To: randalls@bioinfo.wsu.edu
cc: pgsql-performance@postgresql.org
Subject: Re: Performance regarding LIKE searches 
In-reply-to: <2071262140.8761269880968618.JavaMail.root@mail> 
References: <2071262140.8761269880968618.JavaMail.root@mail>
Comments: In-reply-to randalls@bioinfo.wsu.edu
	message dated "Mon, 29 Mar 2010 08:42:48 -0800"
Date: Mon, 29 Mar 2010 13:00:03 -0400
Message-ID: <16513.1269882003@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

randalls@bioinfo.wsu.edu writes:
> I can see I am hitting an index using an index that I created using the varchar_pattern_ops setting.  This is very fast and performs like I would expect.  However, when my application, GBrowse, access the database, I see in my slow query log this:

> 2010-03-29 09:34:38.083 PDT,"gdr_gbrowse_live","gdr_gbrowse_live",11649,"10.0.0.235:59043",4bb0399d.2d81,8,"SELECT",2010-03-28 22:24:45 PDT,4/118607,0,LOG,00000,"duration: 21467.467 ms  execute dbdpg_p25965_9: SELECT f.id,f.object,f.typeid,f.seqid,f.start,f.end,f.strand
>   FROM feature as f, name as n
>   WHERE (n.id=f.id AND lower(n.name) LIKE $1)
  
> ","parameters: $1 = 'Scaffold:scaffold\_163:1000..1199%'",,,,,,,

> GBrowse is a perl based application.  Looking at the duration for this query is around 21 seconds.  That is a bit long.  Does anyone have any ideas why the query duration is so different?

You're not going to get an index optimization when the LIKE pattern
isn't a constant (and left-anchored, but this is).

It is possible to get the planner to treat a query parameter as a
constant (implying a re-plan on each execution instead of having a
cached plan).  I believe what you have to do at the moment is use
unnamed rather than named prepared statements.  The practicality of
this would depend a lot on your client-side software stack, which
you didn't mention.

			regards, tom lane