There is no foolproof way of detecting what has gone wrong, so most systems simply use a timeout: nodes frequently bounce messages back and forth between each other, and if a node doesn’t respond for some period of time — say, 30 seconds — it is assumed to be dead.
