Profiles and clustering

[pmosca]

I'm a little confuse regarding the relationship between profiles and clustering.  I'm looking for some feedback if I should use a profile to define subsystems in general or one for each set of servers cluster, or a combination of both.  Here is my scenario:

Custom Profile A --based on full-ha profile--

  Server group 1

       Server X1, Server X2

       Deployed applications

       App1, App2

  Server group 2

       Server Y1, Server Y2, Server Y3, Server Y4

       Deployed applications

       App3, App4

  Datasources

       DS-App1, DS-App2, DS-App3, DS-App4

Additionally we are not using udp for communication within the cluster but switched to tcp.  Because this change we have to register each server under the profile for the JGroupds, so it looks something like:

     initial_hosts = x.x.x.1[7600],x.x.x.2[7600],x.x.x.3[7600],x.x.x.4[7600]

These are my concerns:

.- Datasources are deployed to all Server groups --even though the related application is not deployed to all servers groups--.

.- All the servers look to be communication between each other and maintaining heartbeat.  I would like that communication would only be between servers on the same server group. 

I'm looking to get some implementation feedback regarding:

Should a scenario like this require having two profiles? one with server group 1 and one with server group 2.  Are there any guidelines that define when to create additional profiles?

Appreciate any feedback that can be provided.

Regards,

[wdfink]

Profiles are a domain thing and only for administration, there is no relation to a cluster.

Having the same configuration is easy to install, but consider that it might be more difficult to understand and to maintain as each change affect both server-groups which might be not wanted in every case.

So it is more a organisation decision than a technical one

You might use the same configuration-profile for both.

The only thing is that you have to clearly separate the clusters.

This can be done by setting a property for initial_hosts. use ${xyz.initial_hosts} inside the JGroups config and add a property with this name to the each server-group and set the correct hosts.

consider, it is recommended that initial_hosts include all hosts of a cluster.

[pmosca]

Fink, thanks for your feedback.

I believe we will split the configuration to use two profiles, one for each custer.  It will also allow to target the datasources to specific servers.

I'm not sure I understood "use ${xyz.initial_hosts} inside the JGroups config and add a property with this name to the each server-group"; I thought JGroups configuration is done at the profile level and would need to have all the servers under that profile, is there a way to define the cluster at the server_group level?