On 16.08.2007 08:42, Mikko Partio wrote:
> I have a mission to implement a two-node active-passive PostgreSQL cluster.
> The databases at the cluster are rather large (hundreds of GB's) which opts
> me to consider a shared disk environment. I know this is not natively
> supported with PostgreSQL, but I have been investigating the Red Hat Cluster
> Suite with GFS. The idea would be that the cluster programs with gfs (and HP
> ilo) would make sure that only one postmaster at a time would be able to
> access the shared disk, and in case the active node fails the cluster
> software would shift the services to the previously passive node. What I'm
> pondering here is that is the cluster able to keep the postmasters
> synchronized at all times so that the database won't get corrupted.
>
> Is there anyone on the list that has seen such configuration, or, even
> better, implemented it themselves? I found a small document by Devrim Gunduz
> describing this scenario but it was rather scant on details.
>
> If shared disk is definitely out of the question, the fallback plan would be
> to use drbd and linux-ha.
The usual setup is DRBD + Heartbeat, which is fast, simple and proven.
Using a shared disk / SAN has disadvantages, such as single point of
failure, (usually) non-native fencing and (usually) way higher latency.
DRBD does handle a lot of stuff by it self, which you need to take care
yourself with a plain shared device. Using a cluster file system such as
GFS2, OCFS2 or Lustre is a waste of resources, as you can't have
active/active with PostgreSQL anyway.
--
Regards,
Hannes Dorbath
