Date: Fri, 10 Nov 2006 12:37:00 +0530
From: Abhijit Menon-Sen <ams@oryx.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-performance@postgresql.org, arnt@oryx.com
Subject: Re: 10x rowcount mis-estimation favouring merge over nestloop
Message-ID: <20061110070700.GA2751@penne.toroid.org>
References: <20061110051245.GA1928@penne.toroid.org>
	<16547.1163139324@sss.pgh.pa.us>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <16547.1163139324@sss.pgh.pa.us>

At 2006-11-10 01:15:24 -0500, tgl@sss.pgh.pa.us wrote:
>
> it seems that "field<=12" is true for a much smaller fraction of the
> rows satisfying (part!='' or value ilike '%,%') than for the general
> population of rows in the header_fields table.

Indeed. One-sixth of the rows in the entire table match field<=12, but
only one-fifteenth of the rows matching the part/value condition also
match field<=12.

> There's been some speculation about obtaining stats on partial indexes
> as a substitute for solving the general problem of correlation stats,

Oh. So my partial index's rowcount isn't being considered at all? That
explains a lot. Ok, I'll just run the query with mergejoin and seqscan
disabled. (I can't think of much else to do to speed it up, anyway.)

Thanks.

-- ams