Under certain circumstances, JGroups can issue a MERGE view when a node is joining the cache. The new node joins the cluster, and all nodes have the same cache topology (not containing the joiner yet).

During the merge, the CH's are joined (through CHFactory.union) and as all report the same topology/hash, the resulting hash is identical. However, the joiner is added to the members list and therefore it can finish the initial state transfer, although no data have been assigned to him.

Later, the coordinator starts rebalance and the node begins to receive some data, but the thread which started the cluster manager (and should wait until the cluster becomes properly replicated through initial ST) is already released.