From: Andreas Joseph Krogh <andreak@officenet.no>
Organization: OfficeNet AS
To: pgsql-sql@postgresql.org,
 depesz@depesz.com
Subject: Re: Index usage in order by with multiple columns in order-by-clause
Date: Sat, 11 Aug 2007 21:45:14 +0200
User-Agent: KMail/1.9.6
References: <200708101653.13011.andreak@officenet.no>
	<20070811190521.GA7915@depesz.com>
In-Reply-To: <20070811190521.GA7915@depesz.com>
MIME-Version: 1.0
Content-Type: text/plain;
  charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: inline
Message-Id: <200708112145.14339.andreak@officenet.no>

On Saturday 11 August 2007 21:05:22 hubert depesz lubaczewski wrote:
> On Fri, Aug 10, 2007 at 04:53:12PM +0200, Andreas Joseph Krogh wrote:
> > I have the following test-case:
> >
> > CREATE TABLE test(
> > name varchar PRIMARY KEY,
> > value varchar NOT NULL,
> > created timestamp not null
> > );
> >
> > create index test_lowernamevalue_idx ON test ((lower(name) ||
> > lower(value))); create index test_lowernamevaluecreated_idx ON test
> > ((lower(name) || lower(value)), created);
> > andreak=3D# EXPLAIN ANALYZE select * from test order by lower(name) ||
> > lower(value) ASC, created DESC;
> >                                                QUERY PLAN
> > -----------------------------------------------------------------------=
=2D-
> >------------------------------- Sort  (cost=3D60.39..62.32 rows=3D770
> > width=3D72) (actual time=3D0.034..0.034 rows=3D0 loops=3D1)
> >    Sort Key: (lower((name)::text) || lower((value)::text)), created
> >    ->  Seq Scan on test  (cost=3D0.00..23.47 rows=3D770 width=3D72) (ac=
tual
> > time=3D0.004..0.004 rows=3D0 loops=3D1)
> >  Total runtime: 0.123 ms
> > (4 rows)
> > In my application I often have a need to sort by more than 3 columns, so
> > I'm really wondering if there is a way to make sorting of multiple
> > columsn (each which may have different sort-order) use an index?
> > Preferrably without having to create 2^N indexes.
>
> first of all - you can try with separate indexes on lower()||lower(),
> and created.
>
> then - you can use a trick.
> create a function that will reverse order of your date (using a simple
> "-" operator)
> and then index your lower() and output of this function.
>
> you will need to modify the query, but it's perfectly doable.
>
> for example:
> create function test_ts(timestamp) returns interval as $BODY$
> begin
> return '2000-01-01 00:00:00'::timestamp-$1;
> end;
> $BODY$ language plpgsql immutable;
>
> of course this particular date is irrelevant, we just have to substract
> from something.
>
> then:
> create index test_lowernamevaluecreated_idx2 ON test ((lower(name) ||
> lower(value)), test_ts(created));
>
> and change your query to:
> select * from test order by lower(name) || lower(value) ASC,
> test_ts(created); it would show you what you need.
>
> depesz

Thanks. I actaully do have an index on lower(a) || lower(b). Then, as Tom L=
ane=20
explained, I need to have lots of indexes if I want to sort with different=
=20
ordering (ASC|DESC) on each column.

=2D-=20
Andreas Joseph Krogh <andreak@officenet.no>
Senior Software Developer / Manager
=2D-----------------------+---------------------------------------------+
OfficeNet AS            | The most difficult thing in the world is to |
Karenslyst All=C3=A9 11      | know how to do a thing and to watch         |
PO. Box 529 Sk=C3=B8yen      | somebody else doing it wrong, without       |
0214 Oslo               | comment.                                    |
NORWAY                  |                                             |
Tlf:    +47 24 15 38 90 |                                             |
=46ax:    +47 24 15 38 91 |                                             |
Mobile: +47 909  56 963 |                                             |
=2D-----------------------+---------------------------------------------+