Re: Fillfactor for GIN indexes

On Mon, Jan 19, 2015 at 5:46 PM, Cédric Villemain <cedric(at)2ndquadrant(dot)com>
wrote:

> Le lundi 19 janvier 2015 08:24:08 Robert Haas a écrit :
>
> > On Sat, Jan 17, 2015 at 4:49 AM, Alexander Korotkov
>
> >
>
> > <aekorotkov(at)gmail(dot)com> wrote:
>
> > > I already wrote quite detailed explanation of subject. Let mel try
>
> > > to
>
> > > explain in shortly. GIN is two level nested btree. Thus, GIN would
>
> > > have absolutely same benefits from fillfactor as btree. Lack of
>
> > > tests showing it is, for sure, fault.
>
> > >
>
> > > However, GIN posting trees are ordered by ItemPointer and this makes
>
> > > some specific. If you have freshly created table and do
>
> > > inserts/updates they would use the end of heap. Thus, inserts would
>
> > > go to the end of GIN posting tree and fillfactor wouldn't affect
>
> > > anything. Fillfactor would give benefits on HOT or heap space
>
> > > re-usage.
>
> >
>
> > Ah, OK. Those tests clarify things considerably; I see the point now.
>
>
>
> So I do.
>
>
>
> Alexander said:
>
> 1) In order to have fully correct support of fillfactor in GIN we need to
> rewrite GIN build algorithm.
>
> 2) Without rewriting GIN build algorithm, not much can be done with entry
> tree. However, you can implement some heuristics.
>
>
>
> The patch is 2), for the posting tree only ?
>

Yes, it's just for posting tree.

------
With best regards,
Alexander Korotkov.