Re: Autovacuum Improvements

Matthew O'Connor wrote:
> Glen Parker wrote:
> >If it isn't there somewhere already, I would ask to add:
> >
> >4) Expose all information used by autovacuum to form its decisions.
>
> You could argue that this is already there, although not easy to get at
> I suppose.  But all table threshold settings are available either in the
> pg_autovacuum relation or the defaults via GUC variables, that plus a
> little math will get the information autovacuum uses to form its decisions.

No, we currently don't expose the number of dead tuples which autovacuum
uses.

> >5) Expose a very easy way to discover autovacuum's opinion about a
> >particular table, for example "table_needs_vacuum(oid)", ignoring any
> >time constraints that may be in place.
>
> This might be a nice feature however in the presence of the much talked
> about but not yet developed maintenance window concept, I'm not sure how
>  this should work.  That is, during business hours the table doesn't
> need vacuuming, but it will when the evening maintenance window opens up.

I intend to work on the maintenance window idea for 8.3.  I'm not sure
if I'll be able to introduce the worker process stuff in there as well.
I actually haven't done much design on the stuff so I can't say.

Now, if you (Matthew, or Glen as well!) were to work on that it'll be
appreciated ;-) and we could team up.

--
Alvaro Herrera                                http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support