Re: Catching up Production from Warm Standby aftermaintenance - Please help

On Tue, Jul 7, 2009 at 11:10 AM, Scott Whitney<swhitney(at)journyx(dot)com> wrote:
> I'd like to phone in with a slightly different opinion on VACUUM FULL. Yeah,
> it should be avoided when possible, but it's not always possible. In our
> case, I've got 300ish databases backing to a single database server. Each of
> those dbs has a couple of hundred tables and a hundred or more views. The
> product (Journyx Timesheet) is pretty complex, and I find that if I do _not_
> perform a full vacuum once per week, my customer dbs start to slow down
> inordinately. Queries which would run in 1-2 seconds will run in 30-40
> seconds after a few weeks of not performing a full vacuum.

Wait, full vacuum on the whole db, or vacuum full?

> I've got autovac
> running on all dbs.
>
> Now, that could well be due to index bloat with complex indexes, or it could
> be due to a variety of other factors, but also my pg_clog directory does not
> clear out, but continues to create new clog segments. Running my weekly
> vac-full-analyze resolves that problem for me. This might not be the case
> for you if you have a less complex schema, especially noting how you say you
> use it.

You likely have very long running transactions. Look for idle in
transaction queries in the pg_stat_activity table.

It may be that right now vacuum full is the only fix but if you can
identify a reason regular vacuum isn't working you could eliminate the
need for vacuum full.

> I _think_ autovacuum, somewhere around early 8.x resolves the transaction
> wrap-around issues, but someone else should verify that.

Ayup.