From: Tom Lane <tgl@sss.pgh.pa.us>
To: Thomas Kellerer <spam_eater@gmx.net>
cc: pgsql-performance@lists.postgresql.org
Subject: Re: Bloom index cost model seems to be wrong
In-reply-to: <dc54f6e8-c6bf-6869-08c2-de855de470ae@gmx.net>
References: <dc54f6e8-c6bf-6869-08c2-de855de470ae@gmx.net>
Comments: In-reply-to Thomas Kellerer <spam_eater@gmx.net>
	message dated "Tue, 12 Feb 2019 16:08:25 +0100"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <2125.1549986118.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Tue, 12 Feb 2019 10:41:58 -0500
Message-ID: <2126.1549986118@sss.pgh.pa.us>
Precedence: bulk

Thomas Kellerer <spam_eater@gmx.net> writes:
> The bloom index is only used if either Seq Scan is disabled or if the ra=
ndom_page_cost is set to 1 (anything about 1 triggers a Seq Scan on my Win=
dows laptop). =


Hm.  blcostestimate is using the default cost calculation, except for

	/* We have to visit all index tuples anyway */
	costs.numIndexTuples =3D index->tuples;

which essentially tells genericcostestimate to assume that every index
tuple will be visited.  This obviously is going to increase the cost
estimate; maybe there's something wrong with that?

I notice that the bloom regression test script is only testing queries
where it forces the choice of plan type, so it really doesn't prove
anything about whether the cost estimates are sane.

			regards, tom lane