[MDEV-20288] Shutdown of database and stop slave do not work in MariaDB 10.3.17 when slave threads are running Created: 2019-08-08  Updated: 2019-08-08  Resolved: 2019-08-08

Status: Closed
Project: MariaDB Server
Component/s: Replication
Affects Version/s: 10.3.17
Fix Version/s: 10.2.27, 10.1.42, 10.3.18, 10.4.8

Type: Bug Priority: Major
Reporter: Volker Klasen Assignee: Unassigned
Resolution: Duplicate Votes: 0
Labels: None

Issue Links:
Duplicate
is duplicated by MDEV-20247 Replication hangs with "preparing" an... Closed

 Description   

Hi,

I just upgraded a slave of a 10.3.15 from 10.3.15 to 10.3.17. Now I cannot shutdown the database or stop the slave threads, if they are running (I need to use kill -9).

When I start the database with skip_slave_start, I can shutdown the database normally.
When I start only the IO_THREAD or only the SQL_THREAD, stop slave works.
When I start all threads, neither stopping them or shutting down the whole database works.
However, when the master is not reachable when starting the slave threads, stopping them is possible.
When the master is shut down while the slave threads are running, stopping the threads will not work. show slave status shows:

Slave_SQL_Running_State: Waiting for worker threads to be idle

Please let me know if you need further information.

Cheers,
Volker



 Comments   
Comment by Elena Stepanova [ 2019-08-08 ]

It could be related to MDEV-20247.
Do you use any non-zero value of innodb-thread-concurrency? Try to disable it and see if the problem goes away.
Also, if you get a readable stack trace of all threads from the hanging slave, we'll probably be able to tell with more certainty. But if you indeed use the value and disabling it helps, with all likelihood it's the same problem.

Comment by Volker Klasen [ 2019-08-08 ]

Yes, I do have innodb-thread-concurrency set to 64 and the issue goes away once I set it to 0. So, I guess, this is a duplicate of MDEV-20247.
As I can't provide a stack trace (without much effort), I propose to close this as duplicate and I'll wait for 10.3.18 to check again.

Thanks,
Volker

Comment by Elena Stepanova [ 2019-08-08 ]

Thanks for the update.

Comment by Elena Stepanova [ 2019-08-08 ]

Should be fixed in the scope of MDEV-20247.

Generated at Thu Feb 08 08:58:19 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.