database/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-07-29 05:21:33 +03:00
Code Activity

MDEV-31655: Parallel replication deadlock victim preference code errorneously removed

Browse Source 
Restore code to make InnoDB choose the second transaction as a deadlock
victim if two transactions deadlock that need to commit in-order for
parallel replication. This code was erroneously removed when VATS was
implemented in InnoDB.

Also add a test case for InnoDB choosing the right deadlock victim.
Also fixes this bug, with testcase that reliably reproduces:

MDEV-28776: rpl.rpl_mark_optimize_tbl_ddl fails with timeout on sync_with_master

Note: This should be null-merged to 10.6, as a different fix is needed
there due to InnoDB locking code changes.

Signed-off-by: Kristian Nielsen <knielsen@knielsen-hq.org>
This commit is contained in:
Kristian Nielsen
2023-07-11 00:31:29 +02:00
parent 920789e9d4
commit 900c4d6920
12 changed files with 393 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
sql/sql_class.cc
View File

@ -5272,6 +5272,49 @@ thd_need_ordering_with(const MYSQL_THD thd, const MYSQL_THD other_thd)
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.
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;
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;
/* 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);