pmooney.net

1 - Further to Paul's tip: note that the servers in the target cluster will be contacted in alphabetic order of names. Server01 will be contacted before Server02, for example.

So it's a good idea when naming servers in a cluster to make sure that, if you want replication load to go to a particular server (assuming nornmal availability), that you name that one first in the collating sequence.

2 - Isn't it a possible source of rep conflicts as the cluster replicator will also try to replicate?

3 - @2 - I believe you should look at this as servers outside the cluster or clients replicating with the cluster for this purpose. Not for servers inside the cluster replicating with each other. Even still, replicating normally, and via the cluster replicator will not create replication conflicts.

Also, when you set up a server outside the cluster, or a client, to replicate with the cluster by cluster name, you should also schedule replication with one named server from inside the cluster as well on a much less frequent schedule (once a day for example). The cluster cache (which is where the cluster name, members servers and contact info is stored) needs to be refreshed occasionally by conatcting a member server of the cluster, and if you haven't connected to a member server of the cluster in a long enough time, the client/server will not know how to contact the cluster by cluster name. Normal replication with the cluster by name will refresh the cluster cache as well, but adding the extra replication schedule will create a "self healing" solution should long communiation breaks occur, etc and is a matter of good practice.

4 - In this case field "Files/Directory paths to NOT replicate" doesnt work correctly. If you provide a directory all working fine but if you provide particular filename, this exclision will not work, database will be replicating anyway (seen on R8.0.2)