Description
Currently, both kube-controller-manager and kube-scheduler perform service checks using the /healthz endpoint without a way of disabling it. The health_url config property defaults to the metrics endpoint with /metrics replaced by /healthz.
We are looking to use both checks with EKS, given that they now support fetching these metrics via a custom APIService as documented here, however they do not expose the container's /healthz endpoint meaning we experience errors on the service check and have it report a failure.
Workaround
At the moment we are (a)busing the fact the regex defaulting logic checks for an exact match of ending with /metrics and using a prometheus_url: https://kubernetes.default.svc/apis/metrics.eks.amazonaws.com/v1/kcm/container/metrics? to invalidate the defaulting logic. This appears to be working with EKS for now.