Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.1(EOL), 10.2(EOL), 10.3(EOL), 10.4(EOL)
-
None
Description
https://github.com/MariaDB/mariadb.org-tools/blob/master/buildbot/maria-master.cfg, starting from row 3255
After mtr tests it creates 3 my.cnfs for rest of tests. However, wsrep_sync_wait is not configured and default is 0 (no waiting). Later it does drop table in one node and then select on other node expecting table to be unavailable, however that drop does not happen immediately and in same time on all nodes sporadically.
Basically, when you issue a statement, in this case DROP TABLE, against one of the nodes, the client receives back the result as soon as the following holds: 1) the command is replicated and has been ordered in the cluster, and 2) obviously, the command is executed locally. So when the client receives back the response, it is possible that the command was not yet executed at all replicas (it was just replicated, and eventually all replicas will apply it in the same order). If you want to make sure that the client is going to see that DROP TABLE, then there's a mechanism in galera, called wsrep_sync_wait, that will wait for events that were replicated but not applied yet.
So, either configuration of the nodes should be changed so that wsrep-sync-wait=15 or it should be set before that select or you need to have a loop waiting for it to happen.
I have marked this as 10.4 but it should be investigated is this change also needed starting from 10.1