When 2 nodes are down, akka has quirky, by-design behavior such that it won't let either node re-join until both become reachable or are declared "down".

So with node1 and node2 down, when node1 is restarted. the shards are actually still operational b/c the lower level akka remoting on node3 does have a connection so node3 is able to send AppendEntries heartbeats to node1 so the shards have an operational leader. We mainly just use akka cluster to get the remote node addresses. So the only problem in this scenario is that node1, while it knows node3 is its leader, it doesn't yet have node3's akka address so it can't initiate messages to node3. Thus transactions initiated on node1 would fail. We can alleviate/work around that issue by sending the leader's full address in the AppendEntries message.