On Fri, Mar 05, 2004 at 09:15:59AM +0800, Jagi C. Sarcilla wrote: > anyone implemented postgresql with fail-over (clustered in think) I > need inputs about it, your help is highly appreciated.
There are at least two approaches available that I'm aware of. The first approach is a general high-availability solution using Heartbeat[1], DRBD[2] and of course, PostgreSQL and the GNU/Linux operating system, preferably with a journalling filesystem like ext3[3] or ReiserFS[4] (I prefer XFS[5], but it won't be supported by DRBD until DRBD v0.7 because XFS uses dynamic block sizes). Heartbeat handles the high-availability component, mounting and unmounting filesystems, bringing up and taking down IP aliases and starting up and shutting down programs as needed. DRBD provides something like a network-based RAID-1, with one node having an active copy and the other node having a realtime mirror that goes live when the primary node goes down. I've tested the Heartbeat+DRBD approach for a Samba fileserver + Cyrus mail server + DHCP3 server, and it worked out very well. Both nodes were connected using GbE via a crossover cable for heartbeat communication and DRBD replication and were connected to the rest of the world via standard 100Mbps Ethernet. Failover and data synchronization was flawless (well... except for WINS propagation for the Samba server which had much to be desired from). PostgreSQL is known to work well with this type of a high-availability setup, and it is recommended if you want other services to be highly-available as well. The main drawback of the Heartbeat approach is that the inactive node is idle most of the time (unless you set up an active-active type of cluster). An alternative approach puts a middle-person in between the rest of the world and your PostgreSQL databases. One such middleware is DBBalancer[6] currently at v0.4.0ALPHA. The DBBalancer approach adds a single point of failure, but provides high-availability in the sense that you have two or more PostgreSQL databases with the same data. One strong point of the DBBalancer approach over the Heartbeat approach is that you have load balancing for a theoritical improvement in read performance. At present the DBBalancer has other drawbacks aside from the obvious SPoF. A major killer for some might be that it only supports trust and cleartext password authentication methods (for now, anyway). --> Jijo [1] http://www.linux-ha.org/heartbeat/ [2] http://www.drbd.org/ [3] http://www.redhat.com/support/wpapers/redhat/ext3/index.html [4] http://www.namesys.com/ [5] http://oss.sgi.com/projects/xfs/ [6] http://dbbalancer.sourceforge.net/ -- Federico Sevilla III : Free & Open Source Software : When we speak of free http://jijo.free.net.ph : Consultant & Administrator : software we refer to GnuPG Key ID : 0x93B746BE : freedom, not price. -- Philippine Linux Users' Group (PLUG) Mailing List [EMAIL PROTECTED] (#PLUG @ irc.free.net.ph) Official Website: http://plug.linux.org.ph Searchable Archives: http://marc.free.net.ph . To leave, go to http://lists.q-linux.com/mailman/listinfo/plug . Are you a Linux newbie? To join the newbie list, go to http://lists.q-linux.com/mailman/listinfo/ph-linux-newbie
