database/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-08-30 11:22:14 +03:00
Code Activity

Patch that refactors global read lock implementation and fixes

Browse Source 
bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ
LOCK" and bug #54673 "It takes too long to get readlock for
'FLUSH TABLES WITH READ LOCK'".

The first bug manifested itself as a deadlock which occurred
when a connection, which had some table open through HANDLER
statement, tried to update some data through DML statement
while another connection tried to execute FLUSH TABLES WITH
READ LOCK concurrently.

What happened was that FTWRL in the second connection managed
to perform first step of GRL acquisition and thus blocked all
upcoming DML. After that it started to wait for table open
through HANDLER statement to be flushed. When the first connection
tried to execute DML it has started to wait for GRL/the second
connection creating deadlock.

The second bug manifested itself as starvation of FLUSH TABLES
WITH READ LOCK statements in cases when there was a constant
stream of concurrent DML statements (in two or more
connections).

This has happened because requests for protection against GRL
which were acquired by DML statements were ignoring presence of
pending GRL and thus the latter was starved.

This patch solves both these problems by re-implementing GRL
using metadata locks.

Similar to the old implementation acquisition of GRL in new
implementation is two-step. During the first step we block
all concurrent DML and DDL statements by acquiring global S
metadata lock (each DML and DDL statement acquires global IX
lock for its duration). During the second step we block commits
by acquiring global S lock in COMMIT namespace (commit code
acquires global IX lock in this namespace).

Note that unlike in old implementation acquisition of
protection against GRL in DML and DDL is semi-automatic.
We assume that any statement which should be blocked by GRL
will either open and acquires write-lock on tables or acquires
metadata locks on objects it is going to modify. For any such
statement global IX metadata lock is automatically acquired
for its duration.

The first problem is solved because waits for GRL become
visible to deadlock detector in metadata locking subsystem
and thus deadlocks like one in the first bug become impossible.

The second problem is solved because global S locks which
are used for GRL implementation are given preference over
IX locks which are acquired by concurrent DML (and we can
switch to fair scheduling in future if needed).

Important change:
FTWRL/GRL no longer blocks DML and DDL on temporary tables.
Before this patch behavior was not consistent in this respect:
in some cases DML/DDL statements on temporary tables were
blocked while in others they were not. Since the main use cases
for FTWRL are various forms of backups and temporary tables are
not preserved during backups we have opted for consistently
allowing DML/DDL on temporary tables during FTWRL/GRL.

Important change:
This patch changes thread state names which are used when
DML/DDL of FTWRL is waiting for global read lock. It is now
either "Waiting for global read lock" or "Waiting for commit
lock" depending on the stage on which FTWRL is.

Incompatible change:
To solve deadlock in events code which was exposed by this
patch we have to replace LOCK_event_metadata mutex with
metadata locks on events. As result we have to prohibit
DDL on events under LOCK TABLES.

This patch also adds extensive test coverage for interaction
of DML/DDL and FTWRL.

Performance of new and old global read lock implementations
in sysbench tests were compared. There were no significant
difference between new and old implementations.
This commit is contained in:
Dmitry Lenev
2010-11-11 20:11:05 +03:00
parent aee8fce233
commit 378cdc58c1
75 changed files with 5492 additions and 1243 deletions
Download Patch File Download Diff File Expand all files Collapse all files

154
sql/sql_parse.cc
View File

@@ -18,12 +18,9 @@
#include "sql_priv.h"
#include "unireg.h" // REQUIRED: for other includes
#include "sql_parse.h" // sql_kill, *_precheck, *_prepare
#include "lock.h" // wait_if_global_read_lock,
// unlock_global_read_lock,
// try_transactional_lock,
#include "lock.h" // try_transactional_lock,
// check_transactional_lock,
// set_handler_table_locks,
// start_waiting_global_read_lock,
// lock_global_read_lock,
// make_global_read_lock_block_commit
#include "sql_base.h" // find_temporary_tablesx
@@ -260,21 +257,20 @@ void init_update_queries(void)
the code, in particular in the Query_log_event's constructor.
*/
sql_command_flags[SQLCOM_CREATE_TABLE]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_AUTO_COMMIT_TRANS | CF_PROTECT_AGAINST_GRL |
CF_AUTO_COMMIT_TRANS |
CF_CAN_GENERATE_ROW_EVENTS;
sql_command_flags[SQLCOM_CREATE_INDEX]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_ALTER_TABLE]= CF_CHANGES_DATA | CF_WRITE_LOGS_COMMAND |
CF_AUTO_COMMIT_TRANS | CF_PROTECT_AGAINST_GRL;
CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_TRUNCATE]= CF_CHANGES_DATA | CF_WRITE_LOGS_COMMAND |
CF_AUTO_COMMIT_TRANS | CF_PROTECT_AGAINST_GRL;
CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_TABLE]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_LOAD]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_PROTECT_AGAINST_GRL |
CF_CAN_GENERATE_ROW_EVENTS;
sql_command_flags[SQLCOM_CREATE_DB]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS | CF_PROTECT_AGAINST_GRL;
sql_command_flags[SQLCOM_DROP_DB]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS | CF_PROTECT_AGAINST_GRL;
sql_command_flags[SQLCOM_ALTER_DB_UPGRADE]= CF_AUTO_COMMIT_TRANS | CF_PROTECT_AGAINST_GRL;
sql_command_flags[SQLCOM_ALTER_DB]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS | CF_PROTECT_AGAINST_GRL;
sql_command_flags[SQLCOM_CREATE_DB]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_DB]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_ALTER_DB_UPGRADE]= CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_ALTER_DB]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_RENAME_TABLE]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_INDEX]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_CREATE_VIEW]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
@@ -285,26 +281,18 @@ void init_update_queries(void)
sql_command_flags[SQLCOM_CREATE_EVENT]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_ALTER_EVENT]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_EVENT]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_CREATE_TRIGGER]= CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_TRIGGER]= CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_UPDATE]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_PROTECT_AGAINST_GRL |
CF_CAN_GENERATE_ROW_EVENTS;
sql_command_flags[SQLCOM_UPDATE_MULTI]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_PROTECT_AGAINST_GRL |
CF_CAN_GENERATE_ROW_EVENTS;
sql_command_flags[SQLCOM_INSERT]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_PROTECT_AGAINST_GRL |
CF_CAN_GENERATE_ROW_EVENTS;
sql_command_flags[SQLCOM_INSERT_SELECT]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_PROTECT_AGAINST_GRL |
CF_CAN_GENERATE_ROW_EVENTS;
sql_command_flags[SQLCOM_DELETE]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_PROTECT_AGAINST_GRL |
CF_CAN_GENERATE_ROW_EVENTS;
sql_command_flags[SQLCOM_DELETE_MULTI]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_PROTECT_AGAINST_GRL |
CF_CAN_GENERATE_ROW_EVENTS;
sql_command_flags[SQLCOM_REPLACE]= CF_CHANGES_DATA | CF_REEXECUTION_FRAGILE |
CF_CAN_GENERATE_ROW_EVENTS;
@@ -366,20 +354,19 @@ void init_update_queries(void)
CF_REEXECUTION_FRAGILE);
sql_command_flags[SQLCOM_CREATE_USER]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL;
sql_command_flags[SQLCOM_RENAME_USER]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL;
sql_command_flags[SQLCOM_DROP_USER]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL;
sql_command_flags[SQLCOM_CREATE_USER]= CF_CHANGES_DATA;
sql_command_flags[SQLCOM_RENAME_USER]= CF_CHANGES_DATA;
sql_command_flags[SQLCOM_DROP_USER]= CF_CHANGES_DATA;
sql_command_flags[SQLCOM_GRANT]= CF_CHANGES_DATA;
sql_command_flags[SQLCOM_REVOKE]= CF_CHANGES_DATA;
sql_command_flags[SQLCOM_REVOKE_ALL]= CF_PROTECT_AGAINST_GRL;
sql_command_flags[SQLCOM_OPTIMIZE]= CF_CHANGES_DATA;
sql_command_flags[SQLCOM_CREATE_FUNCTION]= CF_CHANGES_DATA;
sql_command_flags[SQLCOM_CREATE_PROCEDURE]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_CREATE_SPFUNCTION]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_PROCEDURE]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_FUNCTION]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_ALTER_PROCEDURE]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_ALTER_FUNCTION]= CF_CHANGES_DATA | CF_PROTECT_AGAINST_GRL | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_CREATE_PROCEDURE]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_CREATE_SPFUNCTION]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_PROCEDURE]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_DROP_FUNCTION]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_ALTER_PROCEDURE]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_ALTER_FUNCTION]= CF_CHANGES_DATA | CF_AUTO_COMMIT_TRANS;
sql_command_flags[SQLCOM_INSTALL_PLUGIN]= CF_CHANGES_DATA;
sql_command_flags[SQLCOM_UNINSTALL_PLUGIN]= CF_CHANGES_DATA;
@@ -1111,7 +1098,7 @@ bool dispatch_command(enum enum_server_command command, THD *thd,
SHOW statements should not add the used tables to the list of tables
used in a transaction.
*/
MDL_ticket *mdl_savepoint= thd->mdl_context.mdl_savepoint();
MDL_savepoint mdl_savepoint= thd->mdl_context.mdl_savepoint();
status_var_increment(thd->status_var.com_stat[SQLCOM_SHOW_FIELDS]);
if (thd->copy_db_to(&db.str, &db.length))
@@ -1754,16 +1741,6 @@ bool sp_process_definer(THD *thd)
/**
Execute command saved in thd and lex->sql_command.
Before every operation that can request a write lock for a table
wait if a global read lock exists. However do not wait if this
thread has locked tables already. No new locks can be requested
until the other locks are released. The thread that requests the
global read lock waits for write locked tables to become unlocked.
Note that wait_if_global_read_lock() sets a protection against a new
global read lock when it succeeds. This needs to be released by
start_waiting_global_read_lock() after the operation.
@param thd Thread handle
@todo
@@ -1797,7 +1774,6 @@ mysql_execute_command(THD *thd)
/* have table map for update for multi-update statement (BUG#37051) */
bool have_table_map_for_update= FALSE;
#endif
/* Saved variable value */
DBUG_ENTER("mysql_execute_command");
#ifdef WITH_PARTITION_STORAGE_ENGINE
thd->work_part_info= 0;
@@ -1989,17 +1965,6 @@ mysql_execute_command(THD *thd)
thd->mdl_context.release_transactional_locks();
}
/*
Check if this command needs protection against the global read lock
to avoid deadlock. See CF_PROTECT_AGAINST_GRL.
start_waiting_global_read_lock() is called at the end of
mysql_execute_command().
*/
if (((sql_command_flags[lex->sql_command] & CF_PROTECT_AGAINST_GRL) != 0) &&
!thd->locked_tables_mode)
if (thd->global_read_lock.wait_if_global_read_lock(thd, FALSE, TRUE))
goto error;
#ifndef DBUG_OFF
if (lex->sql_command != SQLCOM_SET_OPTION)
DEBUG_SYNC(thd,"before_execute_sql_command");
@@ -2074,10 +2039,6 @@ mysql_execute_command(THD *thd)
if (res)
break;
if (!thd->locked_tables_mode && lex->protect_against_global_read_lock &&
thd->global_read_lock.wait_if_global_read_lock(thd, FALSE, TRUE))
break;
res= execute_sqlcom_select(thd, all_tables);
break;
}
@@ -2336,20 +2297,6 @@ case SQLCOM_PREPARE:
create_info.default_table_charset= create_info.table_charset;
create_info.table_charset= 0;
}
/*
The create-select command will open and read-lock the select table
and then create, open and write-lock the new table. If a global
read lock steps in, we get a deadlock. The write lock waits for
the global read lock, while the global read lock waits for the
select table to be closed. So we wait until the global readlock is
gone before starting both steps. Note that
wait_if_global_read_lock() sets a protection against a new global
read lock when it succeeds. This needs to be released by
start_waiting_global_read_lock(). We protect the normal CREATE
TABLE in the same way. That way we avoid that a new table is
created during a global read lock.
Protection against grl is covered by the CF_PROTECT_AGAINST_GRL flag.
*/
#ifdef WITH_PARTITION_STORAGE_ENGINE
{
@@ -3143,9 +3090,6 @@ end_with_restore_list:
if (check_table_access(thd, LOCK_TABLES_ACL | SELECT_ACL, all_tables,
FALSE, UINT_MAX, FALSE))
goto error;
if (lex->protect_against_global_read_lock &&
thd->global_read_lock.wait_if_global_read_lock(thd, FALSE, TRUE))
goto error;
thd->variables.option_bits|= OPTION_TABLE_LOCK;
thd->in_lock_tables=1;
@@ -3801,13 +3745,22 @@ end_with_restore_list:
Security_context *backup= NULL;
LEX_USER *definer= thd->lex->definer;
/*
We're going to issue an implicit GRANT statement.
It takes metadata locks and updates system tables.
Make sure that sp_create_routine() did not leave any
locks in the MDL context, so there is no risk to
deadlock.
We're going to issue an implicit GRANT statement so we close all
open tables. We have to keep metadata locks as this ensures that
this statement is atomic against concurent FLUSH TABLES WITH READ
LOCK. Deadlocks which can arise due to fact that this implicit
statement takes metadata locks should be detected by a deadlock
detector in MDL subsystem and reported as errors.
No need to commit/rollback statement transaction, it's not started.
TODO: Long-term we should either ensure that implicit GRANT statement
is written into binary log as a separate statement or make both
creation of routine and implicit GRANT parts of one fully atomic
statement.
*/
close_mysql_tables(thd);
DBUG_ASSERT(thd->transaction.stmt.is_empty());
close_thread_tables(thd);
/*
Check if the definer exists on slave,
then use definer privilege to insert routine privileges to mysql.procs_priv.
@@ -4067,13 +4020,22 @@ create_sp_error:
#ifndef NO_EMBEDDED_ACCESS_CHECKS
/*
We're going to issue an implicit REVOKE statement.
It takes metadata locks and updates system tables.
Make sure that sp_create_routine() did not leave any
locks in the MDL context, so there is no risk to
deadlock.
We're going to issue an implicit REVOKE statement so we close all
open tables. We have to keep metadata locks as this ensures that
this statement is atomic against concurent FLUSH TABLES WITH READ
LOCK. Deadlocks which can arise due to fact that this implicit
statement takes metadata locks should be detected by a deadlock
detector in MDL subsystem and reported as errors.
No need to commit/rollback statement transaction, it's not started.
TODO: Long-term we should either ensure that implicit REVOKE statement
is written into binary log as a separate statement or make both
dropping of routine and implicit REVOKE parts of one fully atomic
statement.
*/
close_mysql_tables(thd);
DBUG_ASSERT(thd->transaction.stmt.is_empty());
close_thread_tables(thd);
if (sp_result != SP_KEY_NOT_FOUND &&
sp_automatic_privileges && !opt_noacl &&
@@ -4362,14 +4324,6 @@ error:
res= TRUE;
finish:
if (thd->global_read_lock.has_protection())
{
/*
Release the protection against the global read lock and wake
everyone, who might want to set a global read lock.
*/
thd->global_read_lock.start_waiting_global_read_lock(thd);
}
DBUG_ASSERT(!thd->in_active_multi_stmt_transaction() ||
thd->in_multi_stmt_transaction_mode());
@@ -4405,6 +4359,11 @@ finish:
close_thread_tables(thd);
thd_proc_info(thd, 0);
#ifndef DBUG_OFF
if (lex->sql_command != SQLCOM_SET_OPTION && ! thd->in_sub_stmt)
DEBUG_SYNC(thd, "execute_command_after_close_tables");
#endif
if (stmt_causes_implicit_commit(thd, CF_IMPLICIT_COMMIT_END))
{
/* No transaction control allowed in sub-statements. */
@@ -4430,6 +4389,10 @@ finish:
*/
thd->mdl_context.release_transactional_locks();
}
else if (! thd->in_sub_stmt)
{
thd->mdl_context.release_statement_locks();
}
DBUG_RETURN(res || thd->is_error());
}
@@ -5899,7 +5862,8 @@ TABLE_LIST *st_select_lex::add_table_to_list(THD *thd,
ptr->next_name_resolution_table= NULL;
/* Link table in global list (all used tables) */
lex->add_to_query_tables(ptr);
ptr->mdl_request.init(MDL_key::TABLE, ptr->db, ptr->table_name, mdl_type);
ptr->mdl_request.init(MDL_key::TABLE, ptr->db, ptr->table_name, mdl_type,
MDL_TRANSACTION);
DBUG_RETURN(ptr);
}