[MXS-1805] MaxScale may hang with multiple concurrent maxadmin calls. Created: 2018-04-16  Updated: 2020-08-25  Resolved: 2018-04-16

Status: Closed
Project: MariaDB MaxScale
Component/s: Core
Affects Version/s: None
Fix Version/s: 2.2.5

Type: Bug Priority: Critical
Reporter: Johan Wikman Assignee: Johan Wikman
Resolution: Fixed Votes: 0
Labels: None


 Description   

Some maxadmin calls, e.g. show dcbs, requires communication between the worker thread handling the maxadmin command and all other worker threads.

This is handled so that DCB information about the current worker thread (the one handling the maxadmin command) are collected directly and the DCB information about all other threads by sending a message via the epoll-mechanism to the other worker threads.

However, if multiple maxadmin commands are handled concurrently by different worker threads, this will lead to a deadlock as all but one worker handling a maxadmin command will sit in the maxadmin spinlock, thus not handling any messages sent to that worker.



 Comments   
Comment by Johan Wikman [ 2018-04-16 ]

All maxadmin connections are handled by the main thread.

Generated at Thu Feb 08 04:09:31 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.