Message-ID: <cf12201c9c3aa4a29a5c2ed26c7853ce.squirrel@sq.gransy.com>
In-Reply-To: <4E5E3E5F.4000807@squeakycode.net>
References: <4E5D485F.8060300@squeakycode.net>
	<4E5D8F68.4010102@ringerc.id.au> <4E5E3E5F.4000807@squeakycode.net>
Date: Wed, 31 Aug 2011 16:19:36 +0200
Subject: Re: IN or EXISTS
From: "Tomas Vondra" <tv@fuzzy.cz>
To: "Andy Colson" <andy@squeakycode.net>
Cc: "Craig Ringer" <ringerc@ringerc.id.au>,
	"pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>
User-Agent: SquirrelMail/1.4.21
MIME-Version: 1.0
Content-Type: text/plain;charset=utf-8
Content-Transfer-Encoding: 8bit
Importance: Normal

On 31 Srpen 2011, 15:59, Andy Colson wrote:
> I assume:
> Buckets: 16384  Batches: 1  Memory Usage: 4531kB
>
> That means a total of 4.5 meg of ram was used for the hash, so if my
> work_mem was lower than that it would swap?  (or choose a different plan?)

Why don't you try that? Just set the work_mem to 1MB or so and run the query.

I think it'll use the same plan but multiple batches - read just part of
the inner table so that the hash table fits into work_mem, scan the outer
table etc. The downside is it'd rescan the outer table several times.

Tomas