Re: Berserk Autovacuum (let's save next Mandrill)

On Sun, Mar 31, 2019 at 1:11 AM Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
>
> On Wed, Mar 27, 2019 at 5:32 PM Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> wrote:
> > * certain tables would have some sort of partial scan that sets the
> > visibility map. There's no reason to invoke the whole vacuuming
> > machinery. I don't think this is limited to append-only tables, but
> > rather those are just the ones that are affected the most.
>
> I think this is a really good idea, but in order for it to work well I
> think we would need to have some kind of estimate of autovacuum
> pressure.
>
> If we know that we're currently fairly on top of things, and there is
> not much for autovacuum to do, periodically vacuuming a chunk of some
> table that has a lot of unset visibility-map bits is probably a good
> idea. However, we can't possibly guess how aggressively to do this if
> we have no idea how long it's going to be before we need to vacuum
> that table for real. If the number of XIDs remaining until the table
> gets a wraparound vacuum is X, and the number of XIDs being consumed
> per day is Y, we can estimate that in roughly X/Y days, we're going to
> need to do a wraparound vacuum. That value might be in the range of
> months, or in the range of hours.
>
> If it's months, we probably want limit vacuum to working at a pretty
> slow rate, say 1% of the table size per hour or something. If it's in
> hours, we need to be a lot more aggressive. Right now we have no
> information to tell us which of those things is the case, so we'd just
> be shooting in the dark.

Sawada-san presented some ideas in his PGCon 2018 talk that may be related.

https://www.pgcon.org/2018/schedule/attachments/488_Vacuum_More_Efficient_Than_Ever

(slide 32~)

Thanks,
Amit