Invaldating a subset of a sufficiently large query cache can take a long time.
During this time the server is efficiently frozen and no other operation can
be executed. This patch addresses this problem by moving the locks which cause
the freezing and also by temporarily disable the query cache while the
invalidation takes place.
The log tables are by nature PERFORMANCE_SCHEMA tables,
which should not be affected by SET GLOBAL READ_ONLY or FLUSH TABLES
WITH READ LOCK.
The implementation of FLUSH TABLES WITH READ LOCK already ignored log tables.
Now with this patch, the implementation of LOCK TABLE also ignore a
global read lock for log tables, which was the missing symmetry.
corruption errors: 126,134,145
When one thread attempts to lock two (or more) tables and another
thread executes statement that aborts these locks (e.g. REPAIR
TABLE) we may get a table object with wrong lock type in a table
cache.
For example if SELECT FROM t1,t2 was aborted, subsequent INSERT
INTO t1 may be executed under read lock.
As a result we may get various table corruptions and even a server
crash.
This is fixed by resetting lock type in case lock was aborted by
another thread.
I failed to create reasonable test case for this bug.
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.
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.
Also note that we decided to postpone introduction of some tests
for concurrent behaviour of CREATE TABLE ... SELECT till 5.1.
The main reason for this is absence in 5.0 ability to set @@debug
variable at runtime, which can be circumvented only by using several
test files with individual .opt files. Since the latter is likely
to slowdown test-suite unnecessary we chose not to push this tests
into 5.0, but run them manually for this version and later push
their optimized version into 5.1
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
what it actually means (Monty approved the renaming)
- correcting description of transaction_alloc command-line options
(our manual is correct)
- fix for a failure of rpl_trigger.
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.
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.
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
Corrected spelling in copyright text
Makefile.am:
Don't update the files from BitKeeper
Many files:
Removed "MySQL Finland AB & TCX DataKonsult AB" from copyright header
Adjusted year(s) in copyright header
Many files:
Added GPL copyright text
Removed files:
Docs/Support/colspec-fix.pl
Docs/Support/docbook-fixup.pl
Docs/Support/docbook-prefix.pl
Docs/Support/docbook-split
Docs/Support/make-docbook
Docs/Support/make-makefile
Docs/Support/test-make-manual
Docs/Support/test-make-manual-de
Docs/Support/xwf
An update that used a join of a table to itself and modified the
table on one side of the join reported the table as crashed or
updated wrong rows.
Fixed by creating temporary table for self-joined multi update statement.
Bug#11733 (COMMITs should not happen if read-only is set)
Bug#22009 (Can write to a read-only server under some circumstances)
See the work log for details
The change consist of
a) acquiring the global read lock in SET GLOBAL READONLY
b) honoring opt_readonly in ha_commit_trans(),
c) honoring opt_readonly in mysql_lock_tables().
a) takes care of the server stability,
b) makes the transactional tables safe (Bug 11733)
c) makes the non transactional tables safe (Bug 22009)
Bug #21785 "Server crashes after rename of the log table" and
Bug #21966 "Strange warnings on create like/repair of the log
tables"
According to the patch, from now on, one should use RENAME to
perform a log table rotation (this should also be reflected in
the manual).
Here is a sample:
use mysql;
CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log;
RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log;
The rules for Rename of the log tables are following:
IF 1. Log tables are enabled
AND 2. Rename operates on the log table and nothing is being
renamed to the log table.
DO 3. Throw an error message.
ELSE 4. Perform rename.
The very RENAME query will go the the old (backup) table. This is
consistent with the behavoiur we have with binlog ROTATE LOGS
statement.
Other problems, which are solved by the patch are:
1) Now REPAIR of the log table is exclusive operation (as it should be), this
also eliminates lock-related warnings. and
2) CREATE LIKE TABLE now usese usual read lock on the source table rather
then name lock, which is too restrictive. This way we get rid of another
log table-related warning, which occured because of the above fact
(as a side-effect, name lock resulted in a warning).
from cache" and #21216 "Simultaneous DROP TABLE and SHOW OPEN TABLES causes
server to crash".
Crash happened when one ran DROP DATABASE or SHOW OPEN TABLES statements
while concurrently doing DROP TABLE (or RENAME TABLE, CREATE TABLE LIKE
or any other command that takes name-lock) in other connection.
This problem was caused by the fact that table placeholders which were
added to table cache in order to obtain name-lock on table had
TABLE_SHARE::db and table_name set to 0. Therefore they broke assumption
that these members are non-0 for all tables in table cache on which some
of our code relies.
The fix sets these members for such placeholders to appropriate value making
this assumption true again. As attempt to avoid such problems in future
we introduce auxiliary TABLE_SHARE::set_table_cache_key() methods which
should be used when one wants to set TABLE_SHARE::table_cache_key and which
ensure that TABLE_SHARE::table_name/db are set properly.
Test cases for these bugs were added to 5.0 test-suite (with 5.0-specific
fix for bug #21216).
server to crash".
Crash caused by assertion failure happened when one ran SHOW OPEN TABLES
while concurrently doing DROP TABLE (or RENAME TABLE, CREATE TABLE LIKE
or any other command that takes name-lock) in other connection.
For non-debug version of server problem exposed itself as wrong output
of SHOW OPEN TABLES statement (it was missing name-locked tables).
Finally in 5.1 both debug and non-debug versions simply crashed in
this situation due to NULL-pointer dereference.
This problem was caused by the fact that table placeholders which were
added to table cache in order to obtain name-lock had TABLE_SHARE::table_name
set to 0. Therefore they broke assumption that this member is non-0 for
all tables in table cache which was checked by assert in list_open_tables()
(in 5.1 this function simply relies on it).
The fix simply sets this member for such placeholders to appropriate value
making this assumption true again.
This patch also includes test for similar bug 12212 "Crash that happens
during removing of database name from cache" reappeared in 5.1 as bug 19403.
Addendum fixes after changing the condition variable
for the global read lock.
The stress test suite revealed some deadlocks. Some were
related to the new condition variable (COND_global_read_lock)
and some were general problems with the global read lock.
It is now necessary to signal COND_global_read_lock whenever
COND_refresh is signalled.
We need to wait for the release of a global read lock if one
is set before every operation that requires a write lock.
But we must not wait if we have locked tables by LOCK TABLES.
After setting a global read lock a thread waits until all
write locks are released.
Replaced COND_refresh with COND_global_read_lock becasue of a bug in NTPL threads when using different mutexes as arguments to pthread_cond_wait()
The original code caused a hang in FLUSH TABLES WITH READ LOCK in some circumstances because pthread_cond_broadcast() was not delivered to other threads.
This fixes:
Bug#16986: Deadlock condition with MyISAM tables
Bug#20048: FLUSH TABLES WITH READ LOCK causes a deadlock
After a locking error the open table(s) were not fully
cleaned up for reuse. But they were put into the open table
cache even before the lock was tried. The next statement
reused the table(s) with a wrong lock type set up. This
tricked MyISAM into believing that it don't need to update
the table statistics. Hence CHECK TABLE reported a mismatch
of record count and table size.
Fortunately nothing worse has been detected yet. The effect
of the test case was that the insert worked on a read locked
table. (!)
I added a new function that clears the lock type from all
tables that were prepared for a lock. I call this function
when a lock failes.
No test case. One test would add 50 seconds to the
test suite. Another test requires file mode modifications.
I added a test script to the bug report. It contains three
cases for failing locks. All could reproduce a table
corruption. All are fixed by this patch.
This bug was not lock timeout specific.
Additional fix for INSERT DELAYED with subselect.
Originally detected in 5.1, but 5.0 could also be affected.
The user thread creates a dummy table object,
which is not included in the lock. The 'real' table is
opened and locked by the 'delayed' system thread.
The dummy object is now marked as not locked and this is
tested in mysql_lock_have_duplicate().
Added missing DBUG_xxx_RETURN statements
Fixed some usage of not initialized variables (as found by valgrind)
Ensure that we don't remove locked tables used as name locks from open table cache until unlock_table_names() are called.
This was fixed by having drop_locked_name() returning any table used as a name lock so that we can free it in unlock_table_names()
This will allow Tomas to continue with his work to use namelocks to syncronize things.
Note: valgrind still produces a lot of warnings about using not initialized code and shows memory loss errors when running the ndb tests
