Otherwise the streams don't advance and steadily fall behind, so when a worker does connect either a) they'll be streamed lots of old updates or b) the connection will fail as the streams are too far behind.