One of the manifestation of the issue:
- start 2 nodes with SLSB with TUNNEL transport
- start both nodes creating 2 clusters (or partitions)
- start ejb client
- start GossipRouter and wait for merge
- ejb client keeps talking only to known node; never receives a topology update
This is because org.wildfly.clustering.server.registry.CacheRegistry#topologyChanged does not handle cluster merges and thus all entries from a given partition will be lost forever.
The effects are especially missing client mappings and broken session stickiness.