I re-create the issue here as suggested by @bowei.

Kubernetes version (use kubectl version): v1.5.2

Environment:

• Cloud provider or hardware configuration: baremetal,
• OS (e.g. from /etc/os-release): Debian GNU/Linux 8 (jessie)
• Kernel (e.g. uname -a): 3.16.7
• Install tools: docker-multinode.

Current status: entries in /etc/hosts of the nodes are not used by kube-dns. There is no straightforward way to replicate this custom DNS configuration for the cluster.

What I would like: Some way to easily define custom DNS entries used by kube-dns on a cluster-wide level, without deploying an additional DNS server.

Already considered solutions:

• Use a Service without selector and external Endpoint. It won't allow full DNS names and that won't change for the time being.
• Deploy an additional DNS server and add it to /etc/resolv.conf in all nodes the node running kube-dns. This works, but it's not convenient for quick/testing deployments. Also, it requires access to the host.
• Define a ConfigMap with the DNS entries and add them to the kube-dns dnsmasq /etc/hosts.d. This is possible, but requires modifying kube-dns launch flags (not obvious in automated deployments like docker-multinode.

Possible solutions:

• Implement a special ConfigMap to declare custom entries on a cluster-wide level and make kube-dns look it up.
• kube-dns imports node's /etc/hosts entries, like it does for /etc/resolv.conf. This is not very elegant and doesn't scale, but it replicates a capability currently existing in non containerized system administration.