← Back to team overview

maria-discuss team mailing list archive

Mariadb 10.4 to 10.6 performance regression


for testing the upgrade path I did an in-place slave server (with optimistic
parallel replication) upgrade from 10.4.26 to 10.6.11, but the performance
difference is so significant/abysmal I don't even know where to look at (see
the attached QPS chart before and after upgrade - it's like 2-3x times

All the previous settings/configuration are kept.

I have various server parameters monitored but none seem to indicate why
would everything become way slower (see for example the disk write
operations also are 2-3 times lower).

Maybe someone can give an idea where to look or what to check what could
make such an impact? 

p.s. I know about https://jira.mariadb.org/browse/MDEV-29988 but I don't
think that I'm hitting it since the cpu load hasn't changed (is even lower).


Attachment: 10.4_to_10.6.png
Description: PNG image

Attachment: diskwrite.png
Description: PNG image

Follow ups