mirror of
https://github.com/MariaDB/server.git
synced 2025-07-29 05:21:33 +03:00
Fix for BUG#8325 "Deadlock in replication thread stops replication":
in slave SQL thread: if a transaction fails because of InnoDB deadlock or innodb_lock_wait_timeout exceeded, optionally retry the transaction a certain number of times (new variable --slave_transaction_retries). sql/mysql_priv.h: new var slave_transaction_retries sql/mysqld.cc: new variable slave_transaction_retries. Plus fixing a typo. sql/set_var.cc: new global variable slave_transaction_retries (will be one per subslave, when we have multimaster). sql/slave.cc: Slave SQL thread: if a transaction fails because of InnoDB deadlock or innodb_lock_wait_timeout exceeded, optionally retry the transaction a certain number of times (--slave_transaction_retries). sql/slave.h: new RELAY_LOG_INFO::trans_retries.
This commit is contained in:
1
mysql-test/t/rpl_deadlock-slave.opt
Normal file
1
mysql-test/t/rpl_deadlock-slave.opt
Normal file
@ -0,0 +1 @@
|
||||
--innodb --loose-innodb_lock_wait_timeout=4 --slave-transaction-retries=2 --max-relay-log-size=4096
|
107
mysql-test/t/rpl_deadlock.test
Normal file
107
mysql-test/t/rpl_deadlock.test
Normal file
@ -0,0 +1,107 @@
|
||||
# See if slave restarts the transaction after failing on an InnoDB deadlock error.
|
||||
|
||||
# Note: testing what happens when too many retries is possible, but
|
||||
# needs large waits when running with --debug, so we don't do it.
|
||||
# The same way, this test may not test what is expected when run
|
||||
# under Valgrind, timings are too short then (with --valgrind I
|
||||
# (Guilhem) have seen the test manage to provoke lock wait timeout
|
||||
# error but not deadlock error; that is ok as code deals with the two
|
||||
# errors in exactly the same way.
|
||||
|
||||
source include/have_innodb.inc;
|
||||
source include/master-slave.inc;
|
||||
|
||||
connection master;
|
||||
create table t1 (a int not null, key(a)) engine=innodb;
|
||||
create table t2 (a int not null, key(a)) engine=innodb;
|
||||
create table t3 (a int) engine=innodb;
|
||||
create table t4 (a int) engine=innodb;
|
||||
sync_slave_with_master;
|
||||
|
||||
show create table t1;
|
||||
show create table t2;
|
||||
stop slave;
|
||||
|
||||
# 1) Test deadlock
|
||||
|
||||
connection master;
|
||||
begin;
|
||||
# Let's keep BEGIN and the locked statement in two different relay logs.
|
||||
let $1=200;
|
||||
disable_query_log;
|
||||
while ($1)
|
||||
{
|
||||
eval insert into t3 values( $1 );
|
||||
dec $1;
|
||||
}
|
||||
enable_query_log;
|
||||
insert into t3 select * from t2 for update;
|
||||
insert into t1 values(1);
|
||||
commit;
|
||||
save_master_pos;
|
||||
|
||||
connection slave;
|
||||
begin;
|
||||
# Let's make our transaction large so that it's slave who is chosen as
|
||||
# victim
|
||||
let $1=1000;
|
||||
disable_query_log;
|
||||
while ($1)
|
||||
{
|
||||
eval insert into t4 values( $1 );
|
||||
dec $1;
|
||||
}
|
||||
enable_query_log;
|
||||
select * from t1 for update;
|
||||
start slave;
|
||||
--sleep 3; # hope that slave is blocked now
|
||||
insert into t2 values(22); # provoke deadlock, slave should be victim
|
||||
commit;
|
||||
sync_with_master;
|
||||
select * from t1; # check that slave succeeded finally
|
||||
select * from t2;
|
||||
# check that no error is reported
|
||||
--replace_column 1 # 8 # 9 # 23 # 33 #
|
||||
--replace_result $MASTER_MYPORT MASTER_MYPORT
|
||||
show slave status;
|
||||
|
||||
# 2) Test lock wait timeout
|
||||
|
||||
stop slave;
|
||||
change master to master_log_pos=401; # the BEGIN log event
|
||||
begin;
|
||||
select * from t2 for update; # hold lock
|
||||
start slave;
|
||||
--sleep 10; # slave should have blocked, and be retrying
|
||||
commit;
|
||||
sync_with_master;
|
||||
select * from t1; # check that slave succeeded finally
|
||||
select * from t2;
|
||||
# check that no error is reported
|
||||
--replace_column 1 # 8 # 9 # 23 # 33 #
|
||||
--replace_result $MASTER_MYPORT MASTER_MYPORT
|
||||
show slave status;
|
||||
|
||||
# Now we repeat 2), but with BEGIN in the same relay log as
|
||||
# COMMIT (to see if seeking into hot log is ok).
|
||||
|
||||
set global max_relay_log_size=0;
|
||||
|
||||
# This is really copy-paste of 2) of above
|
||||
stop slave;
|
||||
change master to master_log_pos=401;
|
||||
begin;
|
||||
select * from t2 for update;
|
||||
start slave;
|
||||
--sleep 10;
|
||||
commit;
|
||||
sync_with_master;
|
||||
select * from t1;
|
||||
select * from t2;
|
||||
--replace_column 1 # 8 # 9 # 23 # 33 #
|
||||
--replace_result $MASTER_MYPORT MASTER_MYPORT
|
||||
show slave status;
|
||||
|
||||
connection master;
|
||||
drop table t1,t2;
|
||||
sync_slave_with_master;
|
Reference in New Issue
Block a user