Currently, the callback for the hello command sent by the SingleServerDiscoveryMonitor takes the mutex before checking for shut down. We could let a thread lock in the case where the ReplicaSetMonitor task executor thread acquires SingleServerDiscoveryMonitor's mutex when calling SingleServerDiscoveryMonitor::shutdown and then tries to acquire it again when calling the callback function after cancelling pending a hello command. To avoid this, the callback should not take the mutex if the status is CallbackCanceled.