[MXS-4209] KILL command doesn't work correctly if persistent connections are enabled Created: 2022-07-19  Updated: 2022-10-11  Resolved: 2022-07-26

Status: Closed
Project: MariaDB MaxScale
Component/s: Core
Affects Version/s: 2.5.20, 6.4.1
Fix Version/s: 2.5.22, 6.4.2

Type: Bug Priority: Major
Reporter: markus makela Assignee: markus makela
Resolution: Fixed Votes: 0
Labels: None

Issue Links:
Problem/Incident
causes MXS-4283 Race condition in KILL command proces... Closed

 Description   

With persistpoolmax enabled, a KILL query will end up hanging on the connection that sends it.



 Comments   
Comment by markus makela [ 2022-07-26 ]

This happened because the old mechanism relied on the EPOLLHUP that the COM_QUIT would end up generating. As this event never arrives when the persistent connection pool is enabled, the KILL execution hanged.

The fix to this was to backport the LocalClient notification mechanism from MaxScale 6 into MaxScale 2.5 and use that to signal the client when the response to the KILL command was complete. This also improves the overall performance of the KILL commands from older versions as there's no need to poll for the completion anymore.

Generated at Thu Feb 08 04:26:59 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.