Right now, there's nothing accounting for the fact that sockets aren't meant to be shared across nodes (the client is meant to be the holder of all state). If a connection were dropped because the server were restarted and Primus reconnected, it wouldn't send a new "join" message, and that's a problem. (Even if it were to send a new "join" message, it'd get all the messages again, and that's a waste of bandwidth: it needs to say what the latest message it got was.)
Right now, there's nothing accounting for the fact that sockets aren't meant to be shared across nodes (the client is meant to be the holder of all state). If a connection were dropped because the server were restarted and Primus reconnected, it wouldn't send a new "join" message, and that's a problem. (Even if it were to send a new "join" message, it'd get all the messages again, and that's a waste of bandwidth: it needs to say what the latest message it got was.)