[Openstack] Slow API operations

Jay Pipes jaypipes at gmail.com
Tue Sep 3 19:42:14 UTC 2013


On 09/03/2013 03:12 PM, Clint Byrum wrote:
> Excerpts from Guilherme Russi's message of 2013-09-03 11:52:39 -0700:
>> Query OK, 502150 rows affected (32 min 2.77 sec) and nothing has changed,
>> lol.
>
> There's also indexes in Havana that help a lot, you might consider adding
> them manually:
>
> ALTER TABLE token ADD INDEX ix_token_valid (valid);
> ALTER TABLE token ADD INDEX ix_token_expires (expires);
>
> Note that a 500,000 row delete is _brutal_ on your server. We use this
> in TripleO:
>
> https://git.openstack.org/cgit/openstack/tripleo-image-elements/tree/elements/keystone/cleanup-keystone-tokens.sh
>
> It allows space in between the deletes for other things to happen,
> and also deletes in a more efficient way to not thrash around the table
> deleting things in index order.
>
> Also, if you don't need the content of your token table for audit purposes
> and you can afford the RAM, you should definitely consider switching to
> the memcached backend for tokens.

+1000. memcached is the superior option for tokens.

-jay





More information about the Openstack mailing list