Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
N/A
-
None
Description
Two examples follow. I have a few more if needed.
Case 1
|
bb-10.6-monty-innodb 790a52644c |
2021-05-11 19:04:37 0 [Note] InnoDB: Starting crash recovery from checkpoint LSN=54303,54303
|
2021-05-11 19:04:41 0 [Note] InnoDB: Starting final batch to recover 2264 pages from redo log.
|
2021-05-11 19:04:44 0 [ERROR] InnoDB: Cannot replay rename of tablespace 81 from './test/tt6.ibd' to './test/table75_innodb_int_autoinc.ibd' because the target file exists
|
2021-05-11 19:04:44 0 [ERROR] InnoDB: Plugin initialization aborted at srv0start.cc[1473] with error Data structure corruption
|
In the case above, the test was running concurrent DDL in 4 threads and was sigkilled in the process. The last executed statements in each thread, according to the general log, were
28 Query DROP TABLE `table53_innodb_int_autoinc`, `table76_innodb_int_autoinc`, vv2 |
30 Query ALTER TABLE `table89_innodb_int_autoinc` DROP KEY `ncol4`, ADD KEY IF NOT EXISTS n ( `ncol4` ) COMMENT 'pad' |
31 Query UNLOCK TABLES
|
29 Query RENAME TABLE `t15` TO tt6 |
The logs, datadirs and rr profiles are available.
Case 2
|
bb-10.6-monty-innodb 790a52644c |
2021-05-11 17:59:28 0 [Note] InnoDB: Starting crash recovery from checkpoint LSN=54303,54303
|
2021-05-11 17:59:29 0 [Note] InnoDB: Starting final batch to recover 456 pages from redo log.
|
2021-05-11 17:59:30 0 [ERROR] InnoDB: Cannot replay rename of tablespace 13 from './test/tt11.ibd' to './test/t5_innodb.ibd' because the target file exists
|
2021-05-11 17:59:30 0 [ERROR] InnoDB: Plugin initialization aborted at srv0start.cc[1473] with error Data structure corruption
|
In the case above, the test was running various DDL in a single thread in PS protocol, got sigkilled upon running a query, and failed with the error above upon recovery.
The logs and the data directories are available. Unfortunately, in PS protocol the last statement may be missing from the general log.
rr profiles are also available, but the profile before crash seems to be broken somehow.
So, this case is better than the previous one in the sense that there was only one thread involved, but worse in everything else.
I think I've seen similar failures on previous releases without atomic DDL, so it's possibly not anything new, but I would expect it to be gone with atomic DDL. Strangely, I didn't see them in previous tests on atomic DDL, but now they have suddenly appeared in bulk.
Attachments
Issue Links
- relates to
-
MDEV-17567 Atomic DDL
-
- Closed
-
-
-
- Closed
-