In incident T367348: Incident: 2024-06-12 toolforge k8s control plane we discovered a failure case where the Kubernetes API server's port is open but it's not responding to any API requests. In that case HAProxy still sees the failed servers as up and tries to send traffic to them. The current HAProxy health check is a simple TCP check, I think it should be replaced by a HTTPS check for the /healthz endpoint.
Also review that we have alerts for the case where the API servers are not healthy according to HAProxy.
