Hi Marko, all,

This patch fixes a severe performance regression introduced in 12.3.1.
Thanks a lot to Jonathan Miller for discovering this and helping track down
the root cause.

I remember we have previously talked several times about finding a way to
clean up the logic around the commit_ordered()/commit() API. This is rather
convoluted currently with multiple flags being set and cleared, and indeed
this complexity is what lead to this bug getting introduced in the first
place.

I still agree that such cleanup would be very desirable. But for this
particular issue, I think it is important to have a simple fix that works
within the current code. This is a very severe and unfortunate performance
regression in 12.3.1 RC, and it is essential to get it fixed before 12.3 GA.

This patch is a minimal fix that works within the current logic. I have
checked the code carefully and believe it is correct, and Jeb has verified
that it fixes the performance regression. I'm happy to discuss a more
thorough refactor/improvement around the commit_ordered API at any time.

Hope this helps,

 - Kristian.

Kristian Nielsen (1):
  MDEV-39277: Performance regression in 12.3.0 when using legacy binlog

 storage/innobase/handler/ha_innodb.cc | 1 +
 storage/innobase/trx/trx0trx.cc       | 4 ----
 2 files changed, 1 insertion(+), 4 deletions(-)

-- 
2.47.3

_______________________________________________
commits mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to