To: Scott Carey <scott@richrelevance.com>
cc: Robert Haas <robertmhaas@gmail.com>, "pgsql-performance@postgresql.org
	Performance" <pgsql-performance@postgresql.org>
Subject: Re: HashJoin order,
	hash the large or small table? Postgres likes to hash the big one,
	why?
In-reply-to: <C9CB28BE.306C0%scott@richrelevance.com> 
References: <C9CB28BE.306C0%scott@richrelevance.com>
Comments: In-reply-to Scott Carey <scott@richrelevance.com>
	message dated "Wed, 13 Apr 2011 10:22:40 -0700"
Date: Wed, 13 Apr 2011 13:35:31 -0400
Message-ID: <8747.1302716131@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Scott Carey <scott@richrelevance.com> writes:
>> On Oct 27, 2010, at 12:56 PM, Tom Lane wrote:
>> Because a poorly distributed inner relation leads to long hash chains.
>> In the very worst case, all the keys are on the same hash chain and it
>> degenerates to a nested-loop join.

> A pathological skew case (all relations with the same key), should be
> _cheaper_ to probe.

I think you're missing the point, which is that all the hash work is
just pure overhead in such a case (and it is most definitely not
zero-cost overhead).  You might as well just do a nestloop join.
Hashing is only beneficial to the extent that it allows a smaller subset
of the inner relation to be compared to each outer-relation tuple.
So I think biasing against skew-distributed inner relations is entirely
appropriate.

			regards, tom lane