Cluster DNS

Basics

For a service called web-service created in the apps namespace the following DNS record would be created. Also showing the DNS record for the pod if that is enabled.

Communication inside the same namespace can use the “Hostname” which is the name of the service. Communication from namespace to namespace must use some or all of the FQDN which would be web-service.apps.svc.cluster.local but could be just web-service.apps

Records for pods are not enabled by default but can be enabled. If enabled, the Hostname for the pod is the ip with the dots replaced with dashes.

Implementation

Kubernetes now uses CoreDNS which is deployed as a replicaset in the kube-system namespace. It is configured via the /etc/coredns/Corfile file which is passed in as a configmap called coredns

CoreDNS works with plugins. The plugin that makes CoreDNS work with Kubernetes is the kubernetes plugin. Others include errors, logging, etc.
The CoreDNS kubernetes plugin config is where the clusters cluster.local top level domain is set and options such as creating DNS records for pods are set.

When the CoreDNS container is deployed, it creates a service called kube-dns by default, and this services IP is used by the kubelet to assign the nameserver when each pod is created, it does this by adding an entry to the /etc/resolv.conf file like this nameserver 10.43.0.10
A search entry is also added with DNS search information like this: default.svc.cluster.local svc.cluster.local cluster.local where the first entry contains the name of the namespace, default in this case.

The kubelet config file at /var/lib/kubelet/config.yaml will contain the clusterDNS and clusterDomain as configured by CoreDNS.