Matt Miller wrote:
>>>http://pgfoundry.org/projects/pgcluster/ which provides syncronous
>>>multi-master clustering.
>>
>>He specifically said that pgcluster did not work for him
>>because ...PGCluster requires that there be a load balancer and a
>>replicator centrally located managing the cluster. If a network
>>problem happens at the centralized location, it would bring down all
>>locations completely.
>
>
> I think the load balancer is an optional component. Clients can connect
> either to a load balancer, or directly to a specific machine.
The replicator surely is not optional, and must be centralized.
> Also, I think pgcluster can operate async. If the system runs in
> "normal" mode then the client gets a response as soon as the transaction
> is committed on the local machine. In "reliable" mode the client waits
> for the commit to happen on all machines.
>
Interesting. I suppose whatever works for your application is the right
answer for replication. PGCluster is query-based, right? I suppose the
question would then be, would he rather use a trigger-based replication
solution, like Slony, or a query-based replication solution.
Regards,
Jeff Davis