On Mon, 2021-09-06 at 12:11 -0700, Peter Geoghegan wrote:
> On Mon, Sep 6, 2021 at 9:21 AM Laurenz Albe <laurenz.albe@cybertec.at> wrote:
> > #define BYPASS_THRESHOLD_PAGES 0.02 /* i.e. 2% of rel_pages */
> >
> > So up to an additional 2% of all pages can have the all-visible bit
> > unset with "index_cleanup = auto".
> >
> > That is probably not worth worrying, right?
>
> I don't think it's worth worrying about.
>
> The bypass-index-vacuuming feature may have had a bit of a messaging
> problem. It was something we usually talked about as being about
> skipping index vacuuming, because that's what it actually does.
> However, the feature isn't really about doing less work during VACUUM.
> It's actually about doing *more* work during VACUUM -- more useful
> work. Especially setting visibility map bits. But also freezing. Now
> you can very aggressively tune VACUUM to do these things more often,
> with no fear of that being way too expensive because of index
> vacuuming that has only marginal value.
That makes sense; thanks for the detailed explanation.
Yours,
Laurenz Albe