X-Original-To: pgsql-admin-postgresql.org@localhost.postgresql.org Received: from localhost (av.hub.org [200.46.204.144]) by postgresql.org (Postfix) with ESMTP id 30D619DCADB for ; Thu, 5 Jan 2006 02:51:29 -0400 (AST) Received: from postgresql.org ([200.46.204.71]) by localhost (av.hub.org [200.46.204.144]) (amavisd-new, port 10024) with ESMTP id 07702-04 for ; Thu, 5 Jan 2006 02:51:29 -0400 (AST) X-Greylist: from auto-whitelisted by SQLgrey- Received: from wolff.to (wolff.to [66.93.197.194]) by postgresql.org (Postfix) with SMTP id 77CAB9DCAD6 for ; Thu, 5 Jan 2006 02:51:26 -0400 (AST) Received: (qmail 22740 invoked by uid 500); 5 Jan 2006 06:57:47 -0000 Date: Thu, 5 Jan 2006 00:57:47 -0600 From: Bruno Wolff III To: Tsirkin Evgeny Cc: "Jim C. Nasby" , pgsql-admin@postgresql.org Subject: Re: preventing deadlocks Message-ID: <20060105065747.GA17322@wolff.to> Mail-Followup-To: Bruno Wolff III , Tsirkin Evgeny , "Jim C. Nasby" , pgsql-admin@postgresql.org References: <43B10E07.7010503@mail.jct.ac.il> <20051228202607.GA5194@wolff.to> <20060104003001.GY82560@pervasive.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.4.2.1i X-Virus-Scanned: by amavisd-new at hub.org X-Spam-Status: No, score=0.037 required=5 tests=[AWL=0.037] X-Spam-Score: 0.037 X-Spam-Level: X-Archive-Number: 200601/63 X-Sequence-Number: 20151 On Wed, Jan 04, 2006 at 10:51:55 +0200, Tsirkin Evgeny wrote: > > > What you want to do in the trigger isinsert a new row into a table that > > contains the change in count, instead of trying to update a single row > > for each value/ident (btw, you'll probably get better performance if you > > make ident an int instead of a numeric). > > Why? Because this doesn't block other processes and still gives you correct results. > > > So now you'll have a list of > > changes to the value, which you will periodically want to roll up into a > > table that just stores the count. > Interesting idea.Thanks.However it pretty complicates things ,maybe there > will be simpler solution.Something i did not thought about at all - i think that Not that avoids blocking. If you don't have a lot of concurrent queries then this may not be an issue for you. > counting is something that everybody does. No it isn't. Smetimes they aren't needed at all, sometimes approximate values are good enough, and even when they are needed, it is often the case that it is better for count queries to run slower so that other queries run faster. > [1] I have also a hope that i can create a trigger that locks counter > table once a 'select for update' was done on one of the tables i count. > However how can i say if a select that fires a trigger is a 'for update' > one? If you have a counter table and do an UPDATE that will lock that row. If you have multiple tables that you keep counts for you will want to lock the whole counter table using a LOCK command or else you can get deadlocks. Note this means that inserts and deletes from any of the tracked tables will block inserts and deletes of those tables in other concurrent queries. > [2] Maybe there is a MVCC or something solution like Bruno suggested (that > i did not realy understood thought). The explanation given at the top is the gist of the solution that uses MVCC advantagesously.