Recovering from a Connection Server Crash

A connection server holding 500,000 live WebSocket connections crashes without warning (hardware failure).

1. Trace what happens to the 500,000 affected clients, step by step, from crash to full recovery.
2. Why is it safe to say no messages are lost, even though the crash is sudden and unannounced?
3. What client-side behavior is essential to prevent the recovery itself from causing an outage?