Re: reloption to prevent VACUUM from truncating empty pages at the end of relation

On Tue, Mar 5, 2019 at 5:11 AM Andres Freund <andres(at)anarazel(dot)de> wrote:
>
> Hi,
>
> On 2019-03-04 20:03:37 +0000, Bossart, Nathan wrote:
> > On 3/3/19, 9:23 PM, "Masahiko Sawada" <sawada(dot)mshk(at)gmail(dot)com> wrote:
> > > FWIW, I agree that we have options for vacuum as vacuum
> > > command options. But for reloptions, I think if the persistence the
> > > setting could be problematic we should not. According to the
> > > discussions so far, I think VACUUM_SHRINK_ENABLED is the one option
> > > that can be available as both vacuum command option and reloptions.
> > > But I'm not sure there is good use case even if we can set
> > > DISABLE_INDEX_CLEANUP as reloptions.
> >
> > +1
> >
> > The DISABLE_INDEX_CLEANUP option is intended to help avoid transaction
> > ID wraparound and should not be used as a long-term VACUUM strategy
> > for a table.
>
> I'm not quite convinced this is right. There's plenty sites that
> practically can't use autovacuum because it might decide to vacuum the
> 5TB index because of 300 dead tuples in the middle of busy periods. And
> without an reloption that's not controllable.

I understood the use case. I'm inclined to add DISABLE_INDEX_CLEANUP
as a reloption.

It's an improvement but it seems to me that the specifying a threshold
or scale factor would be more useful for that case than just turning
on and off. It's something like autovaucum_index_vacuum_scale_factor,
0 by default means always trigger index vacuuming and -1 means never
trigger.

Regards,

--
Masahiko Sawada
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center