Received: from localhost (maia-5.hub.org [200.46.204.182]) by postgresql.org (Postfix) with ESMTP id 7A4E39FBA80 for ; Thu, 31 May 2007 11:43:23 -0300 (ADT) Received: from postgresql.org ([200.46.204.71]) by localhost (mx1.hub.org [200.46.204.182]) (amavisd-maia, port 10024) with ESMTP id 90853-03 for ; Thu, 31 May 2007 11:43:12 -0300 (ADT) X-Greylist: from auto-whitelisted by SQLgrey-1.7.5 Received: from main2.mycybernet.net (main2.mycybernet.net [209.222.63.140]) by postgresql.org (Postfix) with ESMTP id 1D5A49FBA7A for ; Thu, 31 May 2007 11:43:18 -0300 (ADT) Received: from 227-54-222-209.mycybernet.net ([209.222.54.227] helo=phlogiston.dydns.org) by main2.mycybernet.net with esmtp (Exim 4.62) (envelope-from ) id 1Htlrx-0001ow-Eh for pgsql-admin@postgresql.org; Thu, 31 May 2007 10:43:17 -0400 Received: by phlogiston.dydns.org (Postfix, from userid 1000) id 118D73FAF; Thu, 31 May 2007 10:43:12 -0400 (EDT) Date: Thu, 31 May 2007 10:43:12 -0400 From: Andrew Sullivan To: pgsql-admin@postgresql.org Subject: Re: High Avaliable in the PostgreSQL. Message-ID: <20070531144312.GE19737@phlogiston.dyndns.org> References: <20070530164208.2657ee91@localhost> <1c66bda80705302230p7ed40acam8f39e3fea2a00ac6@mail.gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1c66bda80705302230p7ed40acam8f39e3fea2a00ac6@mail.gmail.com> User-Agent: Mutt/1.5.9i X-Virus-Scanned: Maia Mailguard 1.0.1 X-Archive-Number: 200705/332 X-Sequence-Number: 25652 On Thu, May 31, 2007 at 05:30:20PM +1200, adey wrote: > HP was providing CA (Continuous Access) software that was claimed to provide > WAN SAN replication by repeating IO in exactly the sequence it was generated > on the master, to the slave. The CA stuff, or anything else built on FCIP, is pretty cool, but I'd worry a little about latency. Certainly your costs are going to be high on transit, but if the data and uptime are worth it, you could do it. The big issue here is the same as in any other failover-to-other-node case -- you have to be _super_ sure that the failing master writer is dead and disconnected, and across a wide area, this is going to be very tough to do. I think you'd have to spend a fair amount of time doing risk analysis of conflicting commits happening in both sites (if the WAN link goes down because, say, some genius decides to run a backhoe through the fiber going out of your "primary" city, you could find your secondary site decides to promote itself. But the primary site might still have transactions in flight.) > system. I'm not sure this would be sufficient for 99.9% uptime though, as > there would be some startup requirements on the slave. Right, and the requirement was actually upped to 99.99%, which is approximately 1 hour of allowable downtime a year. That is a very high bar. A -- Andrew Sullivan | ajs@crankycanuck.ca Users never remark, "Wow, this software may be buggy and hard to use, but at least there is a lot of code underneath." --Damien Katz