How do I use Hazelcast on JBoss 7 (and newer) right way?

[Viktor Alexandrov]

Hello all.

Did anyone have experience with running hazelcast inside managed classloading environment like JBoss modules? As I found out there are several options to choose from, and I can't decide to pick one of them as "right". Options are:

1. Include hazelcast jar as library of deployment (web-app or ear or something specific to app-server)

  Pros:

    - no classloading problems

    - any application specific java object can be seamlessly stored in hazelcast map, queue or topic

  Cons:

    - initializing a "Cluster Node" per-application, not per-server (jvm), so there might be several nodes in one jvm which may cause data loss on jvm/server crash

    - can not share instance between applications (deployments)

2. Make a "module" of hazelcast and include it as a dependency in each deployment

  Pros:

    - single instance of HZ per JVM

    - shared resource between deployments

    - local invocations do not perform any serialization

    - easy to maintain all apps to use the same version of HZ

  Cons:

    - classloading hell when trying to put to topic/map an application-specific bean (the queue works ok, it's a kind of magic!) - tested with HZ3.1. Can't get it working, so decided to use HashMaps insteand of beans.

    - hence the latter point the only option to use HZ as a data-transfer-layer is to use standard classes or primitives

3. Fix option 2's issues with adding a jar with app-specific beans to hazelcast module to make them available for HZ classloader

  Pros:

    - fixed classloding issues

  Cons:

    - need to restart JBoss on module change

    - need to maintain right versions of DTO beans in app and in module

    - when several apps use different DTOs the HZ module become unsupportable

My heart choose option 2 :) But I have to win the fight with classloading to make it to production. Any thoughts?

[marc....@gmail.com]

Hi There,

How would one go about performing option number 2?

Thanks!