I'm currently seeing very similar behavior for the latest version of BDR where cluster nodes are very confused about the state of their connections/replication slots when a new node joins with a completely different local_node_name but same IP address as previous cluster member.
