[Pgcluster-general] Question about setting up PGCluster

[Sean Brown]

On Nov 1, 2007, at 10:24 AM, Holger Lehmann wrote:

> Hi Sean,
>
> Am Donnerstag, 1. November 2007 15:13:24 schrieb Sean Brown:
>> I am working on setting up a PostgreSQL cluster using PGCluster. I
>> have one question about the replication server.
>>
>> Is the replication server supposed to run on another system or is it
>> a daemon that runs on each Node? If it is supposed to run on a
>
> only one pgreplicate instance is needed, although you can run more  
> than one if
> you want to be failsafe. Where you start them is completely up to  
> you. You
> can start them on the same machine where your postgresql runs or on  
> another
> machine.
>
>> separate system can anyone provide pointers on sizing the system?
>
> Gee, for pgreplicate, make some space for the logfile, you need  
> some good I/O
> troughput, make shure memory is not too tight. It all depends on  
> how many
> data must be replicated in what time ... I guess you might want to  
> play
> around with pgcbench (several times in a loop) and see what happens.

Ok, so if the idea is that we don't loose any data in the event that  
one node goes down, and when it comes back it can be restored to  
where it should be and I don't want (currently) a separate machine to  
run pgreplicate I would choose to run pgreplicate on each node in the  
cluster.

Do I understand and have that straight?

>
>>
>> I'm planning on setting up a Master/Slave Cluster so I won't be using
>> the load balancer.
>
> Um, sorry, Master-Slave Replications are this way:
> http://pgfoundry.org/projects/slony1/ (Slony-I), just kidding.  
> PGReplicate is
> a full Multi-Master Replicator and cannot be configured to be a  
> Master-Slave
> Replicator. You can force you applications to behave so, but you do  
> not need
> to :-)

There were other reasons why Slony wasn't going to work out for us so  
yes, while we could set up multimaster, we will be choosing not to  
right now.