Re: parallel vacuum comments

On Wed, Nov 24, 2021 at 7:48 AM Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
>
> On Fri, Nov 19, 2021 at 11:25 AM houzj(dot)fnst(at)fujitsu(dot)com
> <houzj(dot)fnst(at)fujitsu(dot)com> wrote:
> >
> > 3)
> >
> > + /*
> > + * Reset all index status back to invalid (while checking that we have
> > + * processed all indexes).
> > + */
> > + for (int i = 0; i < pvs->nindexes; i++)
> > + {
> > + PVIndStats *stats = &(pvs->indstats[i]);
> > +
> > + Assert(stats->status == INDVAC_STATUS_COMPLETED);
> > + stats->status = INDVAC_STATUS_INITIAL;
> > + }
> >
> > Would it be safer if we report an error if any index's status is not
> > INDVAC_STATUS_COMPLETED ?
>
> Agreed. It'd be safer since even if some indexes are vacuumed due to a
> bug vacuum errored out rather than continue it (and cause index
> corruption).
>

I think if we want to report an error in this case, we should use elog
as this is an internal error.

--
With Regards,
Amit Kapila.