Content-Type: text/plain; charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 10.1 \(3251\))
Subject: Re: Slow query question
From: Andrey Povazhnyi <w0rse.t@gmail.com>
In-Reply-To: <26351.1481038435@sss.pgh.pa.us>
Date: Wed, 7 Dec 2016 12:51:07 +0300
Cc: pgsql-performance@postgresql.org
Content-Transfer-Encoding: quoted-printable
Message-Id: <17350FB2-A88E-414D-9CAB-96102394E62E@gmail.com>
References: <48E80576-4047-436C-97DD-59EA7BE4DE34@gmail.com>
 <26351.1481038435@sss.pgh.pa.us>
To: Tom Lane <tgl@sss.pgh.pa.us>
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

Tom,

Thank you for a thorough answer. We=E2=80=99ll try the 2-column index.

Regards,
Andrey Povazhnyi

> On Dec 6, 2016, at 6:33 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>=20
> Andrey Povazhnyi <w0rse.t@gmail.com> writes:
>> We=E2=80=99ve got a strange planner behavior on a query to one of our bi=
gger tables after we upgraded to postgres 9.6.1 recently.
>=20
> The basic problem with this query is that there are no good alternatives.
> The planner believes there are about 53K rows matching the WHERE
> condition.  (I assume this estimate is roughly in line with reality,
> else we have different problems to talk about.)  It can either scan down
> the "id" index and stop when it finds the 30th row matching WHERE, or
> it can use the "symbol" index to read all 53K rows matching WHERE and
> then sort them by "id".  Neither one of those is going to be speedy;
> but the more rows there are matching WHERE, the better the first way
> is going to look.
>=20
> If you're worried about doing this a lot, it might be worth your while
> to provide a 2-column index on (source, id) --- in that order --- which
> would allow a query plan that directly finds the required 30 rows as
> consecutive index entries.  Possibly this could replace your index on
> "source" alone, depending on how much bigger the 2-col index is and
> how many queries have no use for the second column.  See
> https://www.postgresql.org/docs/current/static/indexes.html
> particularly 11.3 - 11.5.
>=20
> 			regards, tom lane


--=20
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance