Re: DB issues with WPMU

[Owen Taylor <otaylor redhat com>]

On Tue, 2009-10-27 at 14:49 +1100, Jeff Waugh wrote:
> <quote who="Owen Taylor">
> 
> >  create index wp_1_options_autoload on wp_1_options (autoload);
> > 
> > And sure enough 'show processlist' suddenly went from 90+ active
> > connections to just a couple.
> 
> Great, thanks.
> 
> > Even with the index added, caching feed data in the options table seems
> > weird and perverse. I wonder if it is expected that for a installation the
> > size of blogs.gnome.org some other object cache is installed, and the
> > default object cache behavior is just a fallback to get something going
> > quickly.
> 
> Yeah, using the database as the persistent cache for feeds and such is there
> mostly for "run a single WordPress easily without doing insane stuff on
> every page load" use cases.
> 
> I'll add the APC-based object cache support to WPMU, and watch how it goes
> to nut out a sensible size for the APC user cache.

Sounds good.

> Ultimately, I'd love to put blogo on another machine, so its rather unique
> abuse of Apache and MySQL -- compared to anything else we run -- can be
> segmented away from more, uh, critical infrastructure. :-) Thoughts on this?

My idea for this would probably be:

 - Put the web front end in a smallish VM

 - Run multiple mysqld instances on drawable.gnome.org. Bugzilla isn't
   stressing this machine out even a bit, and we could easily host more
   databases in the same mysqld instance, but it would be nice to
   segment connection and other resources limits.

   Running multiple mysqld instances will take a fair bit of effort and
   configuration, so I don't think it's a short-term project compared
   to some of the other outstanding work.

- Owen