This patch also fixes Bug#55452 "SET PASSWORD is
replicated twice in RBR mode".
The goal of this patch is to remove the release of
metadata locks from close_thread_tables().
This is necessary to not mistakenly release
the locks in the course of a multi-step
operation that involves multiple close_thread_tables()
or close_tables_for_reopen().
On the same token, move statement commit outside
close_thread_tables().
Other cleanups:
Cleanup COM_FIELD_LIST.
Don't call close_thread_tables() in COM_SHUTDOWN -- there
are no open tables there that can be closed (we leave
the locked tables mode in THD destructor, and this
close_thread_tables() won't leave it anyway).
Make open_and_lock_tables() and open_and_lock_tables_derived()
call close_thread_tables() upon failure.
Remove the calls to close_thread_tables() that are now
unnecessary.
Simplify the back off condition in Open_table_context.
Streamline metadata lock handling in LOCK TABLES
implementation.
Add asserts to ensure correct life cycle of
statement transaction in a session.
Remove a piece of dead code that has also become redundant
after the fix for Bug 37521.
mysql-test/r/variables.result:
Update results: set @@autocommit and statement transaction/
prelocked mode.
mysql-test/r/view.result:
A harmless change in CHECK TABLE <view> status for a broken view.
If previously a failure to prelock all functions used in a view
would leave the connection in LTM_PRELOCKED mode, now we call
close_thread_tables() from open_and_lock_tables()
and leave prelocked mode, thus some check in mysql_admin_table() that
works only in prelocked/locked tables mode is no longer activated.
mysql-test/suite/rpl/r/rpl_row_implicit_commit_binlog.result:
Fixed Bug#55452 "SET PASSWORD is replicated twice in
RBR mode": extra binlog events are gone from the
binary log.
mysql-test/t/variables.test:
Add a test case: set autocommit and statement transaction/prelocked
mode.
sql/event_data_objects.cc:
Simplify code in Event_job_data::execute().
Move sp_head memory management to lex_end().
sql/event_db_repository.cc:
Move the release of metadata locks outside
close_thread_tables().
Make sure we call close_thread_tables() when
open_and_lock_tables() fails and remove extra
code from the events data dictionary.
Use close_mysql_tables(), a new internal
function to properly close mysql.* tables
in the data dictionary.
Contract Event_db_repository::drop_events_by_field,
drop_schema_events into one function.
When dropping all events in a schema,
make sure we don't mistakenly release all
locks acquired by DROP DATABASE. These
include locks on the database name
and the global intention exclusive
metadata lock.
sql/event_db_repository.h:
Function open_event_table() does not require an instance
of Event_db_repository.
sql/events.cc:
Use close_mysql_tables() instead of close_thread_tables()
to bootstrap events, since the latter no longer
releases metadata locks.
sql/ha_ndbcluster.cc:
- mysql_rm_table_part2 no longer releases
acquired metadata locks. Do it in the caller.
sql/ha_ndbcluster_binlog.cc:
Deploy the new protocol for closing thread
tables in run_query() and ndb_binlog_index
code.
sql/handler.cc:
Assert that we never call ha_commit_trans/
ha_rollback_trans in sub-statement, which
is now the case.
sql/handler.h:
Add an accessor to check whether THD_TRANS object
is empty (has no transaction started).
sql/log.cc:
Update a comment.
sql/log_event.cc:
Since now we commit/rollback statement transaction in
mysql_execute_command(), we need a mechanism to communicate
from Query_log_event::do_apply_event() to mysql_execute_command()
that the statement transaction should be rolled back, not committed.
Ideally it would be a virtual method of THD. I hesitate
to make THD a virtual base class in this already large patch.
Use a thd->variables.option_bits for now.
Remove a call to close_thread_tables() from the slave IO
thread. It doesn't open any tables, and the protocol
for closing thread tables is more complicated now.
Make sure we properly close thread tables, however,
in Load_data_log_event, which doesn't
follow the standard server execution procedure
with mysql_execute_command().
@todo: this piece should use Server_runnable
framework instead.
Remove an unnecessary call to mysql_unlock_tables().
sql/rpl_rli.cc:
Update Relay_log_info::slave_close_thread_tables()
to follow the new close protocol.
sql/set_var.cc:
Remove an unused header.
sql/slave.cc:
Remove an unnecessary call to
close_thread_tables().
sql/sp.cc:
Remove unnecessary calls to close_thread_tables()
from SP DDL implementation. The tables will
be closed by the caller, in mysql_execute_command().
When dropping all routines in a database, make sure
to not mistakenly drop all metadata locks acquired
so far, they include the scoped lock on the schema.
sql/sp_head.cc:
Correct the protocol that closes thread tables
in an SP instruction.
Clear lex->sphead before cleaning up lex
with lex_end to make sure that we don't
delete the sphead twice. It's considered
to be "cleaner" and more in line with
future changes than calling delete lex->sphead
in other places that cleanup the lex.
sql/sp_head.h:
When destroying m_lex_keeper of an instruction,
don't delete the sphead that all lex objects
share.
@todo: don't store a reference to routine's sp_head
instance in instruction's lex.
sql/sql_acl.cc:
Don't call close_thread_tables() where the caller will
do that for us.
Fix Bug#55452 "SET PASSWORD is replicated twice in RBR
mode" by disabling RBR replication in change_password()
function.
Use close_mysql_tables() in bootstrap and ACL reload
code to make sure we release all metadata locks.
sql/sql_base.cc:
This is the main part of the patch:
- remove manipulation with thd->transaction
and thd->mdl_context from close_thread_tables().
Now this function is only responsible for closing
tables, nothing else.
This is necessary to be able to easily use
close_thread_tables() in procedures, that
involve multiple open/close tables, which all
need to be protected continuously by metadata
locks.
Add asserts ensuring that TABLE object
is only used when is protected by a metadata lock.
Simplify the back off condition of Open_table_context,
we no longer need to look at the autocommit mode.
Make open_and_lock_tables() and open_normal_and_derived_tables()
close thread tables and release metadata locks acquired so-far
upon failure. This simplifies their usage.
Implement close_mysql_tables().
sql/sql_base.h:
Add declaration for close_mysql_tables().
sql/sql_class.cc:
Remove a piece of dead code that has also become redundant
after the fix for Bug 37521.
The code became dead when my_eof() was made a non-protocol method,
but a method that merely modifies the diagnostics area.
The code became redundant with the fix for Bug#37521, when
we started to cal close_thread_tables() before
Protocol::end_statement().
sql/sql_do.cc:
Do nothing in DO if inside a substatement
(the assert moved out of trans_rollback_stmt).
sql/sql_handler.cc:
Add comments.
sql/sql_insert.cc:
Remove dead code.
Release metadata locks explicitly at the
end of the delayed insert thread.
sql/sql_lex.cc:
Add destruction of lex->sphead to lex_end(),
lex "reset" method called at the end of each statement.
sql/sql_parse.cc:
Move close_thread_tables() and other related
cleanups to mysql_execute_command()
from dispatch_command(). This has become
possible after the fix for Bug#37521.
Mark federated SERVER statements as DDL.
Next step: make sure that we don't store
eof packet in the query cache, and move
the query cache code outside mysql_parse.
Brush up the code of COM_FIELD_LIST.
Remove unnecessary calls to close_thread_tables().
When killing a query, don't report "OK"
if it was a suicide.
sql/sql_parse.h:
Remove declaration of a function that is now static.
sql/sql_partition.cc:
Remove an unnecessary call to close_thread_tables().
sql/sql_plugin.cc:
open_and_lock_tables() will clean up
after itself after a failure.
Move close_thread_tables() above
end: label, and replace with close_mysql_tables(),
which will also release the metadata lock
on mysql.plugin.
sql/sql_prepare.cc:
Now that we no longer release locks in close_thread_tables()
statement prepare code has become more straightforward.
Remove the now redundant check for thd->killed() (used
only by the backup project) from Execute_server_runnable.
Reorder code to take into account that now mysql_execute_command()
performs lex->unit.cleanup() and close_thread_tables().
sql/sql_priv.h:
Add a new option to server options to interact
between the slave SQL thread and execution
framework (hack). @todo: use a virtual
method of class THD instead.
sql/sql_servers.cc:
Due to Bug 25705 replication of
DROP/CREATE/ALTER SERVER is broken.
Make sure at least we do not attempt to
replicate these statements using RBR,
as this violates the assert in close_mysql_tables().
sql/sql_table.cc:
Do not release metadata locks in mysql_rm_table_part2,
this is done by the caller.
Do not call close_thread_tables() in mysql_create_table(),
this is done by the caller.
Fix a bug in DROP TABLE under LOCK TABLES when,
upon error in wait_while_table_is_used() we would mistakenly
release the metadata lock on a non-dropped table.
Explicitly release metadata locks when doing an implicit
commit.
sql/sql_trigger.cc:
Now that we delete lex->sphead in lex_end(),
zero the trigger's sphead in lex after loading
the trigger, to avoid double deletion.
sql/sql_udf.cc:
Use close_mysql_tables() instead of close_thread_tables().
sql/sys_vars.cc:
Remove code added in scope of WL#4284 which would
break when we perform set @@session.autocommit along
with setting other variables and using tables or functions.
A test case added to variables.test.
sql/transaction.cc:
Add asserts.
sql/tztime.cc:
Use close_mysql_tables() rather than close_thread_tables().
This patch also fixes Bug#55452 "SET PASSWORD is
replicated twice in RBR mode".
The goal of this patch is to remove the release of
metadata locks from close_thread_tables().
This is necessary to not mistakenly release
the locks in the course of a multi-step
operation that involves multiple close_thread_tables()
or close_tables_for_reopen().
On the same token, move statement commit outside
close_thread_tables().
Other cleanups:
Cleanup COM_FIELD_LIST.
Don't call close_thread_tables() in COM_SHUTDOWN -- there
are no open tables there that can be closed (we leave
the locked tables mode in THD destructor, and this
close_thread_tables() won't leave it anyway).
Make open_and_lock_tables() and open_and_lock_tables_derived()
call close_thread_tables() upon failure.
Remove the calls to close_thread_tables() that are now
unnecessary.
Simplify the back off condition in Open_table_context.
Streamline metadata lock handling in LOCK TABLES
implementation.
Add asserts to ensure correct life cycle of
statement transaction in a session.
Remove a piece of dead code that has also become redundant
after the fix for Bug 37521.
If the expression for a virtual column of table contained datetime
comparison then the execution of the second query that used this
virtual column caused a crash. It happened because the execution
of the first query that used this virtual column inserted a cached
item into the expression tree. The cached tree was allocated in
the statement memory while the expression tree was allocated in
the table memory.
Now the cached items that are inserted into expressions for virtual
columns with datetime comparisons are always allocated in the same
mem_root as the expressions for virtual columns. So now the inserted
cached items are valid for any queries that use these virtual columns.
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.
client/mysqldump.c:
Pass my_free directly as its signature is compatible with the
callback type -- which wasn't the case for free_table_ent.
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.
This assert checks that the server does not try to send EOF to the
client if there has been some error during processing. This to make
sure that the error is in fact sent to the client.
The problem was that any errors during processing of WHERE conditions
in HANDLER ... READ statements where not detected by the handler code.
The handler code therefore still tried to send EOF to the client,
triggering the assert. The bug was only noticeable in debug builds.
This patch fixes the problem by making sure that the handler code
checks for errors during condition processing and acts accordingly.
This assert checks that the server does not try to send EOF to the
client if there has been some error during processing. This to make
sure that the error is in fact sent to the client.
The problem was that any errors during processing of WHERE conditions
in HANDLER ... READ statements where not detected by the handler code.
The handler code therefore still tried to send EOF to the client,
triggering the assert. The bug was only noticeable in debug builds.
This patch fixes the problem by making sure that the handler code
checks for errors during condition processing and acts accordingly.
Problem: the server missed the fact that one can read from
2 indexes alternately using HANDLER interface.
Fix: check if the same (initialized) index is involved
reading next/prev values from the index.
mysql-test/r/handler_myisam.result:
Fix for bug #54007: assert in ha_myisam::index_next, HANDLER
- test result.
mysql-test/t/handler_myisam.test:
Fix for bug #54007: assert in ha_myisam::index_next, HANDLER
- test case.
sql/sql_handler.cc:
Fix for bug #54007: assert in ha_myisam::index_next, HANDLER
- check if we use the same (initialized) index
to read next/prev values from the index.
Problem: the server missed the fact that one can read from
2 indexes alternately using HANDLER interface.
Fix: check if the same (initialized) index is involved
reading next/prev values from the index.
mutex protecting thd->open_tables".
We should not manipulate with table->s->version outside the
table definition cache code, but use the TDC API
to achieve the desired result.
Fix one violation: close_all_tables_for_name().
sql/sql_base.cc:
Use tdc_remove_table(TDC_RT_REMOVE_ALL) to expel the
table share in close_all_tables_for_name, rather
than manipulate with the table cache explicitly.
Make sure that close_cached_tables() calls
close_all_tables_for_name() after closing all the involved
handlers. The rest of the code was inspected to
make sure that mysql_ha_rm_tables() is called.
sql/sql_handler.cc:
Add a method to close all HANDLER cursors for a list
of tables.
sql/sql_handler.h:
Add declaration for mysql_ha_flush_tables().
mutex protecting thd->open_tables".
We should not manipulate with table->s->version outside the
table definition cache code, but use the TDC API
to achieve the desired result.
Fix one violation: close_all_tables_for_name().
thd->open_tables"
thd->open_tables list is not normally accessed concurrently
except for one case: when the connection has open SQL
HANDLER tables, and we want to perform a DDL on the table,
we want to abort waits on MyISAM thr_lock of those connections
that prevent the DDL from proceeding, and iterate
over thd->open_tables list to find out the tables on which
the thread is waiting.
In 5.5 we mostly use deadlock detection and soft deadlock
prevention, as opposed to "hard" deadlock prevention
of 5.1, which would abort any transaction that
may cause a deadlock. The only remaining case when
neither deadlock detection nor deadlock prevention
is implemented in 5.5 is HANDLER SQL, where we use
old good thr_lock_abort() technique form 5.1.
Thus, replace use of LOCK_open to protect thd->open_tables
with thd->LOCK_ha_data (a lock protecting various session
private data).
This is a port of the work done for 5.5.4 for review
and inclusion into 5.5.5.
sql/sql_base.cc:
Use thd->LOCK_ha_data (systematically) to set
thd->open_tables.
sql/sql_class.h:
Implement THD::set_open_tables().
sql/sql_cursor.cc:
Use thd->LOCK_ha_data (systematically) to set
thd->open_tables.
sql/sql_handler.cc:
Use thd->LOCK_ha_data (systematically) to set
thd->open_tables.
Acquisition of LOCK_open is moved inside close_thread_table().
sql/sql_table.cc:
Acquisition of LOCK_open is moved inside close_thread_tables().
thd->open_tables"
thd->open_tables list is not normally accessed concurrently
except for one case: when the connection has open SQL
HANDLER tables, and we want to perform a DDL on the table,
we want to abort waits on MyISAM thr_lock of those connections
that prevent the DDL from proceeding, and iterate
over thd->open_tables list to find out the tables on which
the thread is waiting.
In 5.5 we mostly use deadlock detection and soft deadlock
prevention, as opposed to "hard" deadlock prevention
of 5.1, which would abort any transaction that
may cause a deadlock. The only remaining case when
neither deadlock detection nor deadlock prevention
is implemented in 5.5 is HANDLER SQL, where we use
old good thr_lock_abort() technique form 5.1.
Thus, replace use of LOCK_open to protect thd->open_tables
with thd->LOCK_ha_data (a lock protecting various session
private data).
This is a port of the work done for 5.5.4 for review
and inclusion into 5.5.5.
This patch:
- Moves all definitions from the mysql_priv.h file into
header files for the component where the variable is
defined
- Creates header files if the component lacks one
- Eliminates all include directives from mysql_priv.h
- Eliminates all circular include cycles
- Rename time.cc to sql_time.cc
- Rename mysql_priv.h to sql_priv.h
This patch:
- Moves all definitions from the mysql_priv.h file into
header files for the component where the variable is
defined
- Creates header files if the component lacks one
- Eliminates all include directives from mysql_priv.h
- Eliminates all circular include cycles
- Rename time.cc to sql_time.cc
- Rename mysql_priv.h to sql_priv.h
DDL no longer aborts mysql_lock_tables(), and hence
we no longer need to support need_reopen flag of this
call.
Remove the flag, and all the code in the server
that was responsible for handling the case when
it was set. This allowed to simplify:
open_and_lock_tables_derived(), the delayed thread,
multi-update.
Rename MYSQL_LOCK_IGNORE_FLUSH to MYSQL_OPEN_IGNORE_FLUSH,
since we now only support this flag in open_table().
Rename MYSQL_LOCK_PERF_SCHEMA to MYSQL_LOCK_LOG_TABLE,
to avoid confusion.
Move the wait for the global read lock for cases
when we do updates in SELECT f1() or DO (UPDATE) to
open_table() from mysql_lock_tables(). When waiting
for the read lock, we could raise need_reopen flag,
which is no longer present in mysql_lock_tables().
Since the block responsible for waiting for GRL
was moved, MYSQL_LOCK_IGNORE_GLOBAL_READ_LOCK
was renamed to MYSQL_OPEN_IGNORE_GLOBAL_READ_LOCK.
mysql-test/r/mdl_sync.result:
Update test results (see comments for mdl_sync.test).
mysql-test/t/mdl_sync.test:
Update tests: an abort mysql_lock_tables() called for an
INSERT no longer auto-closes SQL HANDLERS, since it
no longer leads to back-off and retry.
sql/ha_ndbcluster_binlog.cc:
Remove unused variables.
sql/lock.cc:
Remove support for need_reopen parameter of mysql_lock_tables().
Update comments.
sql/log_event_old.cc:
Remove the loop responsible for handling need_reopen
out parameter of mysql_lock_tables().
sql/mysql_priv.h:
Update open and lock tables flag names.
sql/share/errmsg-utf8.txt:
Add a new error message to report when
thr_multi_lock() is aborted.
sql/sql_base.cc:
Update comments. Rename MYSQL_LOCK_IGNORE_FLUSH
to MYSQL_OPEN_IGNORE_FLUSH.
sql/sql_class.h:
Remove unused code.
sql/sql_db.cc:
Remove an unused bit of code.
sql/sql_handler.cc:
For backward compatibility, we still want to back off and
retry when a call to mysql_lock_tables() is aborted
from within an SQL HANDLER. Write an internal error
handler to support the case.
sql/sql_insert.cc:
Call mysql_lock_tables() no longer has need_reopen
out parameter. Simplify the code by removing
the crud that took care of it.
MYSQL_LOCK_IGNORE_FLUSH is now only supported by
open_tables().
sql/sql_show.cc:
Rename MYSQL_LOCK_IGNORE_FLUSH to MYSQL_OPEN_IGNORE_FLUSH
sql/sql_table.cc:
Remove an unused parameter.
sql/sql_update.cc:
Remove the need_reopen loop from multi-update.
We no also longer need to cleanup the parse tree in case
when mysql_lock_tables() is aborted and thus an infinite
source of multi-update bugs is gone.
sql/tztime.cc:
Rename MYSQL_LOCK_IGNORE_FLUSH to MYSQL_OPEN_IGNORE_FLUSH,
since from now on this flag is only supported by open_table().
DDL no longer aborts mysql_lock_tables(), and hence
we no longer need to support need_reopen flag of this
call.
Remove the flag, and all the code in the server
that was responsible for handling the case when
it was set. This allowed to simplify:
open_and_lock_tables_derived(), the delayed thread,
multi-update.
Rename MYSQL_LOCK_IGNORE_FLUSH to MYSQL_OPEN_IGNORE_FLUSH,
since we now only support this flag in open_table().
Rename MYSQL_LOCK_PERF_SCHEMA to MYSQL_LOCK_LOG_TABLE,
to avoid confusion.
Move the wait for the global read lock for cases
when we do updates in SELECT f1() or DO (UPDATE) to
open_table() from mysql_lock_tables(). When waiting
for the read lock, we could raise need_reopen flag,
which is no longer present in mysql_lock_tables().
Since the block responsible for waiting for GRL
was moved, MYSQL_LOCK_IGNORE_GLOBAL_READ_LOCK
was renamed to MYSQL_OPEN_IGNORE_GLOBAL_READ_LOCK.
bool MDL_context::try_acquire_lock(MDL_request*)
This assert was triggered in the following way:
1) HANDLER OPEN t1 from connection 1
2) DROP TABLE t1 from connection 2. This will block due to the metadata lock
held by the open handler in connection 1.
3) DML statement (e.g. INSERT) from connection 1. This will close the table
opened by the HANDLER in 1) and release its metadata lock. This is done due
to the pending exclusive metadata lock from 2).
4) DROP TABLE t1 from connection 2 now completes and removes table t1.
5) HANDLER READ from connection 1. Since the handler table was closed in 3),
the handler code will try to reopen the table. First a new metadata lock on
t1 will be granted before the command fails since the table was removed in 4).
6) HANDLER READ from connection 1. This caused the assert.
The reason for the assert was that the MDL_request's pointer to the lock
ticket was not reset when the statement failed. HANDLER READ then tried to
acquire a lock using the same MDL_request object, triggering the assert.
This bug was only noticeable on debug builds and did not cause any problems
on release builds.
This patch fixes the problem by assuring that the pointer to the metadata
lock ticket is reset when reopening of handler tables fails.
Test case added to handler.inc
bool MDL_context::try_acquire_lock(MDL_request*)
This assert was triggered in the following way:
1) HANDLER OPEN t1 from connection 1
2) DROP TABLE t1 from connection 2. This will block due to the metadata lock
held by the open handler in connection 1.
3) DML statement (e.g. INSERT) from connection 1. This will close the table
opened by the HANDLER in 1) and release its metadata lock. This is done due
to the pending exclusive metadata lock from 2).
4) DROP TABLE t1 from connection 2 now completes and removes table t1.
5) HANDLER READ from connection 1. Since the handler table was closed in 3),
the handler code will try to reopen the table. First a new metadata lock on
t1 will be granted before the command fails since the table was removed in 4).
6) HANDLER READ from connection 1. This caused the assert.
The reason for the assert was that the MDL_request's pointer to the lock
ticket was not reset when the statement failed. HANDLER READ then tried to
acquire a lock using the same MDL_request object, triggering the assert.
This bug was only noticeable on debug builds and did not cause any problems
on release builds.
This patch fixes the problem by assuring that the pointer to the metadata
lock ticket is reset when reopening of handler tables fails.
Test case added to handler.inc
failed in enter_locked_tables_mode".
Server was aborted due to assertion failure when one tried to
execute statement requiring prelocking (i.e. firing triggers
or using stored functions) while having open HANDLERs.
The problem was that THD::enter_locked_tables_mode() method
which was called at the beginning of execution of prelocked
statement assumed there are no open HANDLERs. It had to do
so because corresponding THD::leave_locked_tables_mode()
method was unable to properly restore MDL sentinel when
leaving LOCK TABLES/prelocked mode in the presence of open
HANDLERs.
This patch solves this problem by changing the latter method
to properly restore MDL sentinel and thus removing need for
this assumption. As a side-effect, it lifts unjustified
limitation by allowing to keep HANDLERs open when entering
LOCK TABLES mode.
mysql-test/include/handler.inc:
Adjusted tests after making LOCK TABLES not to close
open HANDLERs. Added coverage for bug #50908
"Assertion `handler_tables_hash.records == 0' failed
in enter_locked_tables_mode".
mysql-test/r/handler_innodb.result:
Updated test results (see include/handler.inc).
mysql-test/r/handler_myisam.result:
Updated test results (see include/handler.inc).
sql/mysql_priv.h:
Introduced mysql_ha_move_tickets_after_trans_sentinel()
routine which allows to move tickets for metadata locks
corresponding to open HANDLERs after transaction sentinel.
sql/sql_class.cc:
Changed THD::leave_locked_tables_mode() to correctly restore
MDL sentinel value in the presence of open HANDLERs.
sql/sql_class.h:
Removed assert from THD::enter_locked_tables_mode() as we
no longer have to close HANDLERs when entering LOCK TABLES
or prelocked modes. Instead we keep them open and correctly
restore MDL sentinel value after leaving them.
Removal of assert also fixes problem from the bug report.
sql/sql_handler.cc:
Introduced mysql_ha_move_tickets_after_trans_sentinel()
routine which allows to move tickets for metadata locks
corresponding to open HANDLERs after transaction sentinel.
sql/sql_parse.cc:
We no longer have to close HANDLERs when entering LOCK TABLES
mode. Instead we keep them open and simply correctly restore
MDL sentinel value after leaving this mode.
failed in enter_locked_tables_mode".
Server was aborted due to assertion failure when one tried to
execute statement requiring prelocking (i.e. firing triggers
or using stored functions) while having open HANDLERs.
The problem was that THD::enter_locked_tables_mode() method
which was called at the beginning of execution of prelocked
statement assumed there are no open HANDLERs. It had to do
so because corresponding THD::leave_locked_tables_mode()
method was unable to properly restore MDL sentinel when
leaving LOCK TABLES/prelocked mode in the presence of open
HANDLERs.
This patch solves this problem by changing the latter method
to properly restore MDL sentinel and thus removing need for
this assumption. As a side-effect, it lifts unjustified
limitation by allowing to keep HANDLERs open when entering
LOCK TABLES mode.
and MDL".
Concurrent execution of a multi-DELETE statement and ALTER
TABLE statement which affected one of the tables used in
the multi-DELETE sometimes led to deadlock.
Similar deadlocks might have occured when one performed
INSERT/UPDATE/DELETE on a view and concurrently executed
ALTER TABLE for the view's underlying table, or when one
concurrently executed TRUNCATE TABLE for InnoDB table and
ALTER TABLE for the same table.
These deadlocks were caused by a discrepancy between types of
metadata and thr_lock.cc locks acquired by those statements.
What happened was that multi-DELETE/TRUNCATE/DML-through-the-
view statement in the first connection acquired SR lock on a
table, then ALTER TABLE would come in in the second connection
and acquire SNW metadata lock and TL_WRITE_ALLOW_READ
thr_lock.c lock and then would start waiting for the first
connection during lock upgrade. After that the statement in
the first connection would try to acquire TL_WRITE lock on
table and would start waiting for the second connection,
creating a deadlock.
This patch solves this problem by ensuring that we acquire
SW metadata lock in all cases in which we acquiring write
thr_lock.c lock. This guarantees that deadlocks like the
one described above won't occur since all lock conflicts
in such situation are resolved within MDL subsystem.
This patch also adds assert which should guarantee that
such situations won't arise in future.
mysql-test/r/lock_multi.result:
Added main test for bug #50913 "Deadlock between
open_and_lock_tables_derived and MDL".
mysql-test/r/mdl_sync.result:
Added additional coverage for bug #50913 "Deadlock
between open_and_lock_tables_derived and MDL".
mysql-test/t/lock_multi.test:
Added main test for bug #50913 "Deadlock between
open_and_lock_tables_derived and MDL".
mysql-test/t/mdl_sync.test:
Added additional coverage for bug #50913 "Deadlock
between open_and_lock_tables_derived and MDL".
sql/lock.cc:
Added assert that enforces that when we are locking
a non-temporary table we have an appropriate type of
metadata lock on this table.
sql/mysql_priv.h:
Added separate flag for open_tables() to be able specify that
SH metadata locks on table to be open should be acquired.
We can no longer use MYSQL_LOCK_IGNORE_FLUSH flag for this
as in addition to use in I_S implementation it is also used
for opening system tables. Since in the latter case we also
acquire thr_lock.c locks using SH metadata lock in it instead
of SR or SW locks may lead to deadlock.
sql/sql_base.cc:
When opening tables don't interpret MYSQL_LOCK_IGNORE_FLUSH
flag as request to acquire SH metadata locks. This flag is
also used for opening system tables for which we also take
thr_lock.c locks and thus proper metadata lock to take in
this case is SR or SW lock (otherwise deadlocks can occur).
In cases when SH lock is really required (e.g. when tables
are open by I_S implementation) we rely on that newly
introduced MYSQL_OPEN_FORCE_SHARED_HIGH_PRIO_MDL flag is
used.
sql/sql_delete.cc:
mysql_truncate_by_delete():
Adjust type of metadata lock to be requested after changing
type of thr_lock.c lock for table list element from one
which was set in parser to TL_WRITE.
This removes discrepancy between types of these locks which
allowed deadlocks to creep in.
sql/sql_handler.cc:
When closing table which was open by HANDLER statement clear
TABLE::open_by_handler flag. This allows to use this flag as
a reliable indication that TABLE instance was open by HANDLER
statement in assert which was added to mysql_lock_tables().
sql/sql_parse.cc:
multi_delete_set_locks_and_link_aux_tables():
Adjust type of metadata lock to be requested after changing
type of thr_lock.c lock for table list element from one
which was set in parser to TL_WRITE.
This removes discrepancy between types of these locks which
allowed deadlocks to creep in.
sql/sql_show.cc:
Use newly introduced MYSQL_OPEN_FORCE_SHARED_HIGH_PRIO_MDL
flag in order to acquire SH metadata locks when opening tables
in I_S implementation.
sql/sql_update.cc:
Added comment explaining why in multi-update after deciding
that we need weaker thr_lock.c lock on a table we don't
downgrade metadata lock on it.
sql/sql_view.cc:
When merging view into main statement adjust type of metadata
lock to be requested after changing type of thr_lock.c lock
for table. This removes discrepancy between types of these
locks which allowed deadlocks to creep in.
and MDL".
Concurrent execution of a multi-DELETE statement and ALTER
TABLE statement which affected one of the tables used in
the multi-DELETE sometimes led to deadlock.
Similar deadlocks might have occured when one performed
INSERT/UPDATE/DELETE on a view and concurrently executed
ALTER TABLE for the view's underlying table, or when one
concurrently executed TRUNCATE TABLE for InnoDB table and
ALTER TABLE for the same table.
These deadlocks were caused by a discrepancy between types of
metadata and thr_lock.cc locks acquired by those statements.
What happened was that multi-DELETE/TRUNCATE/DML-through-the-
view statement in the first connection acquired SR lock on a
table, then ALTER TABLE would come in in the second connection
and acquire SNW metadata lock and TL_WRITE_ALLOW_READ
thr_lock.c lock and then would start waiting for the first
connection during lock upgrade. After that the statement in
the first connection would try to acquire TL_WRITE lock on
table and would start waiting for the second connection,
creating a deadlock.
This patch solves this problem by ensuring that we acquire
SW metadata lock in all cases in which we acquiring write
thr_lock.c lock. This guarantees that deadlocks like the
one described above won't occur since all lock conflicts
in such situation are resolved within MDL subsystem.
This patch also adds assert which should guarantee that
such situations won't arise in future.
HANDLER OPEN
The problem was a too restrictive assert in the code for
HANDLER ... OPEN and HANDLER ... READ that checked table->next
to verify that we didn't open views or merge tables.
This pointer is also used to link temporary tables together
(see thd->temporary_tables). In this case TABLE::next can be
set even if we're trying to open a single table.
This patch adjust the two asserts to also check for the presence
of temporary tables.
Test case added to handler_myisam.test.
HANDLER OPEN
The problem was a too restrictive assert in the code for
HANDLER ... OPEN and HANDLER ... READ that checked table->next
to verify that we didn't open views or merge tables.
This pointer is also used to link temporary tables together
(see thd->temporary_tables). In this case TABLE::next can be
set even if we're trying to open a single table.
This patch adjust the two asserts to also check for the presence
of temporary tables.
Test case added to handler_myisam.test.
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.
mysql-test/include/handler.inc:
Adjusted test case to trigger an execution path on which bug 41110
"crash with handler command when used concurrently with alter
table" and bug 41112 "crash in mysql_ha_close_table/get_lock_data
with alter table" were originally discovered. Left old test case
which no longer triggers this execution path for the sake of
coverage.
Added test coverage for HANDLER SQL statements and type-aware
metadata locks.
Added a test for the global shared lock and HANDLER SQL.
Updated tests to take into account that the old simple deadlock
detection heuristics was replaced with a graph-based deadlock
detector.
mysql-test/r/debug_sync.result:
Updated results (see debug_sync.test).
mysql-test/r/handler_innodb.result:
Updated results (see handler.inc test).
mysql-test/r/handler_myisam.result:
Updated results (see handler.inc test).
mysql-test/r/innodb-lock.result:
Updated results (see innodb-lock.test).
mysql-test/r/innodb_mysql_lock.result:
Updated results (see innodb_mysql_lock.test).
mysql-test/r/lock.result:
Updated results (see lock.test).
mysql-test/r/lock_multi.result:
Updated results (see lock_multi.test).
mysql-test/r/lock_sync.result:
Updated results (see lock_sync.test).
mysql-test/r/mdl_sync.result:
Updated results (see mdl_sync.test).
mysql-test/r/sp-threads.result:
SHOW PROCESSLIST output has changed due to the fact that waiting
for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/r/truncate_coverage.result:
Updated results (see truncate_coverage.test).
mysql-test/suite/funcs_1/datadict/processlist_val.inc:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/funcs_1/r/processlist_val_no_prot.result:
SELECT FROM I_S.PROCESSLIST output has changed due to fact that
waiting for LOCK TABLES WRITE now happens within metadata locking
subsystem.
mysql-test/suite/rpl/t/rpl_sp.test:
Updated to a new SHOW PROCESSLIST state name.
mysql-test/t/debug_sync.test:
Use LOCK TABLES READ instead of LOCK TABLES WRITE as the latter
no longer allows to trigger execution path involving waiting on
thr_lock.c lock and therefore reaching debug sync-point covered
by this test.
mysql-test/t/innodb-lock.test:
Adjusted test case to the fact that innodb_table_locks=0 option is
no longer supported, since LOCK TABLES WRITE handles all its
conflicts within MDL subsystem.
mysql-test/t/innodb_mysql_lock.test:
Added test for bug #37346 "innodb does not detect deadlock between
update and alter table".
mysql-test/t/lock.test:
Added test coverage which checks the fact that we no longer support
DDL under LOCK TABLES on tables which were locked implicitly.
Adjusted existing test cases accordingly.
mysql-test/t/lock_multi.test:
Added test for bug #46272 "MySQL 5.4.4, new MDL: unnecessary
deadlock". Adjusted other test cases to take into account the
fact that waiting for LOCK TABLES ... WRITE now happens within MDL
subsystem.
mysql-test/t/lock_sync.test:
Since LOCK TABLES ... WRITE now takes SNRW metadata lock for
tables locked explicitly we have to implicitly lock InnoDB tables
(through view) to trigger the table-level lock conflict between
TL_WRITE and TL_WRITE_ALLOW_WRITE.
mysql-test/t/mdl_sync.test:
Added basic test coverage for type-of-operation-aware metadata
locks. Also covered with tests some use cases involving HANDLER
statements in which a deadlock could arise.
Adjusted existing tests to take type-of-operation-aware MDL into
account.
mysql-test/t/multi_update.test:
Update to a new SHOW PROCESSLIST state name.
mysql-test/t/truncate_coverage.test:
Adjusted test case after making LOCK TABLES WRITE to wait until
transactions that use the table to be locked are completed.
Updated to the changed name of DEBUG_SYNC point.
sql/handler.cc:
Global read lock functionality has been
moved into a class.
sql/lock.cc:
Global read lock functionality has been
moved into a class.
Updated code to use the new MDL API.
sql/mdl.cc:
Introduced new type-of-operation aware metadata locks.
To do this:
- Changed MDL_lock to use one list for waiting requests and one
list for granted requests. For each list, added a bitmap
that holds information what lock types a list contains.
Added a helper class MDL_lock::List to manipulate with granted
and waited lists while keeping the bitmaps in sync
with list contents.
- Changed lock-compatibility functions to use bitmaps that
define compatibility.
- Introduced a graph based deadlock detector inspired by
waiting_threads.c from Maria implementation.
- Now that we have a deadlock detector, and no longer have
a global lock to protect individual lock objects, but rather
use an rw lock per object, removed redundant code for upgrade,
and the global read lock. Changed the MDL API to
no longer require the caller to acquire the global
intention exclusive lock by means of a separate method.
Removed a few more methods that became redundant.
- Removed deadlock detection heuristic, it has been made
obsolete by the deadlock detector.
- With operation-type-aware metadata locks, MDL subsystem has
become aware of potential conflicts between DDL and open
transactions. This made it possible to remove calls to
mysql_abort_transactions_with_shared_lock() from acquisition
paths for exclusive lock and lock upgrade. Now we can simply
wait for these transactions to complete without fear of
deadlock. Function mysql_lock_abort() has also become
unnecessary for all conflicting cases except when a DDL
conflicts with a connection that has an open HANDLER.
sql/mdl.h:
Introduced new type-of-operation aware metadata locks.
Introduced a graph based deadlock detector and supporting
methods.
Added comments.
God rid of redundant API calls.
Renamed m_lt_or_ha_sentinel to m_trans_sentinel,
since now it guards the global read lock as well as
LOCK TABLES and HANDLER locks.
sql/mysql_priv.h:
Moved the global read lock functionality into a
class.
Added MYSQL_OPEN_FORCE_SHARED_MDL flag which forces
open_tables() to take MDL_SHARED on tables instead of
metadata locks specified in the parser. We use this to
allow PREPARE run concurrently in presence of
LOCK TABLES ... WRITE.
Added signature for find_table_for_mdl_ugprade().
sql/set_var.cc:
Global read lock functionality has been
moved into a class.
sql/sp_head.cc:
When creating TABLE_LIST elements for prelocking or
system tables set the type of request for metadata
lock according to the operation that will be performed
on the table.
sql/sql_base.cc:
- Updated code to use the new MDL API.
- In order to avoid locks starvation we take upgradable
locks all at once. As result implicitly locked tables no
longer get an upgradable lock. Consequently DDL and FLUSH
TABLES for such tables is prohibited.
find_write_locked_table() was replaced by
find_table_for_mdl_upgrade() function.
open_table() was adjusted to return TABLE instance with
upgradable ticket when necessary.
- We no longer wait for all locks on OT_WAIT back off
action -- only on the lock that caused the wait
conflict. Moreover, now we distinguish cases when we
have to wait due to conflict in MDL and old version
of table in TDC.
- Upate mysql_notify_threads_having_share_locks()
to only abort thr_lock.c waits of threads that
have open HANDLERs, since lock conflicts with only
these threads now can lead to deadlocks not detectable
by the MDL deadlock detector.
- Remove mysql_abort_transactions_with_shared_locks()
which is no longer needed.
sql/sql_class.cc:
Global read lock functionality has been moved into a class.
Re-arranged code in THD::cleanup() to simplify assert.
sql/sql_class.h:
Introduced class to incapsulate global read lock
functionality.
Now sentinel in MDL subsystem guards the global read lock
as well as LOCK TABLES and HANDLER locks. Adjusted code
accordingly.
sql/sql_db.cc:
Global read lock functionality has been moved into a class.
sql/sql_delete.cc:
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
TRUNCATE TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_handler.cc:
Inform MDL_context about presence of open HANDLERs.
Since HANLDERs break MDL protocol by acquiring table-level
lock while holding only S metadata lock on a table MDL
subsystem should take special care about such contexts (Now
this is the only case when mysql_lock_abort() is used).
sql/sql_parse.cc:
Global read lock functionality has been moved into a class.
Do not take upgradable metadata locks when opening tables
for CREATE TABLE SELECT as it is not necessary and limits
concurrency.
When initializing TABLE_LIST objects before adding them
to the table list set the type of request for metadata lock
according to the operation that will be performed on the
table.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result FLUSH
TABLES is no longer allowed for such tables.
sql/sql_prepare.cc:
Use MYSQL_OPEN_FORCE_SHARED_MDL flag when opening
tables during PREPARE. This allows PREPARE to run
concurrently in presence of LOCK TABLES ... WRITE.
sql/sql_rename.cc:
Global read lock functionality has been moved into a class.
sql/sql_show.cc:
Updated code to use the new MDL API.
sql/sql_table.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result DROP
TABLE is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_trigger.cc:
Global read lock functionality has been moved into a class.
We no longer acquire upgradable metadata locks on tables
which are locked by LOCK TABLES implicitly. As result
CREATE/DROP TRIGGER is no longer allowed for such tables.
Updated code to use the new MDL API.
sql/sql_view.cc:
Global read lock functionality has been moved into a class.
Fixed results of wrong merge that led to misuse of GLR API.
CREATE VIEW statement is not a commit statement.
sql/table.cc:
When resetting TABLE_LIST objects for PS or SP re-execution
set the type of request for metadata lock according to the
operation that will be performed on the table. Do the same
in auxiliary function initializing metadata lock requests
in a table list.
sql/table.h:
When initializing TABLE_LIST objects set the type of request
for metadata lock according to the operation that will be
performed on the table.
sql/transaction.cc:
Global read lock functionality has been moved into a class.
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.
"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.
mysql-test/include/handler.inc:
Add test coverage for Bug#46224 "HANDLER statements within a
transaction might lead to deadlocks".
Extended HANDLER coverage to cover a mix of HANDLER, transactions
and DDL statements.
mysql-test/r/handler_innodb.result:
Update results (Bug#46224).
mysql-test/r/handler_myisam.result:
Update results (Bug#46224).
sql/lock.cc:
Remove thd->some_tables_deleted, it's never used.
sql/log_event.cc:
No need to check for thd->locked_tables_mode,
it's done inside release_transactional_locks().
sql/mdl.cc:
Implement the concept of HANDLER and LOCK TABLES "sentinel".
Implement a method to clone an acquired ticket.
Do not return tickets beyond the sentinel when acquiring
locks, create a copy.
Remove methods to merge and backup MDL_context, they are now
not used (Hurra!). This opens a path to a proper constructor
and destructor of class MDL_context (to be done in a separate
patch).
Modify find_ticket() to provide information about where
the ticket position is with regard to the sentinel.
sql/mdl.h:
Add declarations necessary for the implementation of the concept
of "sentinel", a dedicated ticket separating transactional and
non-transactional locks.
sql/mysql_priv.h:
Add mark_tmp_table_for_reuse() declaration,
a function to "close" a single session (temporary) table.
sql/sql_base.cc:
Remove thd->some_tables_deleted.
Modify deadlock-prevention asserts and deadlock detection
heuristics to take into account that from now on HANDLER locks
reside in the same locking context.
Add broadcast_refresh() to mysql_notify_thread_having_shared_lock():
this is necessary for the case when a thread having a shared lock
is asleep in tdc_wait_for_old_versions(). This situation is only
possible with HANDLER t1 OPEN; FLUSH TABLE (since all over code paths
that lead to tdc_wait_for_old_versions() always have an
empty MDL_context). Previously the server would simply deadlock
in this situation.
sql/sql_class.cc:
Remove now unused member "THD::some_tables_deleted".
Move mysql_ha_cleanup() a few lines above in THD::cleanup()
to make sure that all handlers are closed when it's time to
destroy the MDL_context of this connection.
Remove handler_mdl_context and handler_tables.
sql/sql_class.h:
Remove THD::handler_tables, THD::handler_mdl_context,
THD::some_tables_deleted.
sql/sql_handler.cc:
Remove thd->handler_tables.
Remove thd->handler_mdl_context.
Rewrite mysql_ha_open() to have no special provision for MERGE
tables, now that we don't have to manipulate with thd->handler_tables
it's easy to do.
Remove dead code.
Fix a bug in mysql_ha_flush() when we would always flush
a temporary HANDLER when mysql_ha_flush() is called (actually
mysql_ha_flush() never needs to flush temporary tables).
sql/sql_insert.cc:
Update a comment, no more thd->some_tables_deleted.
sql/sql_parse.cc:
Implement an incompatible change: entering LOCK TABLES closes
active HANDLERs, if any.
Now that we have a sentinel, we don't need to check
for thd->locked_tables_mode when releasing metadata locks in
COMMIT/ROLLBACK.
sql/sql_plist.h:
Add new (now necessary) methods to the list class.
sql/sql_prepare.cc:
Make sure we don't release HANDLER locks when rollback to a
savepoint, set to not keep locks taken at PREPARE.
sql/sql_servers.cc:
Update to a new signature of MDL_context::release_all_locks().
sql/sql_table.cc:
Remove thd->some_tables_deleted.
sql/transaction.cc:
Add comments.
Make sure rollback to (MDL) savepoint works under LOCK TABLES and
with HANDLER tables.
"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.
Bug #46654 False deadlock on concurrent DML/DDL with partitions,
inconsistent behavior
The problem was that if one connection is running a multi-statement
transaction which involves a single partitioned table, and another
connection attempts to alter the table, the first connection gets
ER_LOCK_DEADLOCK and cannot proceed anymore, even when the ALTER TABLE
statement in another connection has timed out or failed.
The reason for this was that the prepare phase for ALTER TABLE for
partitioned tables removed all instances of the table from the table
definition cache before it started waiting on the lock. The transaction
running in the first connection would notice this and report ER_LOCK_DEADLOCK.
This patch changes the prep_alter_part_table() ALTER TABLE code so that
tdc_remove_table() is no longer called. Instead, only the TABLE instance
changed by prep_alter_part_table() is marked as needing reopen.
The patch also removes an unnecessary call to tdc_remove_table() from
mysql_unpack_partition() as the changed TABLE object is destroyed by the
caller at a later point.
Test case added in partition_sync.test.
Bug #46654 False deadlock on concurrent DML/DDL with partitions,
inconsistent behavior
The problem was that if one connection is running a multi-statement
transaction which involves a single partitioned table, and another
connection attempts to alter the table, the first connection gets
ER_LOCK_DEADLOCK and cannot proceed anymore, even when the ALTER TABLE
statement in another connection has timed out or failed.
The reason for this was that the prepare phase for ALTER TABLE for
partitioned tables removed all instances of the table from the table
definition cache before it started waiting on the lock. The transaction
running in the first connection would notice this and report ER_LOCK_DEADLOCK.
This patch changes the prep_alter_part_table() ALTER TABLE code so that
tdc_remove_table() is no longer called. Instead, only the TABLE instance
changed by prep_alter_part_table() is marked as needing reopen.
The patch also removes an unnecessary call to tdc_remove_table() from
mysql_unpack_partition() as the changed TABLE object is destroyed by the
caller at a later point.
Test case added in partition_sync.test.
------------------------------------------------------------
revno: 2617.68.25
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-next-bg-pre2-2
timestamp: Wed 2009-09-16 18:26:50 +0400
message:
Follow-up for one of pre-requisite patches for fixing bug #30977
"Concurrent statement using stored function and DROP FUNCTION
breaks SBR".
Made enum_mdl_namespace enum part of MDL_key class and removed MDL_
prefix from the names of enum members. In order to do the latter
changed name of PROCEDURE symbol to PROCEDURE_SYM (otherwise macro
which was automatically generated for this symbol conflicted with
MDL_key::PROCEDURE enum member).
------------------------------------------------------------
revno: 2617.68.25
committer: Dmitry Lenev <dlenev@mysql.com>
branch nick: mysql-next-bg-pre2-2
timestamp: Wed 2009-09-16 18:26:50 +0400
message:
Follow-up for one of pre-requisite patches for fixing bug #30977
"Concurrent statement using stored function and DROP FUNCTION
breaks SBR".
Made enum_mdl_namespace enum part of MDL_key class and removed MDL_
prefix from the names of enum members. In order to do the latter
changed name of PROCEDURE symbol to PROCEDURE_SYM (otherwise macro
which was automatically generated for this symbol conflicted with
MDL_key::PROCEDURE enum member).
