Info
In AF-196, we found a rare case that a node in quiesce mode can go into rollback, which will interrupt the shutdown and crash the server. This bug has been there since we introduced quiesce mode in 5.0 so I don't think need to rush to fix this.
One possible solution is to check if the node is already in quiesce before we transition to rollback here.
The reproducer may add more complexity to this ticket where we need to control the quiesce node to go into rollback, which could be tricky since once the node enters quiesce mode, heartbeats will be stopped so it's usually hard for a quiesce node to find a valid sync source and rollback.
Top User Comments
JIRAUSER1272726 commented on Thu, 13 Jun 2024 14:23:58 +0000:
BACKPORT-21306 will get the issue in AF-196 fixed so this ticket is no longer needed.