[MDEV-28211] Type conversion error on ALTER TABLE on Galera cluster in TOI mode Created: 2022-04-01 Updated: 2022-05-02 Resolved: 2022-04-25 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | Data Definition - Alter Table, Galera |
| Affects Version/s: | 10.2.32 |
| Fix Version/s: | N/A |
| Type: | Bug | Priority: | Minor |
| Reporter: | Hartmut Holzgraefe | Assignee: | Jan Lindström (Inactive) |
| Resolution: | Cannot Reproduce | Votes: | 0 |
| Labels: | None | ||
| Description |
|
When doing a schema change in rolling schema upgrade (RSU) mode on a galera cluster, changing the type or length of a column, it is known that errors like:
when changing data while only part of the nodes have been upgraded yet, unless slave_type_conversions=ALL_LOSSY,ALL_NON_LOSSY is set. Now we seem to have had a case where the above error occured on a cluster in TOI mode. The ALTER TABLE was using ALGORITHM=INSTANT to change the length of a VARCHAR column, and wsrep_slave_threads was set to 8, so this might be a parallel execution race condition? |
| Comments |
| Comment by Hartmut Holzgraefe [ 2022-04-01 ] |
|
I tried to reproduce this using same settings and multiple clients changing rows in the table in parallel continuously, then issue the ALTER TABLE, but could not reproduce the problem yet ... |
| Comment by Hartmut Holzgraefe [ 2022-04-22 ] |
|
problem is that I can't even be sure that this really happened in TOI mode and not RSU. Customer is rather confident that nobody on their side switched to RSU, but not 100% So no reproducible test case, and so far they were only hit by it once ... |