When the nodes change, outstanding delayed calls must be cancelled.

Since the HTTP polling of the health check is performed in an asynchronous manner, it is possible that the nodes are changed while that check is in process. That must be detected and the outstanding HTTP poll cancelled.