Hi all,

The DB is a 3 nodes galera prodcution cluster with 2 data nodes and 1 arbitrator.

Encountered an incident that many SQL hung at one point. The SQL stopped at "Sending data" / "Updating".

We cannot figured out resason. The final action is to restart DB node to resume.

During the review, we found that binlog content has tranaction not in time order.

Xid = 200421161 timestamp is 1:50:07 but it is earlier than
Xid = 200421179 timestamp is 1:49:42

The DB has 4 Galera slave thread. Is it related?

Any problem of this behavior?

In case we need to recovery db to 1:50:00, will Xid = 200421179 transaction be applied? We worry recovery will stop at Xid = 200421161 since its time is 1:50:07.

We captured processlist every 1min. In the first checking, time is 01:50:01 and found oldest pending SQL waited for 19s. Thus, old SQL started at 1:49:42. This matched with Xid = 200421179.

Kindly help.

attached files:
binlog-mask.txt - binlog content extracted
processlist.txt - process list of node 1
mariadb.node1.cnf - db node 1 my.cnf
mariadb.node2.cnf - db node 2 my.cnf

Regards,
William Wong