[MDEV-30267] All queries in all databases are suspended, but the db server can be connected Created: 2022-12-19  Updated: 2023-08-07  Resolved: 2023-08-07

Status: Closed
Project: MariaDB Server
Component/s: None
Affects Version/s: 10.5.13
Fix Version/s: N/A

Type: Bug Priority: Critical
Reporter: HaoGuo Assignee: Marko Mäkelä
Resolution: Incomplete Votes: 0
Labels: hang, suspended
Environment:

docker image tag: mariadb:10.5
host: centos

Attachments: Text File mariadb-error.log    

 Description   

Recently, we have had several instances where the database service is not responding.

The specific situation is: all clients can still establish new connections with the database, but when any query operation is performed on any table, the query will continue forever, neither timeout nor return results.

The attachment is the error log, this problem first appeared around 2022-12-16 8:20:00

Do you have any clues on this issue, we are currently at a loss

 Comments   
Comment by Marko Mäkelä [ 2022-12-19 ]

Without having full stack traces of the mariadbd process during the time of the observed hangs, it is impossible to diagnose this. The InnoDB bug MDEV-29835 could match the reported symptoms, but there could be other causes as well.

Comment by HaoGuo [ 2022-12-20 ]

At present, it is found that before these two occurrences, the Linux system time has been set. First, the time is set to 2022/12/31 23:59:00, and the time is set back after a period of time. Can this provide any information?

Comment by HaoGuo [ 2022-12-21 ]

marko

Comment by Sergei Golubchik [ 2022-12-26 ]

We had bugs like that before, for example, MDEV-17481. So, yes, it's plausible that moving the time backwards could trigger a bug.

Comment by Marko Mäkelä [ 2023-05-08 ]

serg, thank you for pointing out MDEV-17481. In its comments I see that extra calls to os_event_set() were proposed to recover from one particular hang.

I think that this type of hangs are impossible in MariaDB Server 10.6 thanks to MDEV-21452 replacing the old InnoDB synchronization primitives. I am reluctant to touch the thread synchronization code in old releases.

Kuuga, sorry, I had missed the follow-ups. Have you experienced hangs related to setting the time backwards when using MariaDB Server 10.6 or later? You may want to wait for the MariaDB Server 10.6.13 release in order to get the fix of MDEV-29835.

Generated at Thu Feb 08 10:14:58 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.