Date: Sat, 27 Aug 2005 11:29:01 +1200
From: Mark Kirkwood <markir@paradise.net.nz>
Subject: Re: Limit + group + join
In-reply-to: <9501.1125071193@sss.pgh.pa.us>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Tobias Brox <tobias@nordicbet.com>,
	pgsql-performance@postgresql.org
Message-id: <430FA5BD.2070407@paradise.net.nz>
MIME-version: 1.0
Content-type: text/plain; format=flowed; charset=ISO-8859-1
Content-transfer-encoding: 7bit
User-Agent: Mozilla Thunderbird 1.0.6 (X11/20050726)
References: <20050826002709.GK10328@tobias.lan>
	<430E85ED.60109@paradise.net.nz> <9501.1125071193@sss.pgh.pa.us>

Tom Lane wrote:
> Mark Kirkwood <markir@paradise.net.nz> writes:
> 
>>What is interesting is why this plan is being rejected...
> 
> 
> Which PG version are you using exactly?  That mistake looks like an
> artifact of the 8.0 "fuzzy plan cost" patch, which we fixed recently:
> http://archives.postgresql.org/pgsql-committers/2005-07/msg00474.php
>

Right on - 8.0.3 (I might look at how CVS tip handles this, could be 
interesting).

> But Tobias wasn't happy with 7.4 either, so I'm not sure that the fuzzy
> cost issue explains his results.
> 
> As far as the "desc" point goes, the problem is that mergejoins aren't
> capable of dealing with backward sort order, so a merge plan isn't
> considered for that case (or at least, it would have to have a sort
> after it, which pretty much defeats the point for a fast-start plan).
> I have some ideas about fixing this but it won't happen before 8.2.

That doesn't explain why the nested loop is being kicked tho', or have I 
missed something obvious? - it's been known to happen :-)...

Cheers

Mark