kubernetes: PodDisruptionBudget maxUnavailable

PodDisruptionBudget allows you to specify minAvailable.

I would like to have, say, a cassandra cluster, and be able to scale it up and down according to some custom metric.

Say my data is replicated 3 ways. So, I don’t ever want more than 2 instances down or I can’t read my data during the disruption.

With the current minAvailable, I have to adjust both the Deployment (or PetSet) scale, and the PodDisruptionBudget at the same time, which prohibits use of generic scaling, e.g. via HPA with custom metrics.

Presumably the validation would only allow one of these options to be specified at a time.

@kubernetes/api-review-team

@kow3ns @foxish we talked about this today.

@davidopp @mwielgus you work on PodDisruptionBudget I think.