NOTE: This undoes changes by BUG#42829 in sql_class.cc:binlog_query().
I will revert the change in a post-push fix (the binlog filter should
be checked in sql_base.cc:decide_logging_format()).
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.
without error
When using quick access methods for searching rows in UPDATE or
DELETE there was no check if a fatal error was not already sent
to the client while evaluating the quick condition.
As a result a false OK (following the error) was sent to the
client and the error was thus transformed into a warning.
Fixed by checking for errors sent to the client during
SQL_SELECT::check_quick() and treating them as real errors.
Fixed a wrong test case in group_min_max.test
Fixed a wrong return code in mysql_update() and mysql_delete()
mutually-nested subqueries
Queries of the form
SELECT * FROM (SELECT 1) AS t1,
(SELECT 2) AS t2,...
(SELECT 32) AS t32
caused the "Too high level of nesting for select" error
as if the query has a form
SELECT * FROM (SELECT 1 FROM (SELECT 2 FROM (SELECT 3 FROM...
The table_factor parser rule has been modified to adjust
the LEX::nest_level variable value after every derived table.
sort_buffer_size cannot allocate
The NULL return from tree_insert() (on low memory) was not
checked for in Item_func_group_concat::add(). As a result
on low memory conditions a crash happens.
Fixed by properly checking the return code.
1. BUG#21704 - Renaming column does not update FK definition
2. Changes in mysql-test/include/mtr_warnings.sql so that the testcase
for BUG#21704 doesn't fail because of the warnings generated.
Detailed revision comments:
r5488 | vasil | 2009-07-09 19:16:44 +0300 (Thu, 09 Jul 2009) | 13 lines
branches/5.1:
Fix Bug#21704 Renaming column does not update FK definition
by checking whether a column that participates in a FK definition is being
renamed and denying the ALTER in this case.
The patch was originally developed by Davi Arnaut <Davi.Arnaut@Sun.COM>:
http://lists.mysql.com/commits/77714
and was later adjusted to conform to InnoDB coding style by me (Vasil),
I also added some more comments and moved the bug specific mysql-test to
a separate file to make it more manageable and flexible.
and base tables
myrg_attach_children() could reuse a buffer that was allocated
previously based on a definition of a child table. The problem
was that the child's definition might have been changed, so
reusing the buffer could lead to crashes or valgrind errors
under some circumstances.
Fixed by changing myrg_attach_children() so that the
rec_per_key_part buffer is reused only when the child table
have not changed, and reallocated otherwise (the old buffer is
deallocated if necessary).
BUG#45749 - Race condition in SET GLOBAL innodb_commit_concurrency=DEFAULT
Detailed revision comments:
r5419 | marko | 2009-06-25 16:11:57 +0300 (Thu, 25 Jun 2009) | 18 lines
branches/5.1: Merge r5418 from branches/zip:
------------------------------------------------------------------------
r5418 | marko | 2009-06-25 15:55:52 +0300 (Thu, 25 Jun 2009) | 5 lines
Changed paths:
M /branches/zip/ChangeLog
M /branches/zip/handler/ha_innodb.cc
M /branches/zip/mysql-test/innodb_bug42101-nonzero.result
M /branches/zip/mysql-test/innodb_bug42101-nonzero.test
M /branches/zip/mysql-test/innodb_bug42101.result
M /branches/zip/mysql-test/innodb_bug42101.test
branches/zip: Fix a race condition caused by
SET GLOBAL innodb_commit_concurrency=DEFAULT. (Bug #45749)
When innodb_commit_concurrency is initially set nonzero,
DEFAULT would change it back to 0, triggering Bug #42101.
rb://139 approved by Heikki Tuuri.
------------------------------------------------------------------------
use partial primary key if another index can prevent filesort
The fix for bug #28404 causes the covering ordering indexes to be
preferred unconditionally over non-covering and ref indexes.
Fixed by comparing the cost of using a covering index to the cost of
using a ref index even for covering ordering indexes.
Added an assertion to clarify the condition the local variables should
be in.
The crash happend because for views which are joins
we have table_list->table == 0 and
table_list->table->'any method' call leads to crash.
The fix is to perform table_list->table->file->extra()
method for all tables belonging to view.
Using DECIMAL constants with more than 65 digits in CREATE
TABLE ... SELECT led to bogus errors in release builds or
assertion failures in debug builds.
The problem was in inconsistency in how DECIMAL constants and
fields are handled internally. We allow arbitrarily long
DECIMAL constants, whereas DECIMAL(M,D) columns are limited to
M<=65 and D<=30. my_decimal_precision_to_length() was used in
both Item and Field code and truncated precision to
DECIMAL_MAX_PRECISION when calculating value length without
adjusting precision and decimals. As a result, a DECIMAL
constant with more than 65 digits ended up having length less
than precision or decimals which led to assertion failures.
Fixed by modifying my_decimal_precision_to_length() so that
precision is truncated to DECIMAL_MAX_PRECISION only for Field
object which is indicated by the new 'truncate' parameter.
Another inconsistency fixed by this patch is how DECIMAL
constants and expressions are handled for CREATE ... SELECT.
create_tmp_field_from_item() (which is used for constants) was
changed as a part of the bugfix for bug #24907 to handle long
DECIMAL constants gracefully. Item_func::tmp_table_field()
(which is used for expressions) on the other hand was still
using a simplistic approach when creating a Field_new_decimal
from a DECIMAL expression.
contains ONLY_FULL_GROUP_BY
The partitioning code needs to issue a Item::fix_fields()
on the partitioning expression in order to prepare
it for being evaluated.
It does this by creating a special table and a table list
for the scope of the partitioning expression.
But when checking ONLY_FULL_GROUP_BY the
Item_field::fix_fields() was relying that there always be
cached_table set and was trying to use it to get the
select_lex of the SELECT the field's table is in.
But the cached_table was not set by the partitioning code
that creates the artificial TABLE_LIST used to resolve the
partitioning expression and this resulted in a crash.
Fixed by rectifying the following errors :
1. Item_field::fix_fields() : the code that check for
ONLY_FULL_GROUP_BY relies on having tables with
cacheable_table set. This is mostly true, the only
two exceptions being the partitioning context table
and the trigger context table.
Fixed by taking the current parsing context if no pointer
to the TABLE_LIST instance is present in the cached_table.
2. fix_fields_part_func() :
2a. The code that adds the table being created to the
scope for the partitioning expression is mostly a copy
of the add_table_to_list and friends with one exception :
it was not marking the table as cacheable (something that
normal add_table_to_list is doing). This caused the
problem in the check for ONLY_FULL_GROUP_BY in
Item_field::fix_fields() to appear.
Fixed by setting the correct members to make the table
cacheable.
The ideal structural fix for this is to use a unified
interface for adding a table to a table list
(add_table_to_list?) : noted in a TODO comment
2b. The Item::fix_fields() was called with a NULL destination
pointer. This causes uninitalized memory reads in the
overloaded ::fix_fields() function (namely
Item_field::fix_fields()) as it expects a non-zero pointer
there. Fixed by passing the source pointer similarly to how
it's done in JOIN::prepare().
The TABLE::reginfo.impossible_range is used by the optimizer to indicate
that the condition applied to the table is impossible. It wasn't initialized
at table opening and this might lead to an empty result on complex queries:
a query might set the impossible_range flag on a table and when the query finishes,
all tables are returned back to the table cache. The next query that uses the table
with the impossible_range flag set and an index over the table will see the flag
and thus return an empty result.
The open_table function now initializes the TABLE::reginfo.impossible_range
variable.
The problem is that the one phase commit function failed to
properly end a empty transaction. The solution is to ensure
that the transaction cleanup procedure is invoked even for
empty transactions.
BUG#40565 - Update Query Results in "1 Row Affected" But Should Be "Zero Rows"
Detailed revision comments:
r5232 | marko | 2009-06-03 14:31:04 +0300 (Wed, 03 Jun 2009) | 21 lines
branches/5.0: Merge r3590 from branches/5.1 in order to fix Bug #40565
(Update Query Results in "1 Row Affected" But Should Be "Zero Rows").
Also, add a test case for Bug #40565.
rb://128 approved by Heikki Tuuri
------------------------------------------------------------------------
r3590 | marko | 2008-12-18 15:33:36 +0200 (Thu, 18 Dec 2008) | 11 lines
branches/5.1: When converting a record to MySQL format, copy the default
column values for columns that are SQL NULL. This addresses failures in
row-based replication (Bug #39648).
row_prebuilt_t: Add default_rec, for the default values of the columns in
MySQL format.
row_sel_store_mysql_rec(): Use prebuilt->default_rec instead of
padding columns.
rb://64 approved by Heikki Tuuri
------------------------------------------------------------------------
In Item_param::set_from_user_var
value.cs_info.character_set_client is set
to 'fromcs' value. It's wrong, it should be set to
thd->variables.character_set_client.
When opening a table, it is imperative that the flag
TABLE::auto_increment_field_not_null be false. But if an error occured during
the creation of a table (e.g. the table exists already) with an auto_increment
column and a BEFORE trigger that used the INSERT ... SELECT construct, the
flag was not reset until after error checking. Thus if an error occured,
select_insert::send_data() returned immediately and it was not reset (see * in
pseudocode below). Crash happened if the table was opened again. Fixed by
resetting the flag after error checking.
nested-loops_join():
for each row in SELECT table {
select_insert::send_data():
if a values is supplied for AUTO_INCREMENT column
table->auto_increment_field_not_null= TRUE
else
table->auto_increment_field_not_null= FALSE
if (error)
return 1; *
if (table->auto_increment_field_not_null == FALSE)
...
table->auto_increment_field_not_null == FALSE
}
<-- table returned to table cache and later retrieved by open_table:
open_table():
assert(table->auto_increment_field_not_null)
1. BUG#45357 - 5.1.35 crashes with Failing assertion: index->type & DICT_CLUSTERED
2. Also fixes the compilation problem when the flag -DUNIV_MUST_NOT_INLINE
Detailed revision comments:
r5340 | marko | 2009-06-17 12:11:49 +0300 (Wed, 17 Jun 2009) | 4 lines
branches/5.1: row_unlock_for_mysql(): When the clustered index is unknown,
refuse to unlock the record.
(Bug #45357, caused by the fix of Bug #39320).
rb://132 approved by Sunny Bains.
r5339 | marko | 2009-06-17 11:01:37 +0300 (Wed, 17 Jun 2009) | 2 lines
branches/5.1: Add missing #include "mtr0log.h" so that the code compiles
with -DUNIV_MUST_NOT_INLINE.
Inconsistent behavior of session variable max_allowed_packet
(and net_buffer_length); only assignment to the global variable
has any effect, without this being obvious to the user.
The patch for Bug#22891 is backported to 5.0, making the two
session variables read-only. As this is a backport to GA
software, the error used when trying to assign to the read-
only variable is ER_UNKNOWN_ERROR. The error message is the
same as in 5.1+.
Large transactions and statements may corrupt the binary log if the size of the
cache, which is set by the max_binlog_cache_size, is not enough to store the
the changes.
In a nutshell, to fix the bug, we save the position of the next character in the
cache before starting processing a statement. If there is a problem, we simply
restore the position thus removing any effect of the statement from the cache.
Unfortunately, to avoid corrupting the binary log, we may end up loosing changes
on non-transactional tables if they do not fit in the cache. In such cases, we
store an Incident_log_event in order to stop the slave and alert users that some
changes were not logged.
Precisely, for every non-transactional changes that do not fit into the cache,
we do the following:
a) the statement is *not* logged
b) an incident event is logged after committing/rolling back the transaction,
if any. Note that if a failure happens before writing the incident event to
the binary log, the slave will not stop and the master will not have reported
any error.
c) its respective statement gives an error
For transactional changes that do not fit into the cache, we do the following:
a) the statement is *not* logged
b) its respective statement gives an error
To work properly, this patch requires two additional things. Firstly, callers to
MYSQL_BIN_LOG::write and THD::binlog_query must handle any error returned and
take the appropriate actions such as undoing the effects of a statement. We
already changed some calls in the sql_insert.cc, sql_update.cc and sql_insert.cc
modules but the remaining calls spread all over the code should be handled in
BUG#37148. Secondly, statements must be either classified as DDL or DML because
DDLs that do not get into the cache must generate an incident event since they
cannot be rolled back.
