Horton

Re: Avoid full GIN index scan when possible

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Julien Rouhaud <rjuju123@gmail.com>
Cc: Thomas Munro <thomas.munro@gmail.com>, Marc Cousin <cousinmarc@gmail.com>, Alexander Korotkov <a.korotkov@postgrespro.ru>, Tomas Vondra <tomas.vondra@2ndquadrant.com>, Nikita Glukhov <n.gluhov@postgrespro.ru>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2019-08-01T14:37:06Z
Lists: pgsql-hackers 
Julien Rouhaud <rjuju123@gmail.com> writes:
> Attached v4 that should address all comments.

Eyeing this a bit further ... doesn't scanPendingInsert also need
to honor so->forcedRecheck?  Something along the lines of

-			tbm_add_tuples(tbm, &pos.item, 1, recheck);
+			tbm_add_tuples(tbm, &pos.item, 1, recheck | so->forcedRecheck);

at line 1837?  (Obviously, there's more than one way you could
write that.)

I'm also not exactly satisfied with the new comments --- they aren't
conveying much, and the XXX in one of them is confusing; does that
mean you're unsure that the comment is correct?

The added test case seems a bit unsatisfying as well, in that it
fails to retrieve any rows.  It's not very clear what it's
trying to test.

			regards, tom lane

Commits

  1. Avoid full scan of GIN indexes when possible