server crash when slow query log is on with additional variables

Description

With the following cnf:

 

The server crash on a loop. Error log msg:

 

while running sysbench with the following params:

And PMM monitoring with local pmm-client in the server

Environment

None

Activity

Show:

Dmitry Lenev February 5, 2024 at 1:16 PM

PR fixing this issue has been pushed to release branch and should appear in Percona Server 8.0.36.

Note the description of the problem from the PR:
When slow log was enabled with --long_query_time option set to 0 (or close value) and with --log_slow_verbosity option including "query_info" mode, server might have crashed when a protocol command other than normal statement execution (COM_QUERY or mysql_query() in C API parlance) was executed in a connection which earlier was used to run some statement. For example, this sometimes happened for COM_QUIT command issue during graceful disconnect.

Daniel Guzman Burgos February 1, 2024 at 6:56 PM

tried by removing “query_info” from log_slow_verbosity and can confirm that workaround works

Dmitry Lenev January 30, 2024 at 11:14 AM

Hello!

Quick crash analysis shows that this seems to be Percona specific issue which is regression from .

So one possible way to temporarily workaround this crash should be removing query_info element from value of --log_slow_verbosity option to disable functionality implemented by PS-7046. If this is not possible, then one could try to increase --long_query_time value from 0 as an alternative (to avoid triggering slow log for commands like COM_QUIT and so on).

Aaditya Dubey January 24, 2024 at 4:34 PM

Hi

Thank you for the report.
Verified as described:

sending the concern to engineering for further review and updates.

Done

Details

Assignee

Reporter

Needs QA

Yes

Time tracking

1d 4h logged

Sprint

Affects versions

Priority

Smart Checklist

Created January 24, 2024 at 1:53 PM
Updated April 15, 2024 at 12:02 PM
Resolved February 5, 2024 at 1:13 PM