There are some cases where RequestCoordinator can fail to advance to the next request.

One cause is that the offliner may not return immediately when canceling the request, and the coordinator's state_ is no longer correct.