Distributed Execution Framework on a remote cluster

[moia]

Is threre a way to submit a DistributedCallable to an infinispan cluster from the outside?

In our environment we have a cluster of infinispan nodes running a distributed cache. We are currently running ISPN 5.3.

We access this cluster from 'client' nodes using hot rod protocol. The client nodes are application servers running some web applications.

Now we need to invoke some tasks directly on remote cluster nodes.

The DistributedCallable API perfectly seems to fit our needs. Is there a way to invoke them in this architecture?

To clarify the situation - technically our client nodes could participate in the cluster (without hotrod).

But as far as I understand, this means they would take on caching themselves as well.

This is unacceptable, as they are restarted quite often, which would cause some data loss.

They also experience long gc pauses, which could cause whole cluster slowdown. And they have different memory requirements.

So we either need to access Distributed Execution Framework from 'outside' of the cluster, or we need to be able to join a cluster,

but without participating in the caching.

If none of these is possible - is there any recommended way to expose our own interface on the infinispan server (rest or any other)?

Best regards,

Mikolaj

[vblagojevic]

Yeah, you can invoke distributed callable remotely from REST. One problem with these remote initiations of distributed execution is that you can not currently invoke them with some complex objects being passed as a parameter and returned as a result. However, you can use primitive parameters as an identity for those complex input objects that could reside in cache or somewhere else in your application scope.

Regards,

Vladimir

[moia]

And there's no way to build a remote cluster's DistributionManager and RpcManager to send commands to a cluster without 'participating' as a cluster member?

Regards,

Mikolaj

[vblagojevic]

No, not at the moment! We are looking into this and should be in place eventually!

Regards,

Vladimir