[MDEV-26099] MariaDB 10.5.10/10.5.11 Galera assertion crash Created: 2021-07-06 Updated: 2021-10-07 Resolved: 2021-10-07 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | Galera |
| Affects Version/s: | 10.5.10 |
| Fix Version/s: | 10.5.13 |
| Type: | Bug | Priority: | Critical |
| Reporter: | Enrico Kern | Assignee: | Jan Lindström (Inactive) |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | None | ||
| Attachments: |
|
||||||||
| Issue Links: |
|
||||||||
| Description |
|
With MariaDB 10.5.10 we recently see alot of full cluster crashes with assertion errors. Workload didnt change really much. The identified queries that do not read are mainly updating some minor session table or cronjob tables, none of which is anything complex. Error in mysql log:
Coredump is attached. I see in gdb:
my.cnf:
|
| Comments |
| Comment by Alice Sherepa [ 2021-07-06 ] | |||||||||||||||||||||
|
Could you please try 10.5.11 - | |||||||||||||||||||||
| Comment by Enrico Kern [ 2021-07-06 ] | |||||||||||||||||||||
|
I will try. Thank you. Sorry that i missed that | |||||||||||||||||||||
| Comment by Enrico Kern [ 2021-07-08 ] | |||||||||||||||||||||
|
Upgrade to 10.5.11 didnt help. Issue is stil present. I try to get a new coredump and will upload it here. | |||||||||||||||||||||
| Comment by Enrico Kern [ 2021-07-09 ] | |||||||||||||||||||||
|
It is basically stil the same, sure that it got fixed in 10.5.11 ?
the actual coredump is to big to upload here even zipped. Guess need to try to figure out how to replicate the crash behavior | |||||||||||||||||||||
| Comment by Mario Karuza (Inactive) [ 2021-07-15 ] | |||||||||||||||||||||
|
Hi, Can you enable WSREP debug mode with wsrep_debug=3 and try to reproduce assert. | |||||||||||||||||||||
| Comment by Enrico Kern [ 2021-07-27 ] | |||||||||||||||||||||
|
Sorry for the late response. The downgrade to 10.5.9 solved the issue for now. So there must be some breaking change between 10.5.9 and 10.5.10 in the new galera plugin. I wasnt able to reproduce a crash in time with the debug parameter, we will create a new cluster with 10.5.11 and copy traffic to reproduce the crashes, then i can hopefully provide the requested informations. | |||||||||||||||||||||
| Comment by Gabor Orosz [ 2021-09-15 ] | |||||||||||||||||||||
|
Hei! We are in the same situation after upgrading from 10.4.17 (+ Thanks and regards, | |||||||||||||||||||||
| Comment by Danilo Spinella [ 2021-09-30 ] | |||||||||||||||||||||
|
Hi there! Is there any plan to work on this issue in the near future? | |||||||||||||||||||||
| Comment by Jan Lindström (Inactive) [ 2021-09-30 ] | |||||||||||||||||||||
|
danyspin97 It will be fixed in | |||||||||||||||||||||
| Comment by Gabor Orosz [ 2021-09-30 ] | |||||||||||||||||||||
|
Hi, My observation, at least in our case, is that the issue concerns transactions that try to modify tables which are in relation with a foreign key constraint defined over a multi-byte (UTF-8) character column. Similar issues that I've found on this topic are Best regards, |