If requests hang for too long, kernel will cancel them and shutdown the associated socket. If this happens, we want to reconfigure nbd device to use a new socket, because currently it can only be fixed with a service restart. Otherwise processes using this device will be stuck in d-state for a dead_conn_timeout, which we currently set to 24 hours