To: Claire McLister <mclister@zeesource.net>
cc: pgsql-performance@postgresql.org
Subject: Re: JDBC/Stored procedure performance issue 
In-reply-to: <018573B2-DC52-4B98-9AB6-1ACB953B0035@zeesource.net> 
References: <018573B2-DC52-4B98-9AB6-1ACB953B0035@zeesource.net>
Comments: In-reply-to Claire McLister <mclister@zeesource.net>
	message dated "Mon, 28 Jan 2008 09:59:07 -0800"
Date: Mon, 28 Jan 2008 15:51:53 -0500
Message-ID: <6915.1201553513@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Claire McLister <mclister@zeesource.net> writes:
> When I do an EXPLAIN ANALYZE on one query that returns 3261 rows, it  
> executes in a reasonable 159ms:
> ...
> If I issue the same query over JDBC or use a PSQL stored procedure, it  
> takes over 3000 ms, which, of course is unacceptable!

I suspect that the problem is with "groupid = $1" instead of
"groupid = 57925".  The planner is probably avoiding an indexscan
in the parameterized case because it's guessing the actual value will
match so many rows as to make a seqscan faster.  Is the distribution
of groupid highly skewed?  You might get better results if you increase
the statistics target for that column.

Switching to something newer than 7.4.x might help too.  8.1 and up
support "bitmap" indexscans which work much better for large numbers
of hits, and correspondingly the planner will use one in cases where
it wouldn't use a plain indexscan.

			regards, tom lane