Re: Re: PANIC: invalid index offnum: 186 when processing BRIN indexes in VACUUM

Tom Lane wrote:
> Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> writes:
> > Tom Lane wrote:
> >> Do we still need the complication in brinsummarize to discriminate
> >> against the last partial range? Now that the lock consideration
> >> is gone, I think that might be a wart.
>
> > In the case of VACUUM, it's not desirable to create a summarization for
> > the last partial range, because if the table is still being filled, that
> > would slow down the insertion process.
>
> Hm. Okay, but you should change the comment then, because "we do not want
> to spend one RelationGetNumberOfBlocks call" is a pretty weak reason.

Changed.

> Also, I think I would accept that argument for autovacuum, but maybe
> not so much for a manual vacuum. Maybe you should drive it off
> IsAutovacuumWorker rather than which operation is being done.

I think your argument is sensible for some uses (where people run manual
VACUUM after loading data) but not others (where people just use manual
VACUUM in place of autovacuuming -- because they don't trust autovac, or
the schedule isn't convenient, or whatever other reason). I've seen
both things being done in production. If we do as you suggest, there is
no way to do the manual vacuum without summarizing the partial also; the
approach of doing the partial only in brin_summarize_new_values lets the
user choose what to do.

Once upon a time I thought about adding a reloption to let the user
choose what to do, but I never got around to writing a patch.

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services