Ideally update service implementations defensively shed load if they feel overwhelmed by traffic, and the cluster-version operator should not get too upset if its local update recommendations are an hour or so stale vs. a minute or so stale. One way to address this would be to teach the CVO to respect Retry-After or other 429 or 503 responses and back off that amount or more before making a follow-up attempt, instead of banging away at the usual frequency.

Definition of done:

• A CVO pointed at a 429ing or 503ing upstream update service will reduce its poll frequency to avoid overloading a troubled service.