[MDEV-8370] mysqld got signal 11 with aggressive parallel replication Created: 2015-06-24 Updated: 2015-06-29 Resolved: 2015-06-29 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | OTHER |
| Affects Version/s: | 10.1.5 |
| Fix Version/s: | 10.1.6 |
| Type: | Bug | Priority: | Major |
| Reporter: | Jean-François Gagné | Assignee: | Sergey Vojtovich |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | None | ||
| Environment: |
$ uname -a $ free -m $ cat /proc/cpuinfo | grep "model name" | uniq -c |
||
| Attachments: |
|
||||||||||||
| Issue Links: |
|
||||||||||||
| Description |
|
Hi, I have a "mysqld got signal 11" in the logs when running aggressive parallel replication with high number of threads. When running with single threaded replication, I have no problem: I ran many times without problems with 5, 10, 20, 40, 80 and 120 thread; but when I run with 320, 240 and 160 threads, sometimes it works, and sometimes mysqld is killed. Stack trace below. Already reported similar problems in the following bugs for mysql_admin_table and execute_sqlcom_select: but I think this might be a different issue as tis one is on mysql_insert. Also, the number of Open Tables as shown by SHOW GLOBAL STATUS is much bigger with aggressive parallel replication than with single threaded replication (see image in attachment where the 1st run is with single threaded replication, and the 3 other runs are with aggressive parallel replication with number of threads of 120, 240 and 320 respectively).
JFG |
| Comments |
| Comment by Elena Stepanova [ 2015-06-24 ] |
|
|
| Comment by Jean-François Gagné [ 2015-06-24 ] |
|
I was surprised by the Open Tables difference between a run without parallel replication and with different number of threads (maybe normal, just asking). Note: all those tests start by restoring a backup and running the same binlogs, so exactly the same workload run in all tests, hence the difference in Open Tables puzzling me. |
| Comment by Sergey Vojtovich [ 2015-06-24 ] |
|
I suspect high number of tables should be expected with parallel replication, same as with high number of connecting clients. |
| Comment by Elena Stepanova [ 2015-06-24 ] |
|
Tests in |
| Comment by Sergey Vojtovich [ 2015-06-29 ] |
|
Fixed in scope of |