[MDEV-24526] binlog rotate via FLUSH LOGS may obsolate binlog file for recovery too early Created: 2021-01-05 Updated: 2022-06-27 Resolved: 2021-04-21 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | Replication |
| Affects Version/s: | None |
| Fix Version/s: | 10.2.38, 10.3.29, 10.4.19, 10.5.10 |
| Type: | Bug | Priority: | Critical |
| Reporter: | Andrei Elkin | Assignee: | Andrei Elkin |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | binlog_checkpoint | ||
| Attachments: |
|
||||||||||||||||
| Issue Links: |
|
||||||||||||||||
| Description |
|
There's a race between FLUSH LOGS and binlog-group commit for Binlog-checkpoint notification which must be done *after* the committed group transactions' affected A test case follows with a patch to be committed soon. |
| Comments |
| Comment by Andrei Elkin [ 2021-01-05 ] |
|
A relate issue is mentioned/linked. |
| Comment by Andrei Elkin [ 2021-01-07 ] |
|
alice: We need to test this and |
| Comment by Andrei Elkin [ 2021-01-07 ] |
|
Attached test is run to pass for over 10 mins. It'd be great to have a RQG-based one and run longer on master and parallel slave, and with actual crash-recovery. alice, we need such test anyway to run periodically, so efforts to that direction now |
| Comment by Sujatha Sivakumar (Inactive) [ 2021-01-13 ] |
|
Hello Andrei, Thank you for working on this issue. Code changes look good to my knowledge. https://github.com/MariaDB/server/commit/3c5d708eeb56fd656a0b270004a16c8f041802b7 Thank you. |
| Comment by Andrei Elkin [ 2021-04-21 ] |
|
2a7dd6442 pushed to 10.2. |