Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0 (Mac OS X Mail 6.2 \(1499\))
Subject: Re: Have: Seq Scan - Want: Index Scan - what am I doing wrong?
From: Chris Ruprecht <chris@cdrbill.com>
In-Reply-To: <9BB26B73-2824-4C97-92B5-5E059108C071@gmail.com>
Date: Tue, 16 Oct 2012 20:19:43 -0400
Cc: "pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>
Content-Transfer-Encoding: quoted-printable
Message-Id: <2A64CF51-CC89-4A0A-89CC-49D7ABFD4603@cdrbill.com>
References: <91417DD6-C9FF-40EA-8EA9-87DF7E696927@cdrbill.com>
	<9BB26B73-2824-4C97-92B5-5E059108C071@gmail.com>
To: Evgeny Shishkin <itparanoia@gmail.com>


On Oct 16, 2012, at 20:01 , Evgeny Shishkin <itparanoia@gmail.com> =
wrote:

> Selecting 5 yours of data is not selective at all, so postgres decides =
it is cheaper to do seqscan.=20
>=20
> Do you have an index on patient.dnsortpersonnumber? Can you post a =
result from=20
> select count(*) from patient where dnsortpersonnumber =3D '347450'; ?
>=20

Yes, there is an index:

"Aggregate  (cost=3D6427.06..6427.07 rows=3D1 width=3D0)"
"  ->  Index Scan using patient_pracsortpatientnumber on patient  =
(cost=3D0.00..6427.06 rows=3D1 width=3D0)"
"        Index Cond: (dnsortpersonnumber =3D '347450'::text)"


In fact, all the other criteria is picked using an index. I fear that =
the >=3D and <=3D on the timestamp is causing the issue. If I do a "=3D" =
of just one of them, I get an index scan. But I need to scan the entire =
range. I get queries like "give me everything that was entered into the =
system for this patient between these two dates". A single date wouldn't =
work.=