Message-ID: <4F99AFBA.4090605@dunslane.net>
Date: Thu, 26 Apr 2012 16:27:38 -0400
From: Andrew Dunstan <andrew@dunslane.net>
User-Agent: Mozilla/5.0 (X11; Linux x86_64;
	rv:11.0) Gecko/20120329 Thunderbird/11.0.1
MIME-Version: 1.0
To: Tom Lane <tgl@sss.pgh.pa.us>
CC: Thomas Kellerer <spam_eater@gmx.net>,
 pgsql-performance@postgresql.org
Subject: Re: query optimization
References: <alpine.DEB.2.02.1204241008570.9591@krichy.tvnetwork.hu>
	<12228.1335467838@sss.pgh.pa.us> <jnc860$q30$1@dough.gmane.org>
	<13222.1335470935@sss.pgh.pa.us>
In-Reply-To: <13222.1335470935@sss.pgh.pa.us>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit


On 04/26/2012 04:08 PM, Tom Lane wrote:
> Thomas Kellerer<spam_eater@gmx.net>  writes:
>> Tom Lane wrote on 26.04.2012 21:17:
>>> Um ... did you analyze all the tables, or just some of them?  I get
>>> sub-millisecond runtimes if all four tables have been analyzed, but it
>>> does seem to pick lousy plans if, say, only a and b have been analyzed.
>> Here it's similar to Richard's experience:
>> Before analyzing the four tables, the first statement yields this plan:
>> [ merge joins ]
>> This continues to stay the plan for about 10-15 repetitions, then it turns to this plan
>> [ hash joins ]
> Hmm.  I see it liking the merge-join plan (with minor variations) with
> or without analyze data, but if just some of the tables have been
> analyzed, it goes for the hash plan which is a good deal slower.  The
> cost estimates aren't that far apart though.  In any case, the only
> reason the merge join is so fast is that the data is perfectly ordered
> in each table; on a less contrived example, it could well be a lot
> slower.
>

It's not so terribly contrived, is it? It's common enough to have tables 
which are append-only and to join them by something that corresponds to 
the append order (serial field, timestamp etc.)

cheers

andrew