kubernetes: kube2sky can not resolve kubernetes service DNS name

There seem to be a chicken and egg problem with the current kube-dns replication controller (v3).

As I mentioned there removing the -kube_master_url flag forces kube2sky to use the kubeconfig file to find the address of the master, but this file contains the DNS name of the service, which can not be resolved:

clusters:
- name: local
  cluster:
     server: https://kubernetes
     insecure-skip-tls-verify: true

As a result kube2sky fails with the error:

I0623 22:22:32.315145       1 kube2sky.go:451] Using https://kubernetes for kubernetes master
I0623 22:22:32.315168       1 kube2sky.go:452] Using kubernetes API 
E0623 22:22:32.322136       1 reflector.go:136] Failed to list *api.Endpoints: Get https://kubernetes/api/v1beta3/endpoints: dial tcp: lookup kubernetes: no such host

This happens in a k8s 0.19.3 cluster powered by CoreOS, on AWS. Surprisingly, running the exact same replication controller with the same secret in a cluster started via kube-up.sh, also on AWS, works perfectly fine:

I0623 22:38:36.231746       1 kube2sky.go:451] Using https://10.0.0.1:443 for kubernetes master
I0623 22:38:36.231760       1 kube2sky.go:452] Using kubernetes API v1beta3

This behaviour feels pretty random and I can’t figure out the reason why in the second scenario https://kubernetes is converted to https://10.0.0.1:443, where is the magic?