Re: SOC & user quotas

On Thu, 2007-03-01 at 14:13 -0800, Joshua D. Drake wrote:
> Andrew Dunstan wrote:
> > Jeff Davis wrote:
> >>
> >> If we allowed an admin to revoke CREATE from the pg_default tablespace,
> >> then anyone who needed that feature could limit people to their own
> >> tablespace (at which point it's the filesystem's job to limit their
> >> usage). Is there some reason that we can't do this now?
> >>
> >>
> >
> >
> > Hitting a file system quota limit is likely to bring you down pretty
> > hard, isn't it?
>
> Yes, and likely corrupt the database.
>

Is it possible to make the database safe in this case? I can see why it
would be difficult to handle pg_xlog filling up (or for that matter
pg_log), but couldn't any other area be turned into just a simple ERROR?

pg_xlog is cluster-wide, so even if you had the databases in separate
quota-limited areas, that wouldn't mean pg_xlog would fill up.

If it's possible to allow disk-is-full to become a simple error, that
might satisfy this request without much effort. If it is a lot of
effort, then I retract my suggestion.

By the way, is there some good documentation about what actually happens
in different cases when the disk is full? I treat it as "never let this
happen" and I don't really know what to expect if it did happen.

Regards,
Jeff Davis