Hi Tom,
On Fri, Jun 10, 2005 at 01:37:54PM -0400, Tom Lane wrote:
> Josh Berkus <josh@agliodbs.com> writes:
> > Yeah. I'd prefer per-database quotas, rather than per-user quotas, which
> > seem kind of useless. The hard part is making any transaction which
> > would exceed the per-database quota roll back cleanly with a
> > comprehensible error message rather than just having the database shut
> > down.
>
> That part doesn't seem hard to me: we already recover reasonably well
> from smgrextend failures. The real difficulty is in monitoring the
> total database size to know when it's time to complain. We don't
> currently make any effort at all to measure that, let alone keep track
> of it in real time.
>
> Given that there might be lots of processes concurrently adding pages
> in different places, I don't think you could hope for an exact
> stop-on-a-dime limit, but maybe if you're willing to accept some fuzz
> it is doable ...
Well I think a fuzzy test is better than none. But I think one should be
able to calculate how much later the quota is detected as exceeded than
it is planed to be. Therefor a threshold is usefull as well (for
alerting)
Regards,
Yann