Bsically, if we have a few seed nodes, a partition happens so that a singleton is running on a "detached cluster", and nodes "come back" by forming a new cluster - they may decide to run the singleton - because they don't know about the node that's running the singleton, nor is any of the seed nodes in contact with it.

This task is about introducing patterns to remove this problem. One pattern is using roles to delay starting the singleton until more seed nodes know the nodes that have joined. Another one is manualy joining nodes and using an external discovery service (zookeeper for example) to find nodes.

Note: We currently do not do autodiscovery in akka, which would also help to kill this problem.

Discussed in detail here: https://groups.google.com/forum/#!topic/akka-user/ns7DPHGYbIk