To: Francois Deliege <fdeliege@gmail.com>
cc: pgsql-performance@postgresql.org
Subject: Re: Distinct + Limit 
In-reply-to: 
 <16737833.463.1332881676120.JavaMail.geo-discussion-forums@pbcpw7>
References: <5833589.541.1332880671745.JavaMail.geo-discussion-forums@pbbnv8>
	<16737833.463.1332881676120.JavaMail.geo-discussion-forums@pbcpw7>
Comments: In-reply-to Francois Deliege <fdeliege@gmail.com>
	message dated "Tue, 27 Mar 2012 13:54:36 -0700"
Date: Wed, 28 Mar 2012 10:13:01 -0400
Message-ID: <22625.1332943981@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Francois Deliege <fdeliege@gmail.com> writes:
> I have the following table with millions of rows:
 
> CREATE TABLE table1
> (
>   col1 text,
>   col2 text,
>   col3 text,
>   col4 text,
>   col5 text,
>   col6 text
> )

> select col1 from table1 group by col1 limit 1;
> select distinct on (col1) col1 from table1 limit 1;

> select col1 from table1 group by col1 limit 2;
> select distinct on (col1) col1 from table1 limit 2;

> Performing any of these following queries results in a full sequential
scan, followed by a hash aggregate, and then the limit.

Well, if you had an index on the column, you would get a significantly
better plan ...

> Similarly, the following query results in a sequential scan:

> select * from table1 where col1 <> col1;

> This query is generated by the Sequel library abstraction layer in Ruby when filtering record based on a empty array of values. We fixed this by handling that case on the client side, but originally thought the server would have rewritten it and sent a empty result set.

It does not, and never will, because that would be an incorrect
optimization.  "col1 <> col1" isn't constant false, it's more like
"col1 is not null".  I'd suggest "WHERE FALSE", or "WHERE 1 <> 1"
if you must, to generate a provably false constraint.

			regards, tom lane