Received: from localhost (unknown [200.46.204.183]) by mail.postgresql.org (Postfix) with ESMTP id 0D256632CB0 for ; Sun, 15 Feb 2009 14:50:45 -0400 (AST) Received: from mail.postgresql.org ([200.46.204.86]) by localhost (mx1.hub.org [200.46.204.183]) (amavisd-maia, port 10024) with ESMTP id 48851-03 for ; Sun, 15 Feb 2009 14:50:26 -0400 (AST) X-Greylist: from auto-whitelisted by SQLgrey-1.7.6 Received: from gwmta.wicourts.gov (gwmta.wicourts.gov [165.219.244.99]) by mail.postgresql.org (Postfix) with ESMTP id 673ED632C51 for ; Sun, 15 Feb 2009 14:50:08 -0400 (AST) Received: from Courts-MTA by gwmta.wicourts.gov with Novell_GroupWise; Sun, 15 Feb 2009 12:50:06 -0600 Message-Id: <49980F79.EE98.0025.0@wicourts.gov> X-Mailer: Novell GroupWise Internet Agent 7.0.3 Date: Sun, 15 Feb 2009 12:50:01 -0600 From: "Kevin Grittner" To: "Scott Marlowe" , "Tom Lane" Cc: "Michael Monnerie" , Subject: Re: Vacuum wait time problem References: <4995ABC0.4090401@musicreports.com> <200902140102.09224@zmi.at> <23106.1234576963@sss.pgh.pa.us> In-Reply-To: Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Disposition: inline X-Virus-Scanned: Maia Mailguard 1.0.1 X-Archive-Number: 200902/100 X-Sequence-Number: 31321 >>> Scott Marlowe wrote: > On Fri, Feb 13, 2009 at 7:02 PM, Tom Lane wrote: >> Scott Marlowe writes: >>> On Fri, Feb 13, 2009 at 5:02 PM, Michael Monnerie >>> wrote: >>>> vacuum_cost_delay = 0 >>>> That was the trick for me. It was set to 250(ms), where it took 5 >>>> hours for a vacuum to run. Now it takes 5-15 minutes. >> >>> Wow!!! 250 ms is HUGE in the scheme of vacuum cost delay. even >>> 10ms is usually plenty to slow down vacuum enough to keep it out >>> of your way and double to quadruple your vacuum times. >> >> I wonder whether we ought to tighten the allowed range of >> vacuum_cost_delay. The upper limit is 1000ms at the moment; >> but that's clearly much higher than is useful, and it seems >> to encourage people to pick silly values ... > > I agree. I can't imagine using a number over 50 or so. I don't know what other people have found useful, but when I experimented with this in our environment, it seemed like I should just treat vacuum_cost_delay as a boolean, where 0 meant off and 10 meant on, and tune it by adjusting vacuum_cost_limit. The granularity of vacuum_cost_delay is course and surprising unpredictable. -Kevin