Multiple service group failover

[shaggy62]

I have some questions about the behavior of service group failover in a N+1 or N+N cluster configuration. I am building two different three node clusters that will run multiple java application servers. My plan is to have each application be bound to a unique IP address. This IP would be an interface alias on a linux bond interface (ie - bond0:0). I've been reading the VCS user guide and it looks like all these things are possible, but I'm looking for confirmation.

The storage for these clusters will be NFS based (mounted on all nodes) with iSCSI used to present the I/O fencing devices.

The questions are:

- Can I configure multiple interface aliases that will move between cluster nodes independent of all other interface aliases?
    - The planned configuration is to have two physical interfaces bonded (Linux bonding) into one interface and then all aliases (bond0:N) would be what applications are bound to.
    - Can one of these interface aliases fail without impacting all other aliases?
    - The goal is independent failover of each interface alias.

- Can I configure multiple applications (and associated resources) to failover independently?
    - Apps A through N run on a given node
    - Goal is failover of any one applications without impact of all other running applications

- Is there any I/O fencing impact?
    - Are the I/O fencing devices always cluster wide?
    - If I configure a cluster with multiple application service groups can one fail independent of the others?

- In an N+1 cluster setup, can I configure things so that specific SGs can run one instance on two nodes and only ever failover to the 3rd node?
    - I'd like to have AppA run an active instance on nodes 1 and 2 and only fail to node 3
    - I'd like to be able to specify that there should never be more than one active instance of AppA on any single node

[Gaurav_S · Accepted solution]

Hello Shaggy,

To answer:

a) Yes, it is possible to configure multiple interface aliases. These interface alias will failover independently (possible) however if both the underneath interfaces are down (i.e a bonded interface failure), then all the aliases will be impacted.

b) Yes, it is possible, best is to configure independent service group for an application. So that in event of a failure, entire service group (containing one application & related resources) will be failed over to other node.

c) Yes, IOfencing is always cluster wide. Regarding second part of this question, ofcourse you can configure cluster with multiple failover application groups & they can fail independently (considering dependencies are configured correctly)

I am little doubtfull on last part, because you can either create a a failover group OR parallel group, you will need to configure a hybrid group so that app runs active on 2 nodes however fails at 3rd node, I have never worked with a hybrid group so I am afraid don't have inputs at this point on this...

Hope this helps

Gaurav

[shaggy62]

Hi Guarav,
Thanks for your answers, these are very helpful. After reading the admin guide, these were mostly the conclusions I drew, but wanted to be sure if I could. I will look into hybrid service groups.