Gcache.page.* files constantly generated on the PXC 5.6.51

MySQL & MariaDB Percona XtraDB Cluster 5.x
1

Hello,

We have a 3-node Percona XTraDB Galera Cluster configuration:

5.6.51-91.0-56-log 
wsrep_provider_version=3.46(r0ef0d79) 
gcache.size = 1G

Files “gcache.page.*” are generated almost every minute and not cleaned which causes high disk space usage. To overcome issue we need to shutdown service and delete files manually.

I don’t see big transactions (large write-sets) which should trigger creation of the gcache.page.* files. Parameter gcache.keep_pages_count does not have any effect on the file generation/retention.

Has anybody had similar experiences with generating “gcache.page.*” files?
How to troubleshoot or resolve this issue?

p.s.
I understand that 5.6 is very old version and that we should upgrade, but still if anybody has any (other) advice I would be grateful.

Best regards,
Marko

1 Like
2

Hello @msutic,
Unfortunately, 5.6 is dead, EOL’d. Any bugs discovered won’t get fixed. There have been many improvements in the Galera library not only between 5.6->5.7, but also from 5.7->8.0. Your best, and honestly, only option, is to upgrade to latest 5.7.* and verify the issue still exists.

2 Likes
3

Thank you for the answer.
We plan upgrade from 5.6 in the near future but till then need to solve issues on 5.6.

It looks like problem with generating gcache.page* files is solved resizing gcache 1GB → 5GB.
We’ll see now after a few days.

Best regards,
Marko

1 Like
4

Take a look at this post that explains the behavior you are seeing: All You Need to Know About GCache (Galera-Cache) - Percona Database Performance Blog

2 Likes
5

I have seen this document and this is why I decided to go with larger gcache.
Still, this behavior does not look good to me. We don’t have so big transactions (write-sets) to cause allocation of additional pages few times per minute.
Also gcache.keep_pages_size does not have any effect on the number of this files.

But, we are on a very old version ant it is best to upgrade and not to troubleshoot further.
We have found a solution (large gcache) and this is good enough for now.

1 Like