|
Since upgrading from MariaDB 10.2.25 to version 10.3.16 we started getting the error bellow, followed by a restart:
mysqld: 2022-07-27 8:38:51 4461671 [ERROR] [FATAL] InnoDB: Unknown error Required history data has been deleted
|
mysqld: 220727 8:38:51 [ERROR] mysqld got signal 6 ;
|
This happens randomly, recently it wasn't that frequent anymore, but it's still there.
We initially logged our problem in MDEV-21100 but as that's linked to Galera, we've been recommended to log this separately.
I've attached here a few stack traces from when we first started seeing this. These are not all crashes but just a few of them.
my.cnf is the configuration we're using for sometime now, not much has changed there over the time.
This (db10) is the "main" master, in classic master-master setup, where the other server is kept as a hot copy and doesn't send any data over.
The table `jos_jobs` is pretty basic and has pretty dynamic content, records are added and removed frequently, as it's used as a sort of job queue.
That's the odd thing, this table was re-created a few times until now (DROP/CREATE - auto_increment reset) and we also run OPTIMIZE on it daily (defragmentation). So it's strange we're getting errors on accessing InnoDB "history" data on a table with very dynamic content which doesn't keep records for too long stored in it.
I've tried to replicate this by running the same set queries for a longer period of time on a similar instance but failed to replicate the crash, so I'm stuck there.
This has been sticking with us even after upgrading to MariaDB 10.4. Upgrades were done incrementally (10.2 -> 10.3 -> 10.4), even for the minor versions (10.4.23 -> 10.4.24 -> 10.4.25), and we didn't have any problems during the upgrades.
|
jos_jobs-schema.sql
mariadb-10.3.18_crash_2019_10_11-1.txt