Re: Template0 datfrozenxid age is 160million and progressing

On 2018-08-01 12:20:15 -0400, Alvaro Herrera wrote:
> On 2018-Aug-01, Andres Freund wrote:
>
> > On 2018-08-01 12:07:16 -0400, Tom Lane wrote:
> > > Andres Freund <andres(at)anarazel(dot)de> writes:
> > > > On 2018-08-01 10:24:24 -0400, Tom Lane wrote:
> > > >> IMO, the action you need to take is enabling autovacuum. We've
> > > >> seen many many people go down the path you are taking, and it's
> > > >> generally led to no good in the end. Manual vacuuming tends
> > > >> to miss stuff, and it cannot react adequately to activity spikes.
> > >
> > > > But it shouldn't matter here, autovacuum will start regardless, no?
> > >
> > > Sure, once it decides that emergency anti-wraparound vacuuming is
> > > necessary. I really doubt the OP wants that to happen; it's the
> > > exact opposite of non-intrusive.
> >
> > That's solely what would trigger it were autovacuum enabled, too? I've
> > complained about "emergency anti-wraparound" beeing anything but
> > emergency (they're largely unavoidable unless you manually script it),
> > but they're what happen once autovacuum_freeze_max_age is reached, and
> > that's the only trigger for vacuuming old relations independent of other
> > activity?
>
> With a small database like template0, it doesn't matter. The vacuuming
> is going to be over before OP realizes it has happened anyway.
> Certainly having it happen on a normal-sized table can become
> problematic, but presumably OP has taken steps to avoid it when
> disabling autovacuum (which is why only template0 is getting into
> trouble.)

Right.

> I think emergency vacuum should behave differently (not scan indexes,
> just apply HOT page prune and clear old XIDs/multixacts), which would
> make it much faster, but that's a separate line of thought (and of
> development).

What I'd love is for freeze_max_age triggered vacuums *not* to be
emergency vacuums. They should just be normal ones. There should be a
separate GUC that triggers the emergency bit. It's really annoying to
get a hard to kill ant-wraparound autovacuum on an insert only table,
where it's the only thing that'll trigger the autovacuum.

Greetings,

Andres Freund