[MDEV-12794] innodb_zip.recover failed in buildbot Created: 2017-05-14  Updated: 2017-05-18  Resolved: 2017-05-18

Status: Closed
Project: MariaDB Server
Component/s: Tests
Affects Version/s: 10.2.6
Fix Version/s: 10.2.7

Type: Bug Priority: Major
Reporter: Elena Stepanova Assignee: Marko Mäkelä
Resolution: Fixed Votes: 0
Labels: None

Issue Links:
Problem/Incident
is caused by MDEV-12720 recovery fails with "Generic error" f... Closed

 Description   

http://buildbot.askmonty.org/buildbot/builders/kvm-bintar-quantal-x86/builds/7100/steps/test/logs/stdio

innodb_zip.recover 'innodb'              w4 [ fail ]  Found warnings/errors in server log file!
        Test ended at 2017-05-11 18:35:21
line
2017-05-11 18:35:19 3057604352 [ERROR] mysqld: Table './mysql/plugin' is marked as crashed and should be repaired
2017-05-11 18:35:19 3057604352 [ERROR] mysqld: Table 'plugin' is marked as crashed and should be repaired
2017-05-11 18:35:19 3057604352 [Warning] Checking table:   './mysql/plugin'
2017-05-11 18:35:19 3057604352 [ERROR] mysql.plugin: 1 client is using or hasn't closed the table properly



 Comments   
Comment by Marko Mäkelä [ 2017-05-16 ]

This test is not writing to any other than InnoDB tables. No call to mtr.add_suppression(), only a server kill+restart.
Maybe this should be fixed somewhere in the mtr infrastructure? If that is not acceptable, then I think that adding
FLUSH TABLES to the start of the test should do the trick.

Comment by Elena Stepanova [ 2017-05-16 ]

I don't think it would be right to suppress an error like that globally, it can hide too many potential problems.
We could consider adding FLUSH directly to include/restart_mysqld.inc, particularly if shutdown_timeout=0.
One problem that I can think of right away is that if restart_mysqld.inc is called concurrently with some workflow (on purpose, to test a specific scenario), FLUSH would totally change the dynamics.
Alternatively, we could add a parameter to restart_mysqld.inc, but then it might be just easier to do FLUSH directly in the test.
Maybe there can be other and better ideas, let's think about it for a while.

Comment by Marko Mäkelä [ 2017-05-18 ]

I added the FLUSH TABLES statement to the test, because that indeed seems to be the best option.

Generated at Thu Feb 08 08:00:31 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.