To: Andreas Kretschmer <akretschmer@spamfence.net>
cc: pgsql-performance@postgresql.org
Subject: Re: bad execution plan for subselects containing windowing-function 
In-reply-to: <20100114173025.GA8675@tux> 
References: <20100114170318.GA7634@tux> <25448.1263489336@sss.pgh.pa.us>
	<20100114173025.GA8675@tux>
Comments: In-reply-to Andreas Kretschmer <akretschmer@spamfence.net>
	message dated "Thu, 14 Jan 2010 18:30:25 +0100"
Date: Thu, 14 Jan 2010 12:42:05 -0500
Message-ID: <25915.1263490925@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Andreas Kretschmer <akretschmer@spamfence.net> writes:
> Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I see no bug here.  Your second query asks for a much more complicated
>> computation, it's not surprising it takes longer.

> But sorry, I disagree. It is the same query with the same result. I can't see
> how the queries should return different results.

In the first query

select id, avg(value) over (partition by value) from values where id = 50 order by id;

the avg() calculations are being done over only rows with id = 50.  In
the second query

select * from (select id, avg(value) over (partition by value) from values  order by id) foo where id = 50;

they are being done over all rows.  In this particular example you
happen to get the same result, but that's just because "avg(foo) over
partition by foo" is a dumb example --- it will necessarily just yield
identically foo.  In more realistic computations the results would be
different.

			regards, tom lane