Re: [PATCH] reduce page overlap of GiST indexes built using sorted method

Hi,

04.02.2022 03:52, Alexander Korotkov wrote:
> Thank you for the revision. This patch looks good to me. I've
> slightly adjusted comments and formatting and wrote the commit
> message.
>
> I'm going to push this if no objections.

While exploring the gist test coverage (that is discussed in [1])
I've found that this block in regress/sql/gist.sql:
-- rebuild the index with a different fillfactor
alter index gist_pointidx SET (fillfactor = 40);
reindex index gist_pointidx;

doesn't do what is declared.
In fact fillfactor is ignored by default now. I've added:
select pg_relation_size('gist_pointidx');
after reindex and get the same size with any fillfactor.
fillfactor = 40: pg_relation_size = 122880
fillfactor = 100: pg_relation_size = 122880

Though size of the index really changes on REL_14_STABLE:
fillfactor = 40: pg_relation_size = 294912
fillfactor = 100: pg_relation_size = 122880

I've found that the behavior changed after f1ea98a79. I see a comment there:
/* fillfactor ignored */
but maybe this change should be reflected on higher levels (tests, docs [2],
RN) too?

For now the fillfactor option still works for the buffering build, but maybe
it could be just made unsupported as it is not supported for gist, brin...

[1] https://www.postgresql.org/message-id/flat/20230331050726.agslrnb7e7sqtpw2%40awork3.anarazel.de
[2] https://www.postgresql.org/docs/devel/sql-createindex.html

Best regards,
Alexander