Currently, CNS will repeatedly try to start controller-manager, even though all errors from it effectively leave controller-manager in a terminal state. This operation should have a timeout that will cause CNS to terminate.

In CNS versions after v1.6.23, this defect is ameliorated by the health check improvements added in #3269. In some cases, this will cause CNS to be considered unhealthy by the API server since it will be unable to perform list requests for NNCs. Since controller-manager performs the same requests to the API server, this specific case could be mitigated by the health check. However, because controller-manager is left in a terminal state, it does not account for scenarios with transient apiserver unavailability.

We should avoid retries of controller-manager.Start entirely, set a reasonable timeout for the context passed in to controller-manager.Start, and ensure that the application is terminated when controller-manager.Start produces an error.