database/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-08-08 11:22:35 +03:00
Code Activity

Merge 10.0.14 into 10.1

Browse Source
This commit is contained in:
Sergei Golubchik
2014-10-15 12:59:13 +02:00
parent f947f73b2b f1afc003ee
commit f62c12b405
2115 changed files with 87968 additions and 80173 deletions
Download Patch File Download Diff File Expand all files Collapse all files

403
sql/sql_class.cc
View File

@@ -1,6 +1,6 @@
/*
Copyright (c) 2000, 2013, Oracle and/or its affiliates.
Copyright (c) 2008, 2013, Monty Program Ab.
Copyright (c) 2008, 2014, SkySQL Ab.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@@ -895,7 +895,6 @@ THD::THD(bool is_wsrep_applier)
accessed_rows_and_keys(0),
m_statement_psi(NULL),
m_idle_psi(NULL),
m_server_idle(false),
thread_id(0),
global_disable_checkpoint(0),
failed_com_change_user(0),
@@ -4346,6 +4345,220 @@ extern "C" int thd_rpl_is_parallel(const MYSQL_THD thd)
return thd->rgi_slave && thd->rgi_slave->is_parallel_exec;
}
/*
This function can optionally be called to check if thd_report_wait_for()
needs to be called for waits done by a given transaction.
If this function returns false for a given thd, there is no need to do any
calls to thd_report_wait_for() on that thd.
This call is optional; it is safe to call thd_report_wait_for() in any case.
This call can be used to save some redundant calls to thd_report_wait_for()
if desired. (This is unlikely to matter much unless there are _lots_ of
waits to report, as the overhead of thd_report_wait_for() is small).
*/
extern "C" int
thd_need_wait_for(const MYSQL_THD thd)
{
rpl_group_info *rgi;
if (!thd)
return false;
rgi= thd->rgi_slave;
if (!rgi)
return false;
return rgi->is_parallel_exec;
}
/*
Used by InnoDB/XtraDB to report that one transaction THD is about to go to
wait for a transactional lock held by another transactions OTHER_THD.
This is used for parallel replication, where transactions are required to
commit in the same order on the slave as they did on the master. If the
transactions on the slave encounters lock conflicts on the slave that did
not exist on the master, this can cause deadlocks.
Normally, such conflicts will not occur, because the same conflict would
have prevented the two transactions from committing in parallel on the
master, thus preventing them from running in parallel on the slave in the
first place. However, it is possible in case when the optimizer chooses a
different plan on the slave than on the master (eg. table scan instead of
index scan).
InnoDB/XtraDB reports lock waits using this call. If a lock wait causes a
deadlock with the pre-determined commit order, we kill the later transaction,
and later re-try it, to resolve the deadlock.
This call need only receive reports about waits for locks that will remain
until the holding transaction commits. InnoDB/XtraDB auto-increment locks
are released earlier, and so need not be reported. (Such false positives are
not harmful, but could lead to unnecessary kill and retry, so best avoided).
*/
extern "C" void
thd_report_wait_for(const MYSQL_THD thd, MYSQL_THD other_thd)
{
rpl_group_info *rgi;
rpl_group_info *other_rgi;
if (!thd || !other_thd)
return;
rgi= thd->rgi_slave;
other_rgi= other_thd->rgi_slave;
if (!rgi || !other_rgi)
return;
if (!rgi->is_parallel_exec)
return;
if (rgi->rli != other_rgi->rli)
return;
if (!rgi->gtid_sub_id || !other_rgi->gtid_sub_id)
return;
if (rgi->current_gtid.domain_id != other_rgi->current_gtid.domain_id)
return;
if (rgi->gtid_sub_id > other_rgi->gtid_sub_id)
return;
/*
This transaction is about to wait for another transaction that is required
by replication binlog order to commit after. This would cause a deadlock.
So send a kill to the other transaction, with a temporary error; this will
cause replication to rollback (and later re-try) the other transaction,
releasing the lock for this transaction so replication can proceed.
*/
other_rgi->killed_for_retry= true;
mysql_mutex_lock(&other_thd->LOCK_thd_data);
other_thd->awake(KILL_CONNECTION);
mysql_mutex_unlock(&other_thd->LOCK_thd_data);
}
/*
This function is called from InnoDB/XtraDB to check if the commit order of
two transactions has already been decided by the upper layer. This happens
in parallel replication, where the commit order is forced to be the same on
the slave as it was originally on the master.
If this function returns false, it means that such commit order will be
enforced. This allows the storage engine to optionally omit gap lock waits
or similar measures that would otherwise be needed to ensure that
transactions would be serialised in a way that would cause a commit order
that is correct for binlogging for statement-based replication.
Since transactions are only run in parallel on the slave if they ran without
lock conflicts on the master, normally no lock conflicts on the slave happen
during parallel replication. However, there are a couple of corner cases
where it can happen, like these secondary-index operations:
T1: INSERT INTO t1 VALUES (7, NULL);
T2: DELETE FROM t1 WHERE b <= 3;
T1: UPDATE t1 SET secondary=NULL WHERE primary=1
T2: DELETE t1 WHERE secondary <= 3
The DELETE takes a gap lock that can block the INSERT/UPDATE, but the row
locks set by INSERT/UPDATE do not block the DELETE. Thus, the execution
order of the transactions determine whether a lock conflict occurs or
not. Thus a lock conflict can occur on the slave where it did not on the
master.
If this function returns true, normal locking should be done as required by
the binlogging and transaction isolation level in effect. But if it returns
false, the correct order will be enforced anyway, and InnoDB/XtraDB can
avoid taking the gap lock, preventing the lock conflict.
Calling this function is just an optimisation to avoid unnecessary
deadlocks. If it was not used, a gap lock would be set that could eventually
cause a deadlock; the deadlock would be caught by thd_report_wait_for() and
the transaction T2 killed and rolled back (and later re-tried).
*/
extern "C" int
thd_need_ordering_with(const MYSQL_THD thd, const MYSQL_THD other_thd)
{
rpl_group_info *rgi, *other_rgi;
DBUG_EXECUTE_IF("disable_thd_need_ordering_with", return 1;);
if (!thd || !other_thd)
return 1;
rgi= thd->rgi_slave;
other_rgi= other_thd->rgi_slave;
if (!rgi || !other_rgi)
return 1;
if (!rgi->is_parallel_exec)
return 1;
if (rgi->rli != other_rgi->rli)
return 1;
if (rgi->current_gtid.domain_id != other_rgi->current_gtid.domain_id)
return 1;
if (!rgi->commit_id || rgi->commit_id != other_rgi->commit_id)
return 1;
/*
Otherwise, these two threads are doing parallel replication within the same
replication domain. Their commit order is already fixed, so we do not need
gap locks or similar to otherwise enforce ordering (and in fact such locks
could lead to unnecessary deadlocks and transaction retry).
*/
return 0;
}
/*
If the storage engine detects a deadlock, and needs to choose a victim
transaction to roll back, it can call this function to ask the upper
server layer for which of two possible transactions is prefered to be
aborted and rolled back.
In parallel replication, if two transactions are running in parallel and
one is fixed to commit before the other, then the one that commits later
will be prefered as the victim - chosing the early transaction as a victim
will not resolve the deadlock anyway, as the later transaction still needs
to wait for the earlier to commit.
Otherwise, a transaction that uses only transactional tables, and can thus
be safely rolled back, will be prefered as a deadlock victim over a
transaction that also modified non-transactional (eg. MyISAM) tables.
The return value is -1 if the first transaction is prefered as a deadlock
victim, 1 if the second transaction is prefered, or 0 for no preference (in
which case the storage engine can make the choice as it prefers).
*/
extern "C" int
thd_deadlock_victim_preference(const MYSQL_THD thd1, const MYSQL_THD thd2)
{
rpl_group_info *rgi1, *rgi2;
bool nontrans1, nontrans2;
if (!thd1 || !thd2)
return 0;
/*
If the transactions are participating in the same replication domain in
parallel replication, then request to select the one that will commit
later (in the fixed commit order from the master) as the deadlock victim.
*/
rgi1= thd1->rgi_slave;
rgi2= thd2->rgi_slave;
if (rgi1 && rgi2 &&
rgi1->is_parallel_exec &&
rgi1->rli == rgi2->rli &&
rgi1->current_gtid.domain_id == rgi2->current_gtid.domain_id)
return rgi1->gtid_sub_id < rgi2->gtid_sub_id ? 1 : -1;
/*
If one transaction has modified non-transactional tables (so that it
cannot be safely rolled back), and the other has not, then prefer to
select the purely transactional one as the victim.
*/
nontrans1= thd1->transaction.all.modified_non_trans_table;
nontrans2= thd2->transaction.all.modified_non_trans_table;
if (nontrans1 && !nontrans2)
return 1;
else if (!nontrans1 && nontrans2)
return -1;
/* No preferences, let the storage engine decide. */
return 0;
}
extern "C" int thd_non_transactional_update(const MYSQL_THD thd)
{
return(thd->transaction.all.modified_non_trans_table);
@@ -4370,9 +4583,18 @@ extern "C" bool thd_binlog_filter_ok(const MYSQL_THD thd)
return binlog_filter->db_ok(thd->db);
}
/*
This is similar to sqlcom_can_generate_row_events, with the expection
that we only return 1 if we are going to generate row events in a
transaction.
CREATE OR REPLACE is always safe to do as this will run in it's own
transaction.
*/
extern "C" bool thd_sqlcom_can_generate_row_events(const MYSQL_THD thd)
{
return sqlcom_can_generate_row_events(thd);
return (sqlcom_can_generate_row_events(thd) && thd->lex->sql_command !=
SQLCOM_CREATE_TABLE);
}
@@ -5888,23 +6110,35 @@ show_query_type(THD::enum_binlog_query_type qtype)
Constants required for the limit unsafe warnings suppression
*/
//seconds after which the limit unsafe warnings suppression will be activated
#define LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT 50
#define LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT 5*60
//number of limit unsafe warnings after which the suppression will be activated
#define LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT 50
#define LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT 10
static ulonglong limit_unsafe_suppression_start_time= 0;
static bool unsafe_warning_suppression_is_activated= false;
static int limit_unsafe_warning_count= 0;
static ulonglong unsafe_suppression_start_time= 0;
static bool unsafe_warning_suppression_active[LEX::BINLOG_STMT_UNSAFE_COUNT];
static ulong unsafe_warnings_count[LEX::BINLOG_STMT_UNSAFE_COUNT];
static ulong total_unsafe_warnings_count;
/**
Auxiliary function to reset the limit unsafety warning suppression.
This is done without mutex protection, but this should be good
enough as it doesn't matter if we loose a couple of suppressed
messages or if this is called multiple times.
*/
static void reset_binlog_unsafe_suppression()
static void reset_binlog_unsafe_suppression(ulonglong now)
{
uint i;
DBUG_ENTER("reset_binlog_unsafe_suppression");
unsafe_warning_suppression_is_activated= false;
limit_unsafe_warning_count= 0;
limit_unsafe_suppression_start_time= my_interval_timer()/10000000;
unsafe_suppression_start_time= now;
total_unsafe_warnings_count= 0;
for (i= 0 ; i < LEX::BINLOG_STMT_UNSAFE_COUNT ; i++)
{
unsafe_warnings_count[i]= 0;
unsafe_warning_suppression_active[i]= 0;
}
DBUG_VOID_RETURN;
}
@@ -5922,95 +6156,94 @@ static void print_unsafe_warning_to_log(int unsafe_type, char* buf,
}
/**
Auxiliary function to check if the warning for limit unsafety should be
thrown or suppressed. Details of the implementation can be found in the
comments inline.
Auxiliary function to check if the warning for unsafe repliction statements
should be thrown or suppressed.
Logic is:
- If we get more than LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT errors
of one type, that type of errors will be suppressed for
LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT.
- When the time limit has been reached, all suppression is reset.
This means that if one gets many different types of errors, some of them
may be reset less than LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT. However at
least one error is disable for this time.
SYNOPSIS:
@params
buf - buffer to hold the warning message text
unsafe_type - The type of unsafety.
query - The actual query statement.
TODO: Remove this function and implement a general service for all warnings
that would prevent flooding the error log.
RETURN:
0 0k to log
1 Message suppressed
*/
static void do_unsafe_limit_checkout(char* buf, int unsafe_type, char* query)
static bool protect_against_unsafe_warning_flood(int unsafe_type)
{
ulonglong now= 0;
DBUG_ENTER("do_unsafe_limit_checkout");
DBUG_ASSERT(unsafe_type == LEX::BINLOG_STMT_UNSAFE_LIMIT);
limit_unsafe_warning_count++;
ulong count;
ulonglong now= my_interval_timer()/1000000000ULL;
DBUG_ENTER("protect_against_unsafe_warning_flood");
count= ++unsafe_warnings_count[unsafe_type];
total_unsafe_warnings_count++;
/*
INITIALIZING:
If this is the first time this function is called with log warning
enabled, the monitoring the unsafe warnings should start.
*/
if (limit_unsafe_suppression_start_time == 0)
if (unsafe_suppression_start_time == 0)
{
limit_unsafe_suppression_start_time= my_interval_timer()/10000000;
print_unsafe_warning_to_log(unsafe_type, buf, query);
reset_binlog_unsafe_suppression(now);
DBUG_RETURN(0);
}
else
{
if (!unsafe_warning_suppression_is_activated)
print_unsafe_warning_to_log(unsafe_type, buf, query);
if (limit_unsafe_warning_count >=
LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT)
/*
The following is true if we got too many errors or if the error was
already suppressed
*/
if (count >= LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT)
{
ulonglong diff_time= (now - unsafe_suppression_start_time);
if (!unsafe_warning_suppression_active[unsafe_type])
{
now= my_interval_timer()/10000000;
if (!unsafe_warning_suppression_is_activated)
/*
ACTIVATION:
We got LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT warnings in
less than LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT we activate the
suppression.
*/
if (diff_time <= LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT)
{
/*
ACTIVATION:
We got LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT warnings in
less than LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT we activate the
suppression.
*/
if ((now-limit_unsafe_suppression_start_time) <=
LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT)
{
unsafe_warning_suppression_is_activated= true;
DBUG_PRINT("info",("A warning flood has been detected and the limit \
unsafety warning suppression has been activated."));
}
else
{
/*
there is no flooding till now, therefore we restart the monitoring
*/
limit_unsafe_suppression_start_time= my_interval_timer()/10000000;
limit_unsafe_warning_count= 0;
}
unsafe_warning_suppression_active[unsafe_type]= 1;
sql_print_information("Suppressing warnings of type '%s' for up to %d seconds because of flooding",
ER(LEX::binlog_stmt_unsafe_errcode[unsafe_type]),
LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT);
}
else
{
/*
Print the suppression note and the unsafe warning.
There is no flooding till now, therefore we restart the monitoring
*/
sql_print_information("The following warning was suppressed %d times \
during the last %d seconds in the error log",
limit_unsafe_warning_count,
(int)
(now-limit_unsafe_suppression_start_time));
print_unsafe_warning_to_log(unsafe_type, buf, query);
/*
DEACTIVATION: We got LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT
warnings in more than LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT, the
suppression should be deactivated.
*/
if ((now - limit_unsafe_suppression_start_time) >
LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT)
{
reset_binlog_unsafe_suppression();
DBUG_PRINT("info",("The limit unsafety warning supression has been \
deactivated"));
}
reset_binlog_unsafe_suppression(now);
}
}
else
{
/* This type of warnings was suppressed */
if (diff_time > LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT)
{
ulong save_count= total_unsafe_warnings_count;
/* Print a suppression note and remove the suppression */
reset_binlog_unsafe_suppression(now);
sql_print_information("Suppressed %lu unsafe warnings during "
"the last %d seconds",
save_count, (int) diff_time);
}
limit_unsafe_warning_count= 0;
}
}
DBUG_VOID_RETURN;
DBUG_RETURN(unsafe_warning_suppression_active[unsafe_type]);
}
/**
@@ -6022,6 +6255,7 @@ deactivated"));
void THD::issue_unsafe_warnings()
{
char buf[MYSQL_ERRMSG_SIZE * 2];
uint32 unsafe_type_flags;
DBUG_ENTER("issue_unsafe_warnings");
/*
Ensure that binlog_unsafe_warning_flags is big enough to hold all
@@ -6029,8 +6263,10 @@ void THD::issue_unsafe_warnings()
*/
DBUG_ASSERT(LEX::BINLOG_STMT_UNSAFE_COUNT <=
sizeof(binlog_unsafe_warning_flags) * CHAR_BIT);
if (!(unsafe_type_flags= binlog_unsafe_warning_flags))
DBUG_VOID_RETURN; // Nothing to do
uint32 unsafe_type_flags= binlog_unsafe_warning_flags;
/*
For each unsafe_type, check if the statement is unsafe in this way
and issue a warning.
@@ -6045,13 +6281,9 @@ void THD::issue_unsafe_warnings()
ER_BINLOG_UNSAFE_STATEMENT,
ER(ER_BINLOG_UNSAFE_STATEMENT),
ER(LEX::binlog_stmt_unsafe_errcode[unsafe_type]));
if (global_system_variables.log_warnings)
{
if (unsafe_type == LEX::BINLOG_STMT_UNSAFE_LIMIT)
do_unsafe_limit_checkout( buf, unsafe_type, query());
else //cases other than LIMIT unsafety
print_unsafe_warning_to_log(unsafe_type, buf, query());
}
if (global_system_variables.log_warnings > 0 &&
!protect_against_unsafe_warning_flood(unsafe_type))
print_unsafe_warning_to_log(unsafe_type, buf, query());
}
}
DBUG_VOID_RETURN;
@@ -6531,6 +6763,7 @@ wait_for_commit::unregister_wait_for_prior_commit2()
this->waitee= NULL;
}
}
wakeup_error= 0;
mysql_mutex_unlock(&LOCK_wait_commit);
}