Bug #43841 SIGHUP during dump playback segfaults mysqld
Submitted: 24 Mar 2009 19:55 Modified: 25 Apr 2009 8:27
Reporter: Adam Erickson Email Updates:
Status: No Feedback Impact on me:
None 
Category:MySQL Server Severity:S1 (Critical)
Version:5.1.30-log OS:Linux (FC5 2.6.18-53.1.21.el5 (64))
Assigned to: CPU Architecture:Any
Tags: crash, mysqld, pthreads, segfault, SIGHUP

[24 Mar 2009 19:55] Adam Erickson 
Description:
mysqld segfaults every night during our automated backup tests.  At the time of each crash there is only one connection that replays mysqldump output.

The only other thing happening at about the time of the crash is a logrotate script that sends SIGHUP to all mysqld processes.  This seems to be the most likely cause of the issue.  I have also noticed that upon receiving SIGHUP mysqld takes a while (5+ seconds) to honor the request and performance drops during this period.  That may be related to this.

Since this is a production system we are not running 5.1.32 yet, I will test against that as soon as I can.

How to repeat:
N/A

Suggested fix:
Not crash.  ;)
[24 Mar 2009 19:55] Adam Erickson 
Segfault details (### are comments by me)

Attachment: segfault.txt (text/plain), 7.33 KiB.
[24 Mar 2009 19:57] Adam Erickson 
Setting severity to reflect a server crash and possible data corruption.
[24 Mar 2009 20:23] Sveta Smirnova 
Thank you for the report.

I can not repeat described behavior. Could you please locate query from the dump file which was running when you sent SIGHUP. Also is interesting which was process state.
[24 Mar 2009 20:31] Adam Erickson 
It's different each time so there isn't any one malicious query.
[24 Mar 2009 21:40] Adam Erickson 
Testing at other times of the day, without SIGHUP essentially, does not exhibit the problem.
[25 Mar 2009 8:27] Sveta Smirnova 
Thank you for the feedback.

I can not repeat described behavior with test data. Please provide output of  `show create table prod_app.am_address`: probably this matters.
[25 Apr 2009 23:00] Bugs System 
No feedback was provided for this bug for over a month, so it is
being suspended automatically. If you are able to provide the
information that was originally requested, please do so and change
the status of the bug back to "Open".