest to setup table list there
from_clause Top-level list of table references in the FROM clause
tables Table list (select_lex->table_list)
conds Condition of current SELECT (can be changed by VIEW)
leaves List of join table leaves list (select_lex->leaf_tables)
refresh It is onle refresh for subquery
select_insert It is SELECT ... INSERT command
want_access what access is needed
NOTE
a wrapper for check_tables that will also check the resulting
table leaves list for access to all the tables that belong to a view
RETURN
FALSE ok; In this case *map will include the chosen index
TRUE error
*/
bool setup_tables_and_check_access(THD *thd,
Name_resolution_context *context,
List<TABLE_LIST> *from_clause,
TABLE_LIST *tables,
TABLE_LIST **leaves,
bool select_insert,
ulong want_access_first,
ulong want_access)
{
TABLE_LIST *leaves_tmp= NULL;
bool first_table= true;
if (setup_tables(thd, context, from_clause, tables,
&leaves_tmp, select_insert))
return TRUE;
2007-03-22 09:19:41 +01:00
if (leaves)
*leaves= leaves_tmp;
for (; leaves_tmp; leaves_tmp= leaves_tmp->next_leaf)
{
if (leaves_tmp->belong_to_view &&
check_single_table_access(thd, first_table ? want_access_first :
want_access, leaves_tmp, FALSE))
{
tables->hide_view_error(thd);
return TRUE;
}
first_table= 0;
}
return FALSE;
}
/*
Create a key_map from a list of index names
SYNOPSIS
get_key_map_from_key_list()
map key_map to fill in
table Table
index_list List of index names
RETURN
0 ok; In this case *map will includes the choosed index
1 error
*/
bool get_key_map_from_key_list(key_map *map, TABLE *table,
2003-10-11 13:06:55 +02:00
List<String> *index_list)
2000-07-31 21:29:14 +02:00
{
List_iterator_fast<String> it(*index_list);
2000-07-31 21:29:14 +02:00
String *name;
uint pos;
2003-10-11 13:06:55 +02:00
map->clear_all();
2000-07-31 21:29:14 +02:00
while ((name=it++))
{
if (table->s->keynames.type_names == 0 ||
(pos= find_type(&table->s->keynames, name->ptr(),
name->length(), 1)) <=
0)
2000-07-31 21:29:14 +02:00
{
my_error(ER_KEY_DOES_NOT_EXITS, MYF(0), name->c_ptr(),
2006-06-01 07:55:45 +02:00
table->pos_in_table_list->alias);
2003-10-11 13:06:55 +02:00
map->set_all();
return 1;
2000-07-31 21:29:14 +02:00
}
2003-10-11 13:06:55 +02:00
map->set_bit(pos-1);
2000-07-31 21:29:14 +02:00
}
return 0;
2000-07-31 21:29:14 +02:00
}
/*
Drops in all fields instead of current '*' field
SYNOPSIS
insert_fields()
thd Thread handler
context Context for name resolution
db_name Database name in case of 'database_name.table_name.*'
table_name Table name in case of 'table_name.*'
it Pointer to '*'
any_privileges 0 If we should ensure that we have SELECT privileges
for all columns
1 If any privilege is ok
RETURN
0 ok 'it' is updated to point at last inserted
1 error. Error message is generated but not sent to client
*/
2000-07-31 21:29:14 +02:00
2001-07-08 03:15:41 +02:00
bool
insert_fields(THD *thd, Name_resolution_context *context, const char *db_name,
2004-07-16 00:15:55 +02:00
const char *table_name, List_iterator<Item> *it,
bool any_privileges)
2000-07-31 21:29:14 +02:00
{
Field_iterator_table_ref field_iterator;
bool found;
char name_buff[NAME_LEN+1];
2000-07-31 21:29:14 +02:00
DBUG_ENTER("insert_fields");
DBUG_PRINT("arena", ("stmt arena: 0x%lx", (ulong)thd->stmt_arena));
2000-07-31 21:29:14 +02:00
if (db_name && lower_case_table_names)
{
/*
convert database to lower case for comparison
We can't do this in Item_field as this would change the
'name' of the item which may be used in the select list
*/
2004-08-26 18:22:22 +02:00
strmake(name_buff, db_name, sizeof(name_buff)-1);
2004-09-02 15:57:26 +02:00
my_casedn_str(files_charset_info, name_buff);
db_name= name_buff;
}
found= FALSE;
/*
If table names are qualified, then loop over all tables used in the query,
else treat natural joins as leaves and do not iterate over their underlying
tables.
*/
for (TABLE_LIST *tables= (table_name ? context->table_list :
context->first_name_resolution_table);
tables;
tables= (table_name ? tables->next_local :
tables->next_name_resolution_table)
)
2000-07-31 21:29:14 +02:00
{
Field *field;
TABLE *table= tables->table;
DBUG_ASSERT(tables->is_leaf_for_name_resolution());
if ((table_name && my_strcasecmp(table_alias_charset, table_name,
tables->alias)) ||
(db_name && strcmp(tables->db,db_name)))
continue;
#ifndef NO_EMBEDDED_ACCESS_CHECKS
/*
Ensure that we have access rights to all fields to be inserted. Under
some circumstances, this check may be skipped.
- If any_privileges is true, skip the check.
- If the SELECT privilege has been found as fulfilled already for both
the TABLE and TABLE_LIST objects (and both of these exist, of
course), the check is skipped.
- If the SELECT privilege has been found fulfilled for the TABLE object
and the TABLE_LIST represents a derived table other than a view (see
below), the check is skipped.
- If the TABLE_LIST object represents a view, we may skip checking if
the SELECT privilege has been found fulfilled for it, regardless of
the TABLE object.
- If there is no TABLE object, the test is skipped if either
* the TABLE_LIST does not represent a view, or
* the SELECT privilege has been found fulfilled.
A TABLE_LIST that is not a view may be a subquery, an
information_schema table, or a nested table reference. See the comment
for TABLE_LIST.
*/
if (!((table && !tables->view && (table->grant.privilege & SELECT_ACL)) ||
2009-06-17 16:56:44 +02:00
(tables->view && (tables->grant.privilege & SELECT_ACL))) &&
!any_privileges)
{
field_iterator.set(tables);
if (check_grant_all_columns(thd, SELECT_ACL, &field_iterator))
DBUG_RETURN(TRUE);
}
#endif
2004-07-16 00:15:55 +02:00
/*
Update the tables used in the query based on the referenced fields. For
views and natural joins this update is performed inside the loop below.
*/
if (table)
thd->used_tables|= table->map;
/*
Initialize a generic field iterator for the current table reference.
Notice that it is guaranteed that this iterator will iterate over the
fields of a single table reference, because 'tables' is a leaf (for
name resolution purposes).
*/
field_iterator.set(tables);
for (; !field_iterator.end_of_fields(); field_iterator.next())
{
Item *item;
if (!(item= field_iterator.create_item(thd)))
DBUG_RETURN(TRUE);
DBUG_ASSERT(item->fixed);
/* cache the table for the Item_fields inserted by expanding stars */
if (item->type() == Item::FIELD_ITEM && tables->cacheable_table)
((Item_field *)item)->cached_table= tables;
if (!found)
{
found= TRUE;
it->replace(item); /* Replace '*' with the first found item. */
}
2004-07-16 00:15:55 +02:00
else
it->after(item); /* Add 'item' to the SELECT list. */
#ifndef NO_EMBEDDED_ACCESS_CHECKS
/*
Set privilege information for the fields of newly created views.
We have that (any_priviliges == TRUE) if and only if we are creating
a view. In the time of view creation we can't use the MERGE algorithm,
therefore if 'tables' is itself a view, it is represented by a
temporary table. Thus in this case we can be sure that 'item' is an
Item_field.
*/
if (any_privileges)
{
DBUG_ASSERT((tables->field_translation == NULL && table) ||
tables->is_natural_join);
DBUG_ASSERT(item->type() == Item::FIELD_ITEM);
Item_field *fld= (Item_field*) item;
const char *field_table_name= field_iterator.get_table_name();
if (!tables->schema_table &&
!(fld->have_privileges=
(get_column_grant(thd, field_iterator.grant(),
field_iterator.get_db_name(),
field_table_name, fld->field_name) &
VIEW_ANY_ACL)))
{
my_error(ER_TABLEACCESS_DENIED_ERROR, MYF(0), "ANY",
thd->security_ctx->priv_user,
thd->security_ctx->host_or_ip,
field_table_name);
DBUG_RETURN(TRUE);
}
}
#endif
if ((field= field_iterator.field()))
2000-07-31 21:29:14 +02:00
{
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
/* Mark fields as used to allow storage engine to optimze access */
bitmap_set_bit(field->table->read_set, field->field_index);
if (table)
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
{
table->covering_keys.intersect(field->part_of_key);
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
table->merge_keys.merge(field->part_of_key);
}
if (tables->is_natural_join)
2004-06-25 15:52:01 +02:00
{
TABLE *field_table;
/*
In this case we are sure that the column ref will not be created
because it was already created and stored with the natural join.
*/
Natural_join_column *nj_col;
if (!(nj_col= field_iterator.get_natural_column_ref()))
DBUG_RETURN(TRUE);
DBUG_ASSERT(nj_col->table_field);
field_table= nj_col->table_ref->table;
if (field_table)
2004-07-16 00:15:55 +02:00
{
thd->used_tables|= field_table->map;
field_table->covering_keys.intersect(field->part_of_key);
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
field_table->merge_keys.merge(field->part_of_key);
field_table->used_fields++;
2004-07-16 00:15:55 +02:00
}
}
2000-07-31 21:29:14 +02:00
}
else
thd->used_tables|= item->used_tables();
thd->lex->current_select->cur_pos_in_select_list++;
2000-07-31 21:29:14 +02:00
}
/*
In case of stored tables, all fields are considered as used,
while in the case of views, the fields considered as used are the
ones marked in setup_tables during fix_fields of view columns.
For NATURAL joins, used_tables is updated in the IF above.
*/
if (table)
table->used_fields= table->s->fields;
2000-07-31 21:29:14 +02:00
}
if (found)
DBUG_RETURN(FALSE);
/*
TODO: in the case when we skipped all columns because there was a
qualified '*', and all columns were coalesced, we have to give a more
meaningful message than ER_BAD_TABLE_ERROR.
*/
if (!table_name)
2004-11-12 13:34:00 +01:00
my_message(ER_NO_TABLES_USED, ER(ER_NO_TABLES_USED), MYF(0));
else
my_error(ER_BAD_TABLE_ERROR, MYF(0), table_name);
DBUG_RETURN(TRUE);
2000-07-31 21:29:14 +02:00
}
/*
Fix all conditions and outer join expressions.
SYNOPSIS
setup_conds()
thd thread handler
tables list of tables for name resolving (select_lex->table_list)
leaves list of leaves of join table tree (select_lex->leaf_tables)
conds WHERE clause
DESCRIPTION
TODO
RETURN
TRUE if some error occured (e.g. out of memory)
FALSE if all is OK
2000-07-31 21:29:14 +02:00
*/
int setup_conds(THD *thd, TABLE_LIST *tables, TABLE_LIST *leaves,
COND **conds)
2000-07-31 21:29:14 +02:00
{
2004-06-25 15:52:01 +02:00
SELECT_LEX *select_lex= thd->lex->current_select;
2004-11-12 12:28:34 +01:00
TABLE_LIST *table= NULL; // For HP compilers
/*
it_is_update set to TRUE when tables of primary SELECT_LEX (SELECT_LEX
which belong to LEX, i.e. most up SELECT) will be updated by
INSERT/UPDATE/LOAD
NOTE: using this condition helps to prevent call of prepare_check_option()
from subquery of VIEW, because tables of subquery belongs to VIEW
(see condition before prepare_check_option() call)
*/
bool it_is_update= (select_lex == &thd->lex->select_lex) &&
thd->lex->which_check_option_applicable();
bool save_is_item_list_lookup= select_lex->is_item_list_lookup;
select_lex->is_item_list_lookup= 0;
2000-07-31 21:29:14 +02:00
DBUG_ENTER("setup_conds");
2004-06-25 15:52:01 +02:00
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
thd->mark_used_columns= MARK_COLUMNS_READ;
DBUG_PRINT("info", ("thd->mark_used_columns: %d", thd->mark_used_columns));
select_lex->cond_count= 0;
select_lex->between_count= 0;
select_lex->max_equal_elems= 0;
2004-07-16 00:15:55 +02:00
for (table= tables; table; table= table->next_local)
{
if (table->prepare_where(thd, conds, FALSE))
goto err_no_arena;
}
2000-07-31 21:29:14 +02:00
if (*conds)
{
thd->where="where clause";
if ((!(*conds)->fixed && (*conds)->fix_fields(thd, conds)) ||
2004-06-25 15:52:01 +02:00
(*conds)->check_cols(1))
goto err_no_arena;
2000-07-31 21:29:14 +02:00
}
/*
Apply fix_fields() to all ON clauses at all levels of nesting,
including the ones inside view definitions.
*/
2004-11-21 19:08:12 +01:00
for (table= leaves; table; table= table->next_leaf)
2000-07-31 21:29:14 +02:00
{
TABLE_LIST *embedded; /* The table at the current level of nesting. */
TABLE_LIST *embedding= table; /* The parent nested table reference. */
do
2004-07-16 00:15:55 +02:00
{
embedded= embedding;
if (embedded->on_expr)
{
/* Make a join an a expression */
thd->where="on clause";
if ((!embedded->on_expr->fixed &&
embedded->on_expr->fix_fields(thd, &embedded->on_expr)) ||
embedded->on_expr->check_cols(1))
goto err_no_arena;
2004-06-25 15:52:01 +02:00
select_lex->cond_count++;
}
embedding= embedded->embedding;
2000-07-31 21:29:14 +02:00
}
while (embedding &&
embedding->nested_join->join_list.head() == embedded);
/* process CHECK OPTION */
if (it_is_update)
{
2005-09-02 08:50:17 +02:00
TABLE_LIST *view= table->top_table();
if (view->effective_with_check)
{
if (view->prepare_check_option(thd))
goto err_no_arena;
thd->change_item_tree(&table->check_option, view->check_option);
}
}
2000-07-31 21:29:14 +02:00
}
2004-06-25 15:52:01 +02:00
if (!thd->stmt_arena->is_conventional())
2004-06-25 15:52:01 +02:00
{
/*
We are in prepared statement preparation code => we should store
WHERE clause changing for next executions.
We do this ON -> WHERE transformation only once per PS/SP statement.
*/
select_lex->where= *conds;
}
thd->lex->current_select->is_item_list_lookup= save_is_item_list_lookup;
DBUG_RETURN(test(thd->is_error()));
2004-06-25 15:52:01 +02:00
err_no_arena:
select_lex->is_item_list_lookup= save_is_item_list_lookup;
2004-06-25 15:52:01 +02:00
DBUG_RETURN(1);
2000-07-31 21:29:14 +02:00
}
/******************************************************************************
** Fill a record with data (for INSERT or UPDATE)
** Returns : 1 if some field has wrong type
******************************************************************************/
2004-11-12 13:34:00 +01:00
/*
Fill fields with given items.
SYNOPSIS
fill_record()
thd thread handler
fields Item_fields list to be filled
values values to fill with
ignore_errors TRUE if we should ignore errors
NOTE
fill_record() may set table->auto_increment_field_not_null and a
caller should make sure that it is reset after their last call to this
function.
2004-11-12 13:34:00 +01:00
RETURN
FALSE OK
TRUE error occured
*/
static bool
2004-11-12 13:34:00 +01:00
fill_record(THD * thd, List<Item> &fields, List<Item> &values,
bool ignore_errors)
2000-07-31 21:29:14 +02:00
{
List_iterator_fast<Item> f(fields),v(values);
Item *value, *fld;
2000-07-31 21:29:14 +02:00
Item_field *field;
TABLE *table= 0;
2000-07-31 21:29:14 +02:00
DBUG_ENTER("fill_record");
/*
Reset the table->auto_increment_field_not_null as it is valid for
only one row.
*/
if (fields.elements)
{
/*
On INSERT or UPDATE fields are checked to be from the same table,
thus we safely can take table from the first field.
*/
fld= (Item_field*)f++;
if (!(field= fld->filed_for_view_update()))
{
my_error(ER_NONUPDATEABLE_COLUMN, MYF(0), fld->name);
goto err;
}
table= field->field->table;
table->auto_increment_field_not_null= FALSE;
f.rewind();
}
while ((fld= f++))
2000-07-31 21:29:14 +02:00
{
if (!(field= fld->filed_for_view_update()))
{
my_error(ER_NONUPDATEABLE_COLUMN, MYF(0), fld->name);
goto err;
}
2000-07-31 21:29:14 +02:00
value=v++;
Field *rfield= field->field;
table= rfield->table;
2003-07-28 16:57:46 +02:00
if (rfield == table->next_number_field)
2004-07-30 09:47:56 +02:00
table->auto_increment_field_not_null= TRUE;
2004-06-25 15:52:01 +02:00
if ((value->save_in_field(rfield, 0) < 0) && !ignore_errors)
2004-11-12 13:34:00 +01:00
{
my_message(ER_UNKNOWN_ERROR, ER(ER_UNKNOWN_ERROR), MYF(0));
goto err;
2004-11-12 13:34:00 +01:00
}
2000-07-31 21:29:14 +02:00
}
DBUG_RETURN(thd->is_error());
err:
if (table)
table->auto_increment_field_not_null= FALSE;
DBUG_RETURN(TRUE);
2000-07-31 21:29:14 +02:00
}
/*
Fill fields in list with values from the list of items and invoke
before triggers.
SYNOPSIS
fill_record_n_invoke_before_triggers()
thd thread context
fields Item_fields list to be filled
values values to fill with
ignore_errors TRUE if we should ignore errors
triggers object holding list of triggers to be invoked
event event type for triggers to be invoked
NOTE
This function assumes that fields which values will be set and triggers
to be invoked belong to the same table, and that TABLE::record[0] and
record[1] buffers correspond to new and old versions of row respectively.
RETURN
FALSE OK
TRUE error occured
*/
bool
fill_record_n_invoke_before_triggers(THD *thd, List<Item> &fields,
List<Item> &values, bool ignore_errors,
Table_triggers_list *triggers,
enum trg_event_type event)
{
return (fill_record(thd, fields, values, ignore_errors) ||
(triggers && triggers->process_triggers(thd, event,
TRG_ACTION_BEFORE, TRUE)));
}
2004-11-12 13:34:00 +01:00
/*
Fill field buffer with values from Field list
SYNOPSIS
fill_record()
thd thread handler
ptr pointer on pointer to record
values list of fields
ignore_errors TRUE if we should ignore errors
NOTE
fill_record() may set table->auto_increment_field_not_null and a
caller should make sure that it is reset after their last call to this
function.
2004-11-12 13:34:00 +01:00
RETURN
FALSE OK
TRUE error occured
*/
bool
fill_record(THD *thd, Field **ptr, List<Item> &values, bool ignore_errors)
2000-07-31 21:29:14 +02:00
{
List_iterator_fast<Item> v(values);
2000-07-31 21:29:14 +02:00
Item *value;
TABLE *table= 0;
2000-07-31 21:29:14 +02:00
DBUG_ENTER("fill_record");
Field *field;
/*
Reset the table->auto_increment_field_not_null as it is valid for
only one row.
*/
if (*ptr)
{
/*
On INSERT or UPDATE fields are checked to be from the same table,
thus we safely can take table from the first field.
*/
table= (*ptr)->table;
table->auto_increment_field_not_null= FALSE;
}
while ((field = *ptr++) && ! thd->is_error())
2000-07-31 21:29:14 +02:00
{
value=v++;
table= field->table;
2003-07-28 16:57:46 +02:00
if (field == table->next_number_field)
2004-07-30 09:47:56 +02:00
table->auto_increment_field_not_null= TRUE;
if (value->save_in_field(field, 0) < 0)
goto err;
2000-07-31 21:29:14 +02:00
}
DBUG_RETURN(thd->is_error());
err:
if (table)
table->auto_increment_field_not_null= FALSE;
DBUG_RETURN(TRUE);
2000-07-31 21:29:14 +02:00
}
/*
Fill fields in array with values from the list of items and invoke
before triggers.
SYNOPSIS
fill_record_n_invoke_before_triggers()
thd thread context
ptr NULL-ended array of fields to be filled
values values to fill with
ignore_errors TRUE if we should ignore errors
triggers object holding list of triggers to be invoked
event event type for triggers to be invoked
NOTE
This function assumes that fields which values will be set and triggers
to be invoked belong to the same table, and that TABLE::record[0] and
record[1] buffers correspond to new and old versions of row respectively.
RETURN
FALSE OK
TRUE error occured
*/
bool
fill_record_n_invoke_before_triggers(THD *thd, Field **ptr,
List<Item> &values, bool ignore_errors,
Table_triggers_list *triggers,
enum trg_event_type event)
{
return (fill_record(thd, ptr, values, ignore_errors) ||
(triggers && triggers->process_triggers(thd, event,
TRG_ACTION_BEFORE, TRUE)));
}
my_bool mysql_rm_tmp_tables(void)
2000-07-31 21:29:14 +02:00
{
2002-10-08 14:39:37 +02:00
uint i, idx;
char filePath[FN_REFLEN], *tmpdir, filePathCopy[FN_REFLEN];
2000-07-31 21:29:14 +02:00
MY_DIR *dirp;
FILEINFO *file;
2007-01-11 13:50:34 +01:00
TABLE_SHARE share;
THD *thd;
2000-07-31 21:29:14 +02:00
DBUG_ENTER("mysql_rm_tmp_tables");
if (!(thd= new THD))
DBUG_RETURN(1);
2007-01-10 11:03:36 +01:00
thd->thread_stack= (char*) &thd;
thd->store_globals();
2002-10-08 14:39:37 +02:00
for (i=0; i<=mysql_tmpdir_list.max; i++)
{
tmpdir=mysql_tmpdir_list.list[i];
/* See if the directory exists */
2002-10-08 14:39:37 +02:00
if (!(dirp = my_dir(tmpdir,MYF(MY_WME | MY_DONT_SORT))))
continue;
2000-07-31 21:29:14 +02:00
2002-10-08 14:39:37 +02:00
/* Remove all SQLxxx tables from directory */
2000-07-31 21:29:14 +02:00
for (idx=0 ; idx < (uint) dirp->number_off_files ; idx++)
{
file=dirp->dir_entry+idx;
2004-06-25 15:52:01 +02:00
/* skiping . and .. */
if (file->name[0] == '.' && (!file->name[1] ||
(file->name[1] == '.' && !file->name[2])))
continue;
2004-06-25 15:52:01 +02:00
if (!memcmp(file->name, tmp_file_prefix,
tmp_file_prefix_length))
{
char *ext= fn_ext(file->name);
uint ext_len= strlen(ext);
uint filePath_len= my_snprintf(filePath, sizeof(filePath),
"%s%c%s", tmpdir, FN_LIBCHAR,
file->name);
if (!memcmp(reg_ext, ext, ext_len))
{
handler *handler_file= 0;
/* We should cut file extention before deleting of table */
memcpy(filePathCopy, filePath, filePath_len - ext_len);
filePathCopy[filePath_len - ext_len]= 0;
Bug#26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Bug 26377 - Deadlock with MERGE and FLUSH TABLE Bug 25038 - Waiting TRUNCATE Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Bug 19627 - temporary merge table locking Bug 27660 - Falcon: merge table possible Bug 30273 - merge tables: Can't lock file (errno: 155) The problems were: Bug 26379 - Combination of FLUSH TABLE and REPAIR TABLE corrupts a MERGE table 1. A thread trying to lock a MERGE table performs busy waiting while REPAIR TABLE or a similar table administration task is ongoing on one or more of its MyISAM tables. 2. A thread trying to lock a MERGE table performs busy waiting until all threads that did REPAIR TABLE or similar table administration tasks on one or more of its MyISAM tables in LOCK TABLES segments do UNLOCK TABLES. The difference against problem #1 is that the busy waiting takes place *after* the administration task. It is terminated by UNLOCK TABLES only. 3. Two FLUSH TABLES within a LOCK TABLES segment can invalidate the lock. This does *not* require a MERGE table. The first FLUSH TABLES can be replaced by any statement that requires other threads to reopen the table. In 5.0 and 5.1 a single FLUSH TABLES can provoke the problem. Bug 26867 - LOCK TABLES + REPAIR + merge table result in memory/cpu hogging Trying DML on a MERGE table, which has a child locked and repaired by another thread, made an infinite loop in the server. Bug 26377 - Deadlock with MERGE and FLUSH TABLE Locking a MERGE table and its children in parent-child order and flushing the child deadlocked the server. Bug 25038 - Waiting TRUNCATE Truncating a MERGE child, while the MERGE table was in use, let the truncate fail instead of waiting for the table to become free. Bug 25700 - merge base tables get corrupted by optimize/analyze/repair table Repairing a child of an open MERGE table corrupted the child. It was necessary to FLUSH the child first. Bug 30275 - Merge tables: flush tables or unlock tables causes server to crash Flushing and optimizing locked MERGE children crashed the server. Bug 19627 - temporary merge table locking Use of a temporary MERGE table with non-temporary children could corrupt the children. Temporary tables are never locked. So we do now prohibit non-temporary chidlren of a temporary MERGE table. Bug 27660 - Falcon: merge table possible It was possible to create a MERGE table with non-MyISAM children. Bug 30273 - merge tables: Can't lock file (errno: 155) This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)". These bugs are fixed by a new implementation of MERGE table open. When opening a MERGE table in open_tables() we do now add the child tables to the list of tables to be opened by open_tables() (the "query_list"). The children are not opened in the handler at this stage. After opening the parent, open_tables() opens each child from the now extended query_list. When the last child is opened, we remove the children from the query_list again and attach the children to the parent. This behaves similar to the old open. However it does not open the MyISAM tables directly, but grabs them from the already open children. When closing a MERGE table in close_thread_table() we detach the children only. Closing of the children is done implicitly because they are in thd->open_tables. For more detail see the comment at the top of ha_myisammrg.cc. Changed from open_ltable() to open_and_lock_tables() in all places that can be relevant for MERGE tables. The latter can handle tables added to the list on the fly. When open_ltable() was used in a loop over a list of tables, the list must be temporarily terminated after every table for open_and_lock_tables(). table_list->required_type is set to FRMTYPE_TABLE to avoid open of special tables. Handling of derived tables is suppressed. These details are handled by the new function open_n_lock_single_table(), which has nearly the same signature as open_ltable() and can replace it in most cases. In reopen_tables() some of the tables open by a thread can be closed and reopened. When a MERGE child is affected, the parent must be closed and reopened too. Closing of the parent is forced before the first child is closed. Reopen happens in the order of thd->open_tables. MERGE parents do not attach their children automatically at open. This is done after all tables are reopened. So all children are open when attaching them. Special lock handling like mysql_lock_abort() or mysql_lock_remove() needs to be suppressed for MERGE children or forwarded to the parent. This depends on the situation. In loops over all open tables one suppresses child lock handling. When a single table is touched, forwarding is done. Behavioral changes: =================== This patch changes the behavior of temporary MERGE tables. Temporary MERGE must have temporary children. The old behavior was wrong. A temporary table is not locked. Hence even non-temporary children were not locked. See Bug 19627 - temporary merge table locking. You cannot change the union list of a non-temporary MERGE table when LOCK TABLES is in effect. The following does *not* work: CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...; However, you can do this with a temporary MERGE table. You cannot create a MERGE table with CREATE ... SELECT, neither as a temporary MERGE table, nor as a non-temporary MERGE table. CREATE TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...; Gives error message: table is not BASE TABLE.
2007-11-15 20:25:43 +01:00
init_tmp_table_share(thd, &share, "", 0, "", filePathCopy);
if (!open_table_def(thd, &share, 0) &&
((handler_file= get_new_handler(&share, thd->mem_root,
share.db_type()))))
{
handler_file->ha_delete_table(filePathCopy);
delete handler_file;
}
free_table_share(&share);
}
/*
File can be already deleted by tmp_table.file->delete_table().
So we hide error messages which happnes during deleting of these
files(MYF(0)).
*/
(void) mysql_file_delete(key_file_misc, filePath, MYF(0));
}
2000-07-31 21:29:14 +02:00
}
my_dirend(dirp);
2002-10-08 14:39:37 +02:00
}
delete thd;
my_pthread_setspecific_ptr(THR_THD, 0);
DBUG_RETURN(0);
2000-07-31 21:29:14 +02:00
}
/*****************************************************************************
unireg support functions
*****************************************************************************/
/*
free all unused tables
NOTE
This is called by 'handle_manager' when one wants to periodicly flush
all not used tables.
2000-07-31 21:29:14 +02:00
*/
void tdc_flush_unused_tables()
2000-07-31 21:29:14 +02:00
{
mysql_mutex_lock(&LOCK_open);
2000-07-31 21:29:14 +02:00
while (unused_tables)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
free_cache_entry(unused_tables);
mysql_mutex_unlock(&LOCK_open);
2000-07-31 21:29:14 +02:00
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
/**
A callback to the server internals that is used to address
special cases of the locking protocol.
Invoked when acquiring an exclusive lock, for each thread that
has a conflicting shared metadata lock.
This function:
- aborts waiting of the thread on a data lock, to make it notice
the pending exclusive lock and back off.
- if the thread is an INSERT DELAYED thread, sends it a KILL
signal to terminate it.
@note This function does not wait for the thread to give away its
locks. Waiting is done outside for all threads at once.
@param thd Current thread context
@param in_use The thread to wake up
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
@param needs_thr_lock_abort Indicates that to wake up thread
this call needs to abort its waiting
on table-level lock.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@retval TRUE if the thread was woken up
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
@retval FALSE otherwise.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@note It is one of two places where border between MDL and the
rest of the server is broken.
*/
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
bool mysql_notify_thread_having_shared_lock(THD *thd, THD *in_use,
bool needs_thr_lock_abort)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
bool signalled= FALSE;
if ((in_use->system_thread & SYSTEM_THREAD_DELAYED_INSERT) &&
!in_use->killed)
{
in_use->killed= THD::KILL_CONNECTION;
2010-02-02 00:22:16 +01:00
mysql_mutex_lock(&in_use->mysys_var->mutex);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if (in_use->mysys_var->current_cond)
2010-02-02 00:22:16 +01:00
mysql_cond_broadcast(in_use->mysys_var->current_cond);
mysql_mutex_unlock(&in_use->mysys_var->mutex);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
signalled= TRUE;
}
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
if (needs_thr_lock_abort)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
mysql_mutex_lock(&in_use->LOCK_thd_data);
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
for (TABLE *thd_table= in_use->open_tables;
thd_table ;
thd_table= thd_table->next)
{
/*
Check for TABLE::needs_reopen() is needed since in some places we call
handler::close() for table instance (and set TABLE::db_stat to 0)
and do not remove such instances from the THD::open_tables
for some time, during which other thread can see those instances
(e.g. see partitioning code).
*/
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
if (!thd_table->needs_reopen())
signalled|= mysql_lock_abort_for_thread(thd, thd_table);
}
mysql_mutex_unlock(&in_use->LOCK_thd_data);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
return signalled;
}
/**
Remove all or some (depending on parameter) instances of TABLE and
TABLE_SHARE from the table definition cache.
@param thd Thread context
@param remove_type Type of removal:
TDC_RT_REMOVE_ALL - remove all TABLE instances and
TABLE_SHARE instance. There
should be no used TABLE objects
and caller should have exclusive
metadata lock on the table.
TDC_RT_REMOVE_NOT_OWN - remove all TABLE instances
except those that belong to
this thread. There should be
no TABLE objects used by other
threads and caller should have
exclusive metadata lock on the
table.
TDC_RT_REMOVE_UNUSED - remove all unused TABLE
instances (if there are no
used instances will also
remove TABLE_SHARE).
@param db Name of database
@param table_name Name of table
@param has_lock If TRUE, LOCK_open is already acquired
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
@note It assumes that table instances are already not used by any
(other) thread (this should be achieved by using meta-data locks).
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
*/
void tdc_remove_table(THD *thd, enum_tdc_remove_table_type remove_type,
const char *db, const char *table_name,
bool has_lock)
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
{
char key[MAX_DBKEY_LENGTH];
uint key_length;
TABLE *table;
TABLE_SHARE *share;
if (! has_lock)
mysql_mutex_lock(&LOCK_open);
else
{
mysql_mutex_assert_owner(&LOCK_open);
}
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
DBUG_ASSERT(remove_type == TDC_RT_REMOVE_UNUSED ||
Implement new type-of-operation-aware metadata locks. Add a wait-for graph based deadlock detector to the MDL subsystem. Fixes bug #46272 "MySQL 5.4.4, new MDL: unnecessary deadlock" and bug #37346 "innodb does not detect deadlock between update and alter table". The first bug manifested itself as an unwarranted abort of a transaction with ER_LOCK_DEADLOCK error by a concurrent ALTER statement, when this transaction tried to repeat use of a table, which it has already used in a similar fashion before ALTER started. The second bug showed up as a deadlock between table-level locks and InnoDB row locks, which was "detected" only after innodb_lock_wait_timeout timeout. A transaction would start using the table and modify a few rows. Then ALTER TABLE would come in, and start copying rows into a temporary table. Eventually it would stumble on the modified records and get blocked on a row lock. The first transaction would try to do more updates, and get blocked on thr_lock.c lock. This situation of circular wait would only get resolved by a timeout. Both these bugs stemmed from inadequate solutions to the problem of deadlocks occurring between different locking subsystems. In the first case we tried to avoid deadlocks between metadata locking and table-level locking subsystems, when upgrading shared metadata lock to exclusive one. Transactions holding the shared lock on the table and waiting for some table-level lock used to be aborted too aggressively. We also allowed ALTER TABLE to start in presence of transactions that modify the subject table. ALTER TABLE acquires TL_WRITE_ALLOW_READ lock at start, and that block all writes against the table (naturally, we don't want any writes to be lost when switching the old and the new table). TL_WRITE_ALLOW_READ lock, in turn, would block the started transaction on thr_lock.c lock, should they do more updates. This, again, lead to the need to abort such transactions. The second bug occurred simply because we didn't have any mechanism to detect deadlocks between the table-level locks in thr_lock.c and row-level locks in InnoDB, other than innodb_lock_wait_timeout. This patch solves both these problems by moving lock conflicts which are causing these deadlocks into the metadata locking subsystem, thus making it possible to avoid or detect such deadlocks inside MDL. To do this we introduce new type-of-operation-aware metadata locks, which allow MDL subsystem to know not only the fact that transaction has used or is going to use some object but also what kind of operation it has carried out or going to carry out on the object. This, along with the addition of a special kind of upgradable metadata lock, allows ALTER TABLE to wait until all transactions which has updated the table to go away. This solves the second issue. Another special type of upgradable metadata lock is acquired by LOCK TABLE WRITE. This second lock type allows to solve the first issue, since abortion of table-level locks in event of DDL under LOCK TABLES becomes also unnecessary. Below follows the list of incompatible changes introduced by this patch: - From now on, ALTER TABLE and CREATE/DROP TRIGGER SQL (i.e. those statements that acquire TL_WRITE_ALLOW_READ lock) wait for all transactions which has *updated* the table to complete. - From now on, LOCK TABLES ... WRITE, REPAIR/OPTIMIZE TABLE (i.e. all statements which acquire TL_WRITE table-level lock) wait for all transaction which *updated or read* from the table to complete. As a consequence, innodb_table_locks=0 option no longer applies to LOCK TABLES ... WRITE. - DROP DATABASE, DROP TABLE, RENAME TABLE no longer abort statements or transactions which use tables being dropped or renamed, and instead wait for these transactions to complete. - Since LOCK TABLES WRITE now takes a special metadata lock, not compatible with with reads or writes against the subject table and transaction-wide, thr_lock.c deadlock avoidance algorithm that used to ensure absence of deadlocks between LOCK TABLES WRITE and other statements is no longer sufficient, even for MyISAM. The wait-for graph based deadlock detector of MDL subsystem may sometimes be necessary and is involved. This may lead to ER_LOCK_DEADLOCK error produced for multi-statement transactions even if these only use MyISAM: session 1: session 2: begin; update t1 ... lock table t2 write, t1 write; -- gets a lock on t2, blocks on t1 update t2 ... (ER_LOCK_DEADLOCK) - Finally, support of LOW_PRIORITY option for LOCK TABLES ... WRITE was abandoned. LOCK TABLE ... LOW_PRIORITY WRITE from now on has the same priority as the usual LOCK TABLE ... WRITE. SELECT HIGH PRIORITY no longer trumps LOCK TABLE ... WRITE in the wait queue. - We do not take upgradable metadata locks on implicitly locked tables. So if one has, say, a view v1 that uses table t1, and issues: LOCK TABLE v1 WRITE; FLUSH TABLE t1; -- (or just 'FLUSH TABLES'), an error is produced. In order to be able to perform DDL on a table under LOCK TABLES, the table must be locked explicitly in the LOCK TABLES list.
2010-02-01 12:43:06 +01:00
thd->mdl_context.is_lock_owner(MDL_key::TABLE, db, table_name,
MDL_EXCLUSIVE));
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
key_length=(uint) (strmov(strmov(key,db)+1,table_name)-key)+1;
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
if ((share= (TABLE_SHARE*) my_hash_search(&table_def_cache,(uchar*) key,
key_length)))
{
if (share->ref_count)
{
I_P_List_iterator<TABLE, TABLE_share> it(share->free_tables);
#ifndef DBUG_OFF
if (remove_type == TDC_RT_REMOVE_ALL)
{
DBUG_ASSERT(share->used_tables.is_empty());
}
else if (remove_type == TDC_RT_REMOVE_NOT_OWN)
{
I_P_List_iterator<TABLE, TABLE_share> it2(share->used_tables);
while ((table= it2++))
if (table->in_use != thd)
{
DBUG_ASSERT(0);
}
}
#endif
/*
Set share's version to zero in order to ensure that it gets
automatically deleted once it is no longer referenced.
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
Note that code in TABLE_SHARE::wait_for_old_version() assumes
Fix for bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are incompatible". The problem was that FLUSH TABLES <list> WITH READ LOCK which was issued when other connection has acquired global read lock using FLUSH TABLES WITH READ LOCK was blocked and has to wait until global read lock is released. This issue stemmed from the fact that FLUSH TABLES <list> WITH READ LOCK implementation has acquired X metadata locks on tables to be flushed. Since these locks required acquiring of global IX lock this statement was incompatible with global read lock. This patch addresses problem by using SNW metadata type of lock for tables to be flushed by FLUSH TABLES <list> WITH READ LOCK. It is OK to acquire them without global IX lock as long as we won't try to upgrade those locks. Since SNW locks allow concurrent statements using same table FLUSH TABLE <list> WITH READ LOCK now has to wait until old versions of tables to be flushed go away after acquiring metadata locks. Since such waiting can lead to deadlock MDL deadlock detector was extended to take into account waits for flush and resolve such deadlocks. As a bonus code in open_tables() which was responsible for waiting old versions of tables to go away was refactored. Now when we encounter old version of table in open_table() we don't back-off and wait for all old version to go away, but instead wait for this particular table to be flushed. Such approach supported by deadlock detection should reduce number of scenarios in which FLUSH TABLES aborts concurrent multi-statement transactions. Note that active FLUSH TABLES <list> WITH READ LOCK still blocks concurrent FLUSH TABLES WITH READ LOCK statement as the former keeps tables open and thus prevents the latter statement from doing flush.
2010-07-27 15:34:58 +02:00
that marking share as old and removal of its unused tables
and of the share itself from TDC happens atomically under
protection of LOCK_open, or, putting it another way, that
TDC does not contain old shares which don't have any tables
used.
*/
share->version= 0;
while ((table= it++))
free_cache_entry(table);
}
else
(void) my_hash_delete(&table_def_cache, (uchar*) share);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
if (! has_lock)
mysql_mutex_unlock(&LOCK_open);
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
}
int setup_ftfuncs(SELECT_LEX *select_lex)
2000-07-31 21:29:14 +02:00
{
List_iterator<Item_func_match> li(*(select_lex->ftfunc_list)),
lj(*(select_lex->ftfunc_list));
Item_func_match *ftf, *ftf2;
2000-07-31 21:29:14 +02:00
while ((ftf=li++))
{
2000-07-31 21:29:14 +02:00
if (ftf->fix_index())
return 1;
lj.rewind();
while ((ftf2=lj++) != ftf)
{
if (ftf->eq(ftf2,1) && !ftf2->master)
ftf2->master=ftf;
}
}
2000-07-31 21:29:14 +02:00
return 0;
}
int init_ftfuncs(THD *thd, SELECT_LEX *select_lex, bool no_order)
{
if (select_lex->ftfunc_list->elements)
{
List_iterator<Item_func_match> li(*(select_lex->ftfunc_list));
2001-12-06 00:05:30 +01:00
Item_func_match *ifm;
DBUG_PRINT("info",("Performing FULLTEXT search"));
Prevent bugs by making DBUG_* expressions syntactically equivalent to a single statement. --- Bug#24795: SHOW PROFILE Profiling is only partially functional on some architectures. Where there is no getrusage() system call, presently Null values are returned where it would be required. Notably, Windows needs some love applied to make it as useful. Syntax this adds: SHOW PROFILES SHOW PROFILE [types] [FOR QUERY n] [OFFSET n] [LIMIT n] where "n" is an integer and "types" is zero or many (comma-separated) of "CPU" "MEMORY" (not presently supported) "BLOCK IO" "CONTEXT SWITCHES" "PAGE FAULTS" "IPC" "SWAPS" "SOURCE" "ALL" It also adds a session variable (boolean) "profiling", set to "no" by default, and (integer) profiling_history_size, set to 15 by default. This patch abstracts setting THDs' "proc_info" behind a macro that can be used as a hook into the profiling code when profiling support is compiled in. All future code in this line should use that mechanism for setting thd->proc_info. --- Tests are now set to omit the statistics. --- Adds an Information_schema table, "profiling" for access to "show profile" data. --- Merge zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-community-3--bug24795 into zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-community --- Fix merge problems. --- Fixed one bug in the query_source being NULL. Updated test results. --- Include more thorough profiling tests. Improve support for prepared statements. Use session-specific query IDs, starting at zero. --- Selecting from I_S.profiling is no longer quashed in profiling, as requested by Giuseppe. Limit the size of captured query text. No longer log queries that are zero length.
2007-02-22 16:03:08 +01:00
thd_proc_info(thd, "FULLTEXT initialization");
2001-12-06 00:05:30 +01:00
while ((ifm=li++))
ifm->init_search(no_order);
}
return 0;
}
2004-07-16 00:15:55 +02:00
/*
open new .frm format table
SYNOPSIS
open_new_frm()
THD thread handler
path path to .frm file (without extension)
2004-07-16 00:15:55 +02:00
alias alias for table
db database
table_name name of table
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
db_stat open flags (for example ->OPEN_KEYFILE|HA_OPEN_RNDFILE..)
2004-07-16 00:15:55 +02:00
can be 0 (example in ha_example_table)
prgflag READ_ALL etc..
ha_open_flags HA_OPEN_ABORT_IF_LOCKED etc..
outparam result table
table_desc TABLE_LIST descriptor
mem_root temporary MEM_ROOT for parsing
*/
bool
open_new_frm(THD *thd, TABLE_SHARE *share, const char *alias,
uint db_stat, uint prgflag,
2004-07-16 00:15:55 +02:00
uint ha_open_flags, TABLE *outparam, TABLE_LIST *table_desc,
MEM_ROOT *mem_root)
{
LEX_STRING pathstr;
File_parser *parser;
char path[FN_REFLEN];
DBUG_ENTER("open_new_frm");
/* Create path with extension */
pathstr.length= (uint) (strxmov(path, share->normalized_path.str, reg_ext,
NullS)- path);
pathstr.str= path;
if ((parser= sql_parse_prepare(&pathstr, mem_root, 1)))
2004-07-16 00:15:55 +02:00
{
if (is_equal(&view_type, parser->type()))
2004-07-16 00:15:55 +02:00
{
if (table_desc == 0 || table_desc->required_type == FRMTYPE_TABLE)
2004-07-16 00:15:55 +02:00
{
my_error(ER_WRONG_OBJECT, MYF(0), share->db.str, share->table_name.str,
"BASE TABLE");
goto err;
2004-07-16 00:15:55 +02:00
}
if (mysql_make_view(thd, parser, table_desc,
(prgflag & OPEN_VIEW_NO_PARSE)))
goto err;
2004-07-16 00:15:55 +02:00
}
else
{
/* only VIEWs are supported now */
my_error(ER_FRM_UNKNOWN_TYPE, MYF(0), share->path.str, parser->type()->str);
goto err;
2004-07-16 00:15:55 +02:00
}
DBUG_RETURN(0);
2004-07-16 00:15:55 +02:00
}
2004-09-03 21:38:45 +02:00
err:
DBUG_RETURN(1);
2004-07-16 00:15:55 +02:00
}
bool is_equal(const LEX_STRING *a, const LEX_STRING *b)
{
return a->length == b->length && !strncmp(a->str, b->str, a->length);
}
/*
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
Tells if two (or more) tables have auto_increment columns and we want to
lock those tables with a write lock.
Initial import of WL#3726 "DDL locking for all metadata objects". Backport of: ------------------------------------------------------------ revno: 2630.4.1 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Fri 2008-05-23 17:54:03 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ------------------------------------------------------------ This is the first patch in series. It transforms the metadata locking subsystem to use a dedicated module (mdl.h,cc). No significant changes in the locking protocol. The import passes the test suite with the exception of deprecated/removed 6.0 features, and MERGE tables. The latter are subject to a fix by WL#4144. Unfortunately, the original changeset comments got lost in a merge, thus this import has its own (largely insufficient) comments. This patch fixes Bug#25144 "replication / binlog with view breaks". Warning: this patch introduces an incompatible change: Under LOCK TABLES, it's no longer possible to FLUSH a table that was not locked for WRITE. Under LOCK TABLES, it's no longer possible to DROP a table or VIEW that was not locked for WRITE. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.2 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:03:45 +0400 message: WL#3726 "DDL locking for all metadata objects". After review fixes in progress. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.3 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 14:08:51 +0400 message: WL#3726 "DDL locking for all metadata objects" Fixed failing Windows builds by adding mdl.cc to the lists of files needed to build server/libmysqld on Windows. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.4 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sat 2008-05-24 21:57:58 +0400 message: WL#3726 "DDL locking for all metadata objects". Fix for assert failures in kill.test which occured when one tried to kill ALTER TABLE statement on merge table while it was waiting in wait_while_table_is_used() for other connections to close this table. These assert failures stemmed from the fact that cleanup code in this case assumed that temporary table representing new version of table was open with adding to THD::temporary_tables list while code which were opening this temporary table wasn't always fulfilling this. This patch changes code that opens new version of table to always do this linking in. It also streamlines cleanup process for cases when error occurs while we have new version of table open. ****** WL#3726 "DDL locking for all metadata objects" Add libmysqld/mdl.cc to .bzrignore. ****** Backport of: ------------------------------------------------------------ revno: 2630.4.6 committer: Dmitry Lenev <dlenev@mysql.com> branch nick: mysql-6.0-3726-w timestamp: Sun 2008-05-25 00:33:22 +0400 message: WL#3726 "DDL locking for all metadata objects". Addition to the fix of assert failures in kill.test caused by changes for this worklog. Make sure we close the new table only once.
2009-11-30 16:55:03 +01:00
SYNOPSIS
has_two_write_locked_tables_with_auto_increment
tables Table list
* Mixed replication mode * : 1) Fix for BUG#19630 "stored function inserting into two auto_increment breaks statement-based binlog": a stored function inserting into two such tables may fail to replicate (inserting wrong data in the slave's copy of the second table) if the slave's second table had an internal auto_increment counter different from master's. Because the auto_increment value autogenerated by master for the 2nd table does not go into binlog, only the first does, so the slave lacks information. To fix this, if running in mixed binlogging mode, if the stored function or trigger plans to update two different tables both having auto_increment columns, we switch to row-based for the whole function. We don't have a simple solution for statement-based binlogging mode, there the bug remains and will be documented as a known problem. Re-enabling rpl_switch_stm_row_mixed. 2) Fix for BUG#20630 "Mixed binlogging mode does not work with stored functions, triggers, views", which was a documented limitation (in mixed mode, we didn't detect that a stored function's execution needed row-based binlogging (due to some UUID() call for example); same for triggers, same for views (a view created from a SELECT UUID(), and doing INSERT INTO sometable SELECT theview; would not replicate row-based). This is implemented by, after parsing a routine's body, remembering in sp_head that this routine needs row-based binlogging. Then when this routine is used, the caller is marked to require row-based binlogging too. Same for views: when we parse a view and detect that its SELECT needs row-based binary logging, we mark the calling LEX as such. 3) Fix for BUG#20499 "mixed mode with temporary table breaks binlog": a temporary table containing e.g. UUID has its changes not binlogged, so any query updating a permanent table with data from the temporary table will run wrongly on slave. Solution: in mixed mode we don't switch back from row-based to statement-based when there exists temporary tables. 4) Attempt to test mysqlbinlog on a binlog generated by mysqlbinlog; impossible due to BUG#11312 and BUG#20329, but test is in place for when they are fixed.
2006-07-09 17:00:47 +02:00
NOTES:
Call this function only when you have established the list of all tables
which you'll want to update (including stored functions, triggers, views
inside your statement).
*/
static bool
has_write_table_with_auto_increment(TABLE_LIST *tables)
* Mixed replication mode * : 1) Fix for BUG#19630 "stored function inserting into two auto_increment breaks statement-based binlog": a stored function inserting into two such tables may fail to replicate (inserting wrong data in the slave's copy of the second table) if the slave's second table had an internal auto_increment counter different from master's. Because the auto_increment value autogenerated by master for the 2nd table does not go into binlog, only the first does, so the slave lacks information. To fix this, if running in mixed binlogging mode, if the stored function or trigger plans to update two different tables both having auto_increment columns, we switch to row-based for the whole function. We don't have a simple solution for statement-based binlogging mode, there the bug remains and will be documented as a known problem. Re-enabling rpl_switch_stm_row_mixed. 2) Fix for BUG#20630 "Mixed binlogging mode does not work with stored functions, triggers, views", which was a documented limitation (in mixed mode, we didn't detect that a stored function's execution needed row-based binlogging (due to some UUID() call for example); same for triggers, same for views (a view created from a SELECT UUID(), and doing INSERT INTO sometable SELECT theview; would not replicate row-based). This is implemented by, after parsing a routine's body, remembering in sp_head that this routine needs row-based binlogging. Then when this routine is used, the caller is marked to require row-based binlogging too. Same for views: when we parse a view and detect that its SELECT needs row-based binary logging, we mark the calling LEX as such. 3) Fix for BUG#20499 "mixed mode with temporary table breaks binlog": a temporary table containing e.g. UUID has its changes not binlogged, so any query updating a permanent table with data from the temporary table will run wrongly on slave. Solution: in mixed mode we don't switch back from row-based to statement-based when there exists temporary tables. 4) Attempt to test mysqlbinlog on a binlog generated by mysqlbinlog; impossible due to BUG#11312 and BUG#20329, but test is in place for when they are fixed.
2006-07-09 17:00:47 +02:00
{
for (TABLE_LIST *table= tables; table; table= table->next_global)
{
/* we must do preliminary checks as table->table may be NULL */
Fix for: Bug #20662 "Infinite loop in CREATE TABLE IF NOT EXISTS ... SELECT with locked tables" Bug #20903 "Crash when using CREATE TABLE .. SELECT and triggers" Bug #24738 "CREATE TABLE ... SELECT is not isolated properly" Bug #24508 "Inconsistent results of CREATE TABLE ... SELECT when temporary table exists" Deadlock occured when one tried to execute CREATE TABLE IF NOT EXISTS ... SELECT statement under LOCK TABLES which held read lock on target table. Attempt to execute the same statement for already existing target table with triggers caused server crashes. Also concurrent execution of CREATE TABLE ... SELECT statement and other statements involving target table suffered from various races (some of which might've led to deadlocks). Finally, attempt to execute CREATE TABLE ... SELECT in case when a temporary table with same name was already present led to the insertion of data into this temporary table and creation of empty non-temporary table. All above problems stemmed from the old implementation of CREATE TABLE ... SELECT in which we created, opened and locked target table without any special protection in a separate step and not with the rest of tables used by this statement. This underminded deadlock-avoidance approach used in server and created window for races. It also excluded target table from prelocking causing problems with trigger execution. The patch solves these problems by implementing new approach to handling of CREATE TABLE ... SELECT for base tables. We try to open and lock table to be created at the same time as the rest of tables used by this statement. If such table does not exist at this moment we create and place in the table cache special placeholder for it which prevents its creation or any other usage by other threads. We still use old approach for creation of temporary tables. Note that we have separate fix for 5.0 since there we use slightly different less intrusive approach.
2007-05-11 19:51:03 +02:00
if (!table->placeholder() &&
* Mixed replication mode * : 1) Fix for BUG#19630 "stored function inserting into two auto_increment breaks statement-based binlog": a stored function inserting into two such tables may fail to replicate (inserting wrong data in the slave's copy of the second table) if the slave's second table had an internal auto_increment counter different from master's. Because the auto_increment value autogenerated by master for the 2nd table does not go into binlog, only the first does, so the slave lacks information. To fix this, if running in mixed binlogging mode, if the stored function or trigger plans to update two different tables both having auto_increment columns, we switch to row-based for the whole function. We don't have a simple solution for statement-based binlogging mode, there the bug remains and will be documented as a known problem. Re-enabling rpl_switch_stm_row_mixed. 2) Fix for BUG#20630 "Mixed binlogging mode does not work with stored functions, triggers, views", which was a documented limitation (in mixed mode, we didn't detect that a stored function's execution needed row-based binlogging (due to some UUID() call for example); same for triggers, same for views (a view created from a SELECT UUID(), and doing INSERT INTO sometable SELECT theview; would not replicate row-based). This is implemented by, after parsing a routine's body, remembering in sp_head that this routine needs row-based binlogging. Then when this routine is used, the caller is marked to require row-based binlogging too. Same for views: when we parse a view and detect that its SELECT needs row-based binary logging, we mark the calling LEX as such. 3) Fix for BUG#20499 "mixed mode with temporary table breaks binlog": a temporary table containing e.g. UUID has its changes not binlogged, so any query updating a permanent table with data from the temporary table will run wrongly on slave. Solution: in mixed mode we don't switch back from row-based to statement-based when there exists temporary tables. 4) Attempt to test mysqlbinlog on a binlog generated by mysqlbinlog; impossible due to BUG#11312 and BUG#20329, but test is in place for when they are fixed.
2006-07-09 17:00:47 +02:00
table->table->found_next_number_field &&
(table->lock_type >= TL_WRITE_ALLOW_WRITE))
return 1;
* Mixed replication mode * : 1) Fix for BUG#19630 "stored function inserting into two auto_increment breaks statement-based binlog": a stored function inserting into two such tables may fail to replicate (inserting wrong data in the slave's copy of the second table) if the slave's second table had an internal auto_increment counter different from master's. Because the auto_increment value autogenerated by master for the 2nd table does not go into binlog, only the first does, so the slave lacks information. To fix this, if running in mixed binlogging mode, if the stored function or trigger plans to update two different tables both having auto_increment columns, we switch to row-based for the whole function. We don't have a simple solution for statement-based binlogging mode, there the bug remains and will be documented as a known problem. Re-enabling rpl_switch_stm_row_mixed. 2) Fix for BUG#20630 "Mixed binlogging mode does not work with stored functions, triggers, views", which was a documented limitation (in mixed mode, we didn't detect that a stored function's execution needed row-based binlogging (due to some UUID() call for example); same for triggers, same for views (a view created from a SELECT UUID(), and doing INSERT INTO sometable SELECT theview; would not replicate row-based). This is implemented by, after parsing a routine's body, remembering in sp_head that this routine needs row-based binlogging. Then when this routine is used, the caller is marked to require row-based binlogging too. Same for views: when we parse a view and detect that its SELECT needs row-based binary logging, we mark the calling LEX as such. 3) Fix for BUG#20499 "mixed mode with temporary table breaks binlog": a temporary table containing e.g. UUID has its changes not binlogged, so any query updating a permanent table with data from the temporary table will run wrongly on slave. Solution: in mixed mode we don't switch back from row-based to statement-based when there exists temporary tables. 4) Attempt to test mysqlbinlog on a binlog generated by mysqlbinlog; impossible due to BUG#11312 and BUG#20329, but test is in place for when they are fixed.
2006-07-09 17:00:47 +02:00
}
* Mixed replication mode * : 1) Fix for BUG#19630 "stored function inserting into two auto_increment breaks statement-based binlog": a stored function inserting into two such tables may fail to replicate (inserting wrong data in the slave's copy of the second table) if the slave's second table had an internal auto_increment counter different from master's. Because the auto_increment value autogenerated by master for the 2nd table does not go into binlog, only the first does, so the slave lacks information. To fix this, if running in mixed binlogging mode, if the stored function or trigger plans to update two different tables both having auto_increment columns, we switch to row-based for the whole function. We don't have a simple solution for statement-based binlogging mode, there the bug remains and will be documented as a known problem. Re-enabling rpl_switch_stm_row_mixed. 2) Fix for BUG#20630 "Mixed binlogging mode does not work with stored functions, triggers, views", which was a documented limitation (in mixed mode, we didn't detect that a stored function's execution needed row-based binlogging (due to some UUID() call for example); same for triggers, same for views (a view created from a SELECT UUID(), and doing INSERT INTO sometable SELECT theview; would not replicate row-based). This is implemented by, after parsing a routine's body, remembering in sp_head that this routine needs row-based binlogging. Then when this routine is used, the caller is marked to require row-based binlogging too. Same for views: when we parse a view and detect that its SELECT needs row-based binary logging, we mark the calling LEX as such. 3) Fix for BUG#20499 "mixed mode with temporary table breaks binlog": a temporary table containing e.g. UUID has its changes not binlogged, so any query updating a permanent table with data from the temporary table will run wrongly on slave. Solution: in mixed mode we don't switch back from row-based to statement-based when there exists temporary tables. 4) Attempt to test mysqlbinlog on a binlog generated by mysqlbinlog; impossible due to BUG#11312 and BUG#20329, but test is in place for when they are fixed.
2006-07-09 17:00:47 +02:00
return 0;
}
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
/*
Open and lock system tables for read.
SYNOPSIS
open_system_tables_for_read()
thd Thread context.
table_list List of tables to open.
backup Pointer to Open_tables_state instance where
information about currently open tables will be
saved, and from which will be restored when we will
end work with system tables.
NOTES
Thanks to restrictions which we put on opening and locking of
system tables for writing, we can open and lock them for reading
even when we already have some other tables open and locked. One
must call close_system_tables() to close systems tables opened
with this call.
RETURN
FALSE Success
TRUE Error
*/
bool
open_system_tables_for_read(THD *thd, TABLE_LIST *table_list,
Open_tables_backup *backup)
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
{
Query_tables_list query_tables_list_backup;
LEX *lex= thd->lex;
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
DBUG_ENTER("open_system_tables_for_read");
/*
Besides using new Open_tables_state for opening system tables,
we also have to backup and reset/and then restore part of LEX
which is accessed by open_tables() in order to determine if
prelocking is needed and what tables should be added for it.
close_system_tables() doesn't require such treatment.
*/
lex->reset_n_backup_query_tables_list(&query_tables_list_backup);
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
thd->reset_n_backup_open_tables_state(backup);
if (open_and_lock_tables(thd, table_list, FALSE,
MYSQL_OPEN_IGNORE_FLUSH |
MYSQL_LOCK_IGNORE_TIMEOUT))
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
{
lex->restore_backup_query_tables_list(&query_tables_list_backup);
thd->restore_backup_open_tables_state(backup);
DBUG_RETURN(TRUE);
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
}
for (TABLE_LIST *tables= table_list; tables; tables= tables->next_global)
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
{
DBUG_ASSERT(tables->table->s->table_category == TABLE_CATEGORY_SYSTEM);
tables->table->use_all_columns();
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
}
lex->restore_backup_query_tables_list(&query_tables_list_backup);
DBUG_RETURN(FALSE);
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
}
/*
Close system tables, opened with open_system_tables_for_read().
SYNOPSIS
close_system_tables()
thd Thread context
backup Pointer to Open_tables_backup instance which holds
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
information about tables which were open before we
decided to access system tables.
*/
void
close_system_tables(THD *thd, Open_tables_backup *backup)
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
{
close_thread_tables(thd);
thd->restore_backup_open_tables_state(backup);
}
/**
A helper function to close a mysql.* table opened
in an auxiliary THD during bootstrap or in the main
connection, when we know that there are no locks
held by the connection due to a preceding implicit
commit.
This function assumes that there is no
statement transaction started for the operation
itself, since mysql.* tables are not transactional
and when they are used the binlog is off (DDL
binlogging is always statement-based.
We need this function since we'd like to not
just close the system table, but also release
the metadata lock on it.
Note, that in LOCK TABLES mode this function
does not release the metadata lock. But in this
mode the table can be opened only if it is locked
explicitly with LOCK TABLES.
*/
void
close_mysql_tables(THD *thd)
{
/* No need to commit/rollback statement transaction, it's not started. */
DBUG_ASSERT(thd->transaction.stmt.is_empty());
close_thread_tables(thd);
thd->mdl_context.release_transactional_locks();
}
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
/*
Open and lock one system table for update.
SYNOPSIS
open_system_table_for_update()
thd Thread context.
one_table Table to open.
NOTES
Table opened with this call should closed using close_thread_tables().
RETURN
0 Error
# Pointer to TABLE object of system table
*/
TABLE *
open_system_table_for_update(THD *thd, TABLE_LIST *one_table)
{
DBUG_ENTER("open_system_table_for_update");
TABLE *table= open_ltable(thd, one_table, one_table->lock_type,
MYSQL_LOCK_IGNORE_TIMEOUT);
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
if (table)
{
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
DBUG_ASSERT(table->s->table_category == TABLE_CATEGORY_SYSTEM);
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
table->use_all_columns();
}
DBUG_RETURN(table);
}
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
/**
Open a log table.
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
Opening such tables is performed internally in the server
implementation, and is a 'nested' open, since some tables
might be already opened by the current thread.
The thread context before this call is saved, and is restored
when calling close_log_table().
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
@param thd The current thread
@param one_table Log table to open
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
@param backup [out] Temporary storage used to save the thread context
*/
TABLE *
2010-02-03 14:43:03 +01:00
open_log_table(THD *thd, TABLE_LIST *one_table, Open_tables_backup *backup)
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
{
uint flags= ( MYSQL_OPEN_IGNORE_GLOBAL_READ_LOCK |
MYSQL_LOCK_IGNORE_GLOBAL_READ_ONLY |
MYSQL_OPEN_IGNORE_FLUSH |
MYSQL_LOCK_IGNORE_TIMEOUT |
MYSQL_LOCK_LOG_TABLE);
TABLE *table;
/* Save value that is changed in mysql_lock_tables() */
ulonglong save_utime_after_lock= thd->utime_after_lock;
DBUG_ENTER("open_log_table");
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
thd->reset_n_backup_open_tables_state(backup);
if ((table= open_ltable(thd, one_table, one_table->lock_type, flags)))
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
{
DBUG_ASSERT(table->s->table_category == TABLE_CATEGORY_LOG);
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
/* Make sure all columns get assigned to a default value */
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
table->use_all_columns();
table->no_replicate= 1;
/*
Don't set automatic timestamps as we may want to use time of logging,
not from query start
*/
table->timestamp_field_type= TIMESTAMP_NO_AUTO_SET;
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
}
else
thd->restore_backup_open_tables_state(backup);
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
thd->utime_after_lock= save_utime_after_lock;
BUG#9953: CONVERT_TZ requires mysql.time_zone_name to be locked The problem was that some facilities (like CONVERT_TZ() function or server HELP statement) may require implicit access to some tables in 'mysql' database. This access was done by ordinary means of adding such tables to the list of tables the query is going to open. However, if we issued LOCK TABLES before that, we would get "table was not locked" error trying to open such implicit tables. The solution is to treat certain tables as MySQL system tables, like we already do for mysql.proc. Such tables may be opened for reading at any moment regardless of any locks in effect. The cost of this is that system table may be locked for writing only together with other system tables, it is disallowed to lock system tables for writing and have any other lock on any other table. After this patch the following tables are treated as MySQL system tables: mysql.help_category mysql.help_keyword mysql.help_relation mysql.help_topic mysql.proc (it already was) mysql.time_zone mysql.time_zone_leap_second mysql.time_zone_name mysql.time_zone_transition mysql.time_zone_transition_type These tables are now opened with open_system_tables_for_read() and closed with close_system_tables(), or one table may be opened with open_system_table_for_update() and closed with close_thread_tables() (the latter is used for mysql.proc table, which is updated as part of normal MySQL server operation). These functions may be used when some tables were opened and locked already. NOTE: online update of time zone tables is not possible during replication, because there's no time zone cache flush neither on LOCK TABLES, nor on FLUSH TABLES, so the master may serve stale time zone data from cache, while on slave updated data will be loaded from the time zone tables.
2007-03-09 11:12:31 +01:00
DBUG_RETURN(table);
}
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
/**
Close a log table.
The last table opened by open_log_table()
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
is closed, then the thread context is restored.
@param thd The current thread
@param backup [in] the context to restore.
*/
2010-02-03 14:43:03 +01:00
void close_log_table(THD *thd, Open_tables_backup *backup)
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
{
close_system_tables(thd, backup);
WL#3984 (Revise locking of mysql.general_log and mysql.slow_log) Bug#25422 (Hang with log tables) Bug 17876 (Truncating mysql.slow_log in a SP after using cursor locks the thread) Bug 23044 (Warnings on flush of a log table) Bug 29129 (Resetting general_log while the GLOBAL READ LOCK is set causes a deadlock) Prior to this fix, the server would hang when performing concurrent ALTER TABLE or TRUNCATE TABLE statements against the LOG TABLES, which are mysql.general_log and mysql.slow_log. The root cause traces to the following code: in sql_base.cc, open_table() if (table->in_use != thd) { /* wait_for_condition will unlock LOCK_open for us */ wait_for_condition(thd, &LOCK_open, &COND_refresh); } The problem with this code is that the current implementation of the LOGGER creates 'fake' THD objects, like - Log_to_csv_event_handler::general_log_thd - Log_to_csv_event_handler::slow_log_thd which are not associated to a real thread running in the server, so that waiting for these non-existing threads to release table locks cause the dead lock. In general, the design of Log_to_csv_event_handler does not fit into the general architecture of the server, so that the concept of general_log_thd and slow_log_thd has to be abandoned: - this implementation does not work with table locking - it will not work with commands like SHOW PROCESSLIST - having the log tables always opened does not integrate well with DDL operations / FLUSH TABLES / SET GLOBAL READ_ONLY With this patch, the fundamental design of the LOGGER has been changed to: - always open and close a log table when writing a log - remove totally the usage of fake THD objects - clarify how locking of log tables is implemented in general. See WL#3984 for details related to the new locking design. Additional changes (misc bugs exposed and fixed): 1) mysqldump which would ignore some tables in dump_all_tables_in_db(), but forget to ignore the same in dump_all_views_in_db(). 2) mysqldump would also issue an empty "LOCK TABLE" command when all the tables to lock are to be ignored (numrows == 0), instead of not issuing the query. 3) Internal errors handlers could intercept errors but not warnings (see sql_error.cc). 4) Implementing a nested call to open tables, for the performance schema tables, exposed an existing bug in remove_table_from_cache(), which would perform: in_use->some_tables_deleted=1; against another thread, without any consideration about thread locking. This call inside remove_table_from_cache() was not required anyway, since calling mysql_lock_abort() takes care of aborting -- cleanly -- threads that might hold a lock on a table. This line (in_use->some_tables_deleted=1) has been removed.
2007-07-27 08:31:06 +02:00
}
/**
@} (end of group Data_Dictionary)
*/