Re: PATCH: index-only scans with partial indexes

Thank you for polishing this.

At Tue, 29 Mar 2016 13:31:19 -0500, Kevin Grittner <kgrittn(at)gmail(dot)com> wrote in <CACjxUsNm9+tn0Hat0p4wR+sF0bfQDYMPLOEw7FyDycQ2UUrbyA(at)mail(dot)gmail(dot)com>
> I tried to whip this into shape, but there were a few areas I
> didn't feel I had the necessary understanding to feel comfortable
> taking on the committer role for it. I've cleaned it up the best I
> could, fixing whitespace and typos, eliminating an unnecessary
> addition of an include, improving C comments (at least to my eye),
> and adding an Assert where it seemed like a good idea to me.

Especially for this one,

===
@@ -2697,6 +2697,7 @@ check_partial_indexes(PlannerInfo *root, RelOptInfo *rel)
continue; /* don't repeat work if already proven OK */

have_partial = true;
+ break;
}
if (!have_partial)
return;
===

The initialization has been moved to set_rel_size so the break
can be restored.

> My own tests and those of others show performance improvements (up
> to 10x for some tests) and no significant performance degradation,
> even with attempts to create "worst case" scenarios.
>
> The only changes to the regression tests are to change an "out"
> file to eliminate re-checking the index predicate against the heap
> on every matching row, which seems like a good thing.
>
> I'm taking my name off as committer and marking it "Ready for
> Committer". If someone else wants to comment on the issues where
> Tom and Kyotaro-san still seem unsatisfied to the point where I
> can get my head around it, I could maybe take it back on as
> committer -- if anyone feels that could be a net win.

FWIW, as mentioned upthread, I added the following condition to
decline ripping index predicates from base restrictinfo without
understanding the reason to do so.

regards,

--
Kyotaro Horiguchi
NTT Open Source Software Center