X-Original-To: pgsql-performance-postgresql.org@localhost.postgresql.org Received: from localhost (unknown [200.46.204.2]) by svr1.postgresql.org (Postfix) with ESMTP id 20D14D1B51E for ; Tue, 21 Oct 2003 16:24:21 +0000 (GMT) Received: from svr1.postgresql.org ([200.46.204.71]) by localhost (neptune.hub.org [200.46.204.2]) (amavisd-new, port 10024) with ESMTP id 95677-09 for ; Tue, 21 Oct 2003 13:23:50 -0300 (ADT) Received: from ihemail2.firewall.lucent.com (ihemail2.lucent.com [192.11.222.163]) by svr1.postgresql.org (Postfix) with ESMTP id 12443D1B55F for ; Tue, 21 Oct 2003 13:23:45 -0300 (ADT) Received: from ihmail.ih.lucent.com (h135-1-218-70.lucent.com [135.1.218.70]) by ihemail2.firewall.lucent.com (Switch-2.2.8/Switch-2.2.0) with ESMTP id h9LGNeN08603; Tue, 21 Oct 2003 11:23:41 -0500 (CDT) Received: from [192.168.10.52] (USS-Excelsior.ih.lucent.com [135.185.171.70]) by ihmail.ih.lucent.com (8.11.7+Sun/EMS-1.5 sol2) id h9LGNdF24594; Tue, 21 Oct 2003 11:23:39 -0500 (CDT) Mime-Version: 1.0 X-Sender: slgan@135.1.218.70 Message-Id: In-Reply-To: <15841.1066685113@sss.pgh.pa.us> References: <19692.1066578442@sss.pgh.pa.us> <200310191204.23888.josh@agliodbs.com> <15841.1066685113@sss.pgh.pa.us> Date: Tue, 21 Oct 2003 11:25:33 -0500 To: Tom Lane From: Seum-Lim Gan Subject: Re: index file bloating still in 7.4 ? Cc: Josh Berkus , pgsql-performance@postgresql.org Content-Type: text/plain; charset="us-ascii" ; format="flowed" X-Virus-Scanned: by amavisd-new at postgresql.org X-Archive-Number: 200310/535 X-Sequence-Number: 4283 Hi Tom, 1.) OK. We have narrowed it down. We did a few (like 5 to 8 times) vacuum analyze (no full), the pg_statistics relfilenode grew. There was no database operation when we did this, no other client connections except the one that does the vacuum. If we do plain simple "vacuum " (again no full), we see pg_statistics_relid_att_index relfilenode grew instead of pg_statistics. So, overtime, these files will grow if we do vacuum. Are these expected ? The question now is, if we are not doing anything to the database, why would they grow after a few vacuums ? 2.) The other problem we have with > DETAIL: 101802 dead row versions cannot be removed yet. > DETAIL: 110900 dead row versions cannot be removed yet. > DETAIL: 753064 dead row versions cannot be removed yet. > DETAIL: 765328 dead row versions cannot be removed yet. We will collect more data and see what we can get from the the process. Offhand, the process is connecting to the database through ODBC and we don't use any BEGIN in our updates, just doing plain UPDATE repeatedly with different keys randomly. The database is defaulted to autocommit=true in postgresql.conf. Thanks. Gan At 5:25 pm -0400 2003/10/20, Tom Lane wrote: >Seum-Lim Gan writes: >> We tried one more thing: with the table not being updated >> at all and we did vacuum. Each time a vacuum is done, >> the index file becomes bigger. > >It is not possible for plain vacuum to make the index bigger. > >VACUUM FULL possibly could make the index bigger, since it has to >transiently create duplicate index entries for every row it moves. > >If you want any really useful comments on your situation, you're going >to have to offer considerably more detail than you have done so far --- >preferably, a test case that lets someone else reproduce your results. >So far, all we can do is guess on the basis of very incomplete >information. When you aren't even bothering to mention whether a vacuum >is FULL or not, I have to wonder whether I have any realistic picture of >what's going on. > > regards, tom lane > >---------------------------(end of broadcast)--------------------------- >TIP 4: Don't 'kill -9' the postmaster -- +--------------------------------------------------------+ | Seum-Lim GAN email : slgan@lucent.com | | Lucent Technologies | | 2000 N. Naperville Road, 6B-403F tel : (630)-713-6665 | | Naperville, IL 60566, USA. fax : (630)-713-7272 | | web : http://inuweb.ih.lucent.com/~slgan | +--------------------------------------------------------+