Configure Master and Worker Nodes to resolve DNS using kube-dns service first then host

[sputnick]

I was testing Pod based NFS for Persistent Volumes and ran into an issue on Kube-Solo (and likely Kube-Cluster has the same). In version 0.9.6 you can successfully create a PersistentVolume that points to an NFS server via its cluster dns service name e.g. "nfs-server.default.svc.cluster.local". You can then also create a PersistentVolumeClaim. However when you create a new Pod/RC that uses that Claim, Kubernetes fails to mount using the NFS provider because the worker/master Nodes cannot resolve the Services' ClusterIP address from the DNS name. This is because /etc/resolve.conf does not contain the kube-dns Service IP address but only the local host IP. I manually changed resolve.conf and kubernetes immediately was able to start everything correctly.

So I guess this is an enhancement/bug. And it seems to be a common issue for k8s users so worth fixing, see: kubernetes/kubernetes#8735

I have attached my test case yamls.
fake-nfs.zip

[rimusz]

@AntonioMeireles is it possible to add kube-dns Service IP address to /etc/resolve.conf?
What I mean is it possible to add it via cloud config?

[AntonioMeireles]

@sputnick thanks for reporting!. going to look.