Not sure why your comment was flagged and marked dead @helpmetest. Workers connect over a persistent streaming connection. Heartbeats are sent every 3s (configu

Not sure why your comment was flagged and marked dead @helpmetest. Workers connect over a persistent streaming connection. Heartbeats are sent every 3s (configurable). If the server is not able to deliver the heartbeat (or any other message), the job is transitioned back from in_flight to ready for other workers to pick up. Workers also kill the connection if they have not received a message within a configurable timeframe. Jobs are not leased for a fixed window. They are considered in_flight for as long as the connection they were sent on is still active. The delivery guarantee is "at least once". Workers should be written in a way that makes them resilient to seeing the same job twice, as there is no way to be certain a worker did or did not process a job in the case of a disconnect if that job was not ack'd or nack'd.