[MDEV-25518] mariadb crash with mysqld got signal 11 Created: 2021-04-26 Updated: 2022-01-03 Resolved: 2022-01-03 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | wsrep |
| Affects Version/s: | 10.5.9, 10.5 |
| Fix Version/s: | 10.2.41, 10.3.32, 10.4.22, 10.5.13, 10.6.5 |
| Type: | Bug | Priority: | Critical |
| Reporter: | Jaroslav | Assignee: | Jan Lindström (Inactive) |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | None | ||
| Attachments: |
|
||||||||
| Issue Links: |
|
||||||||
| Description |
|
We've seems to hit over and over mariadb crashing with following on our multi-master setup. We've tried to decrease memory limits on the nodes but seems issue still occurs.
Our specific config options:
Performance details from mysqtuner
Can you please check if this is a bug or some configuration issue? The crash file is one-time download in https://file.io/UfHRT94iSLvI |
| Comments |
| Comment by Marko Mäkelä [ 2021-04-26 ] | |||||||||||||||||||||||||||||||||||
|
jaroslav, can you please try to provide a full stack trace of the crash that would include all threads that were executing during the time of the crash? The stack trace includes a handle_manager thread that is consuming events that were produced by mysql_manager_submit() calls. Because Galera is involved, I suspect that this has something to do with | |||||||||||||||||||||||||||||||||||
| Comment by Jaroslav [ 2021-04-26 ] | |||||||||||||||||||||||||||||||||||
|
Marko it seems that our current design won't allow us to install debug version of mariadb. We use stripped version (docker)
I can't find and easy way how to do this for this kind of architecture. Is there some additional way how to get you more details from crash? | |||||||||||||||||||||||||||||||||||
| Comment by Jaroslav [ 2021-04-26 ] | |||||||||||||||||||||||||||||||||||
|
We tried to add some more debug to the core file we added as description and this is the most we could get (see attachment gdb_output). Will this be sufficient Marko/Sergei? | |||||||||||||||||||||||||||||||||||
| Comment by Sergei Golubchik [ 2021-04-28 ] | |||||||||||||||||||||||||||||||||||
|
yes, thanks. let me see what we can get from that. | |||||||||||||||||||||||||||||||||||
| Comment by Sergei Golubchik [ 2021-05-02 ] | |||||||||||||||||||||||||||||||||||
|
The important part apparently is
and the line in question is
If the highlighted line 18690 has indeed caused the crash, it means that thd pointer was invalid. It couldn't have been NULL, this was checked on the line 18681. And it was still valid on the line 18677. Generally find_thread_by_id() returns a thd with its thd->LOCK_thd_kill locked, so thd couldn't have been deleted after find_thread_by_id(), if that protection is broken we would've likely had lots of bug reports and tests would've likely detected it too. Another option is that the crash was actually caused by the line 18691, meaning thd is fine, but trx is invalid. | |||||||||||||||||||||||||||||||||||
| Comment by Jaroslav [ 2021-05-03 ] | |||||||||||||||||||||||||||||||||||
|
Thanks Sergei. Is there something we can do in this case to confirm this theory? The signal 11 seems to be ongoing for us so if you would like to see "fresh" coredump we can maybe provide those. However they were created on stripped version. | |||||||||||||||||||||||||||||||||||
| Comment by Sergei Golubchik [ 2021-12-25 ] | |||||||||||||||||||||||||||||||||||
|
This code no longer exists in MariaDB it was completely rewritten in the last release. Do you still experience crashes? | |||||||||||||||||||||||||||||||||||
| Comment by Jaroslav [ 2022-01-03 ] | |||||||||||||||||||||||||||||||||||
|
I think this one is good to be closed. Haven't seen this one to occur recently |