X-Original-To: pgsql-performance-postgresql.org@localhost.postgresql.org Received: from localhost (unknown [200.46.204.144]) by svr1.postgresql.org (Postfix) with ESMTP id C18CE3A4805 for ; Thu, 20 Jan 2005 15:34:10 +0000 (GMT) Received: from svr1.postgresql.org ([200.46.204.71]) by localhost (av.hub.org [200.46.204.144]) (amavisd-new, port 10024) with ESMTP id 15733-05 for ; Thu, 20 Jan 2005 15:33:58 +0000 (GMT) Received: from bast.unixathome.org (bast.unixathome.org [66.11.174.150]) by svr1.postgresql.org (Postfix) with ESMTP id 13B243A47D4 for ; Thu, 20 Jan 2005 15:34:01 +0000 (GMT) Received: from wocker (wocker.unixathome.org [192.168.0.99]) by bast.unixathome.org (Postfix) with ESMTP id 724D03D37; Thu, 20 Jan 2005 10:34:02 -0500 (EST) From: "Dan Langille" To: Stephan Szabo Date: Thu, 20 Jan 2005 10:36:04 -0500 MIME-Version: 1.0 Subject: Re: index scan of whole table, can't see why Cc: pgsql-performance@postgresql.org Message-ID: <41EF8994.24822.D79F9C9@localhost> In-reply-to: <20050120072205.G38768@megazone.bigpanda.com> References: <41EF7C85.4180.D46F54D@localhost> X-mailer: Pegasus Mail for Windows (4.21c) Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Content-description: Mail message body X-Virus-Scanned: by amavisd-new at hub.org X-Spam-Status: No, hits=0 tagged_above=0 required=5 tests= X-Spam-Level: X-Archive-Number: 200501/302 X-Sequence-Number: 9992 On 20 Jan 2005 at 7:26, Stephan Szabo wrote: > On Thu, 20 Jan 2005, Dan Langille wrote: > > > On 20 Jan 2005 at 6:14, Stephan Szabo wrote: > > > > > On Wed, 19 Jan 2005, Dan Langille wrote: > > > > > > > Hi folks, > > > > > > > > Running on 7.4.2, recently vacuum analysed the three tables in > > > > question. > > > > > > > > The query plan in question changes dramatically when a WHERE clause > > > > changes from ports.broken to ports.deprecated. I don't see why. > > > > Well, I do see why: a sequential scan of a 130,000 rows. The query > > > > goes from 13ms to 1100ms because the of this. The full plans are at > > > > http://rafb.net/paste/results/v8ccvQ54.html > > > > > > > > I have tried some tuning by: > > > > > > > > set effective_cache_size to 4000, was 1000 > > > > set random_page_cost to 1, was 4 > > > > > > > > The resulting plan changes, but no speed improvment, are at > > > > http://rafb.net/paste/results/rV8khJ18.html > > > > > > > > Any suggestions please? > > > > > > As a question, what does it do if enable_hashjoin is false? I'm wondering > > > if it'll pick a nested loop for that step for the element/ports join and > > > what it estimates the cost to be. > > > > With enable_hashjoin = false, no speed improvement. Execution plan > > at http://rafb.net/paste/results/qtSFVM72.html > > Honestly I expected it to be slower (which it was), but I figured it's > worth seeing what alternate plans it'll generate (specifically to see how > it cost a nested loop on that join to compare to the fast plan). > Unfortunately, it generated a merge join, so I think it might require both > enable_hashjoin=false and enable_mergejoin=false to get it which is likely > to be even slower in practice but still may be useful to see. Setting both to false gives a dramatic performance boost. See http://rafb.net/paste/results/b70KAi42.html This gives suitable speed, but why does the plan vary so much with such a minor change in the WHERE clause? -- Dan Langille : http://www.langille.org/ BSDCan - The Technical BSD Conference - http://www.bsdcan.org/