kops/dns-controller

dns-controller

dns-controller creates DNS records.

Motivation

In the bring-up of a new cluster, protokube has already ensured that we have an etcd cluster and an apiserver. It also sets up DNS records for the etcd nodes (this is a much simpler problem, because we have a 1:1 mapping from an etcd node to a DNS name.)

However, none of the nodes can reach the API server to register. Nor can end-users reach the API. In future we might expose the API server as a normal service via Type=LoadBalancer or via a normal Ingress, but for now we just expose it via DNS.

How it works

Pods with hostNetworking

dns-controller can add DNS records that point to nodes with hostNetworknig enabled.

The dns-controller recognizes annotations on pod.

• dns.alpha.kubernetes.io/external will set up records for accessing the resource using the node's public IP.
• dns.alpha.kubernetes.io/internal will set up records for accessing the resource using the node's private IP.

Services

NodePort

The controller also recognizes these annotations on NodePort services:

• dns.alpha.kubernetes.io/external creates a Route53 A record with public IPs of all the nodes
• dns.alpha.kubernetes.io/internal creates a Route53 A record with private IPs of all the nodes

Loadbalancer

If either of the two annotations are set on a LoadBalancer service, it will create a CNAME for the load balancer hostname or it will create an A record if the load balancer has an IP.

Ingress

dns-controller can optionally watch Ingress resources. To enable this, you need to add the following to the cluster spec:

spec:
  externalDns:
    watchIngress: true

dns-controller will then map the specified ingress hostname and the LoadBalancer assigned to the ingress.