Reliable worker-primary connection

[asonnino]

Now that we want to be able to crash-recover and rotate network addresses, we should re-think the connection between primary and its workers. Currently the workers assume the primary is always live and ready to receive their message; same goes for the primary.

[allan-bailey]

Could workers just read from a queue and not be attached to a specific primary?

[asonnino]

that is what they do (if by queue you mean a TCP connection)