1
0
mirror of https://github.com/MariaDB/server.git synced 2026-01-06 05:22:24 +03:00

MDEV-5914: Parallel replication deadlock due to InnoDB lock conflicts

Due to how gap locks work, two transactions could group commit together on the
master, but get lock conflicts and then deadlock due to different thread
scheduling order on slave.

For now, remove these deadlocks by running the parallel slave in READ
COMMITTED mode. And let InnoDB/XtraDB allow statement-based binlogging for the
parallel slave in READ COMMITTED.

We are also investigating a different solution long-term, which is based on
relaxing the gap locks only between the transactions running in parallel for
one slave, but not against possibly external transactions.
This commit is contained in:
unknown
2014-03-21 13:30:55 +01:00
parent a5418c5540
commit b352969118
10 changed files with 174 additions and 6 deletions

View File

@@ -1172,6 +1172,84 @@ START SLAVE SQL_THREAD;
SELECT * FROM t3 WHERE a >= 110 ORDER BY a;
--echo ***MDEV-5914: Parallel replication deadlock due to InnoDB lock conflicts ***
--connection server_2
--source include/stop_slave.inc
--connection server_1
CREATE TABLE t4 (a INT PRIMARY KEY, b INT, KEY b_idx(b)) ENGINE=InnoDB;
INSERT INTO t4 VALUES (1,NULL), (2,2), (3,NULL), (4,4), (5, NULL), (6, 6);
# Create a group commit with UPDATE and DELETE, in that order.
# The bug was that while the UPDATE's row lock does not block the DELETE, the
# DELETE's gap lock _does_ block the UPDATE. This could cause a deadlock
# on the slave.
--connection con1
SET debug_sync='commit_after_release_LOCK_prepare_ordered SIGNAL master_queued1 WAIT_FOR master_cont1';
send UPDATE t4 SET b=NULL WHERE a=6;
--connection server_1
SET debug_sync='now WAIT_FOR master_queued1';
--connection con2
SET debug_sync='commit_after_release_LOCK_prepare_ordered SIGNAL master_queued2';
send DELETE FROM t4 WHERE b <= 3;
--connection server_1
SET debug_sync='now WAIT_FOR master_queued2';
SET debug_sync='now SIGNAL master_cont1';
--connection con1
REAP;
--connection con2
REAP;
SET debug_sync='RESET';
--save_master_pos
--connection server_2
--source include/start_slave.inc
--sync_with_master
--source include/stop_slave.inc
SELECT * FROM t4 ORDER BY a;
# Another example, this one with INSERT vs. DELETE
--connection server_1
DELETE FROM t4;
INSERT INTO t4 VALUES (1,NULL), (2,2), (3,NULL), (4,4), (5, NULL), (6, 6);
# Create a group commit with INSERT and DELETE, in that order.
# The bug was that while the INSERT's insert intention lock does not block
# the DELETE, the DELETE's gap lock _does_ block the INSERT. This could cause
# a deadlock on the slave.
--connection con1
SET debug_sync='commit_after_release_LOCK_prepare_ordered SIGNAL master_queued1 WAIT_FOR master_cont1';
send INSERT INTO t4 VALUES (7, NULL);
--connection server_1
SET debug_sync='now WAIT_FOR master_queued1';
--connection con2
SET debug_sync='commit_after_release_LOCK_prepare_ordered SIGNAL master_queued2';
send DELETE FROM t4 WHERE b <= 3;
--connection server_1
SET debug_sync='now WAIT_FOR master_queued2';
SET debug_sync='now SIGNAL master_cont1';
--connection con1
REAP;
--connection con2
REAP;
SET debug_sync='RESET';
--save_master_pos
--connection server_2
--source include/start_slave.inc
--sync_with_master
SELECT * FROM t4 ORDER BY a;
--connection server_2
--source include/stop_slave.inc
SET GLOBAL slave_parallel_threads=@old_parallel_threads;
@@ -1180,7 +1258,7 @@ SET DEBUG_SYNC= 'RESET';
--connection server_1
DROP function foo;
DROP TABLE t1,t2,t3;
DROP TABLE t1,t2,t3,t4;
SET DEBUG_SYNC= 'RESET';
--source include/rpl_end.inc