launchpad-dev team mailing list archive

[Danilo Šegan <danilo@xxxxxxxxxxxxx>]

У пет, 12. 11 2010. у 20:10 +0700, Stuart Bishop пише:

> There is a lot of disk activity on the database server I think. I'm
> seeing %wait time at up to 10%, when it normally is low. poimport and
> one of the karma seem to be where the load is coming from, but it
> doesn't seem that unusual. It might just be are batch jobs having
> trouble on a database server with a fraction of the production RAM and
> probably slower disk.

Isn't part of the problem that we are running both qastaging and staging
DBs on the same server as well?  Is there some cache contention there as
well?  Before we set-up qastaging, DB cache behaviour was much more
predictable (i.e. for +translate pages, I knew that I need to load it
once and expect it to time out, and it gets all the relevant indexes
into RAM, with further refreshes on any other +translate pages working:
today, caches get purged much more frequently).

Also, if we are indeed seeing big load from poimport on qastaging, are
we perhaps running this script on both staging and qastaging?  Perhaps
it's a script that's best run on demand (I'd instead suggest cleaning up
the import queue, but having it duplicated on staging is such a useful
debugging tool that I'd hate to lose that).

Cheers,
Danilo