DistributedReplicantManagerImpl can't expose lifecycle methods to the microcontainer. Instead the DRM needs to be injected into ClusterPartition before it's create() phase, and ClusterPartition should take DRM through its lifecycle.

Otherwise you can get into a situation where services like HAJNDI depend on ClusterPartition and start before DRM. If this happens and they invoke on an unstarted DRM, the deployment will block (forever). Avoiding that problem requires adding a dependency on DRM, which is way too fragile.

DRM is really a subcomponent of ClusterPartition and should be managed that way.