percona-cluster: table_cache is not configurable

[Nobuto Murata]

Currently percona-cluster charm uses "table_cache = 512" put in templates/my.cnf.

In a kind of large OpenStack deployment, "table_cache = 512" may not be enough. With tens of compute nodes and hundreds of instances, I saw 884 open tables at maximum.

[Nobuto Murata (nobuto)]

Note: I didn't take benchmark between small and large table_cache.

[Mario Splivalo (mariosplivalo)]

Nobuto, do you have any tests to prove that table_cache needs to be larger?

Can you provide output of 'mysqladmin status' from a busy/large OpenStack deployment? Also, I'd be intereste in output of this command:

SHOW GLOBAL STATUS LIKE 'Opened_tables';

possibly repeated few times with 5-10 minutes of delay.

While 512 may seem low, default can safely be set to 4k or even 8k, but rarely need to go over it. There is usually less than 300 tables in typical OpenStack deployment, therefore 512 should be ok.

[Nobuto Murata (nobuto)]

@Mario,

I don't have an access to the environment anymore, but as I commented on #1 munin monotoring graph output[1] by mysql plugin[2] shows around 800 as open tables for 3 days.

[1] https://launchpadlibrarian.net/192677209/mysql_files_tables-pinpoint%3D1418357068%2C1418632684.png
[2] http://bazaar.launchpad.net/~ubuntu-branches/ubuntu/trusty/munin/trusty/view/head:/plugins/node.d/mysql_.in