Hi All,

We run around 60 Mariadb 10.6 servers on EL in vms and varying workloads.

Some time after they updated from 10.6.12 to 10.6.13, four of the busiest started showing higher disk usage than expected. Upon investigation, /var/lib/mysql/ibdata1 was exceptionally large, up to 335Gb, and was still growing at several Gb/day.

Note: All servers have only run with “innodb_file_per_table” set and only have innodb engines for user DBs.

I cobbled together some data collection just before I went on holiday and checked the graphs today. (Pic attached – if not stripped by the list software)

Four machines growing, rest static.

Two affected machines stopped growing a week ago – and on checking, Mariadb was upgraded from 10.6.13 to 10.6.14 on them and they were rebooted then.   The two still growing are still running 10.6.13

My questions!

1. Am I right that this was a known bug in Mariadb and has been resolved with the .14 release?

2. The .14 release notes mention “InnoDB does not free UNDO after the fix of MDEV-30671 (MDEV-31234)” – is the failure to free UNDO logs a likely cause for ibdata1 to grow, given that it stores the undo log? (The two worst affected machines process many million queries a day)

3. To recover this space, I will need to do a full mysqldump, wipe and reimport of the data, correct?

I guess I’m mostly asking for reassurance that this issue is now resolved and that once I schedule in the downtime to recover the space, that it’s not going to resurface…

Thanks,

Simon