Manage CoreDNS for DNS in Amazon EKS clusters

To improve the stability and availability of the CoreDNS Deployment, versions v1.9.3-eksbuild.6 and later and v1.10.1-eksbuild.3 are deployed with a PodDisruptionBudget. If you’ve deployed an existing PodDisruptionBudget, your upgrade to these versions might fail. If the upgrade fails, completing one of the following tasks should resolve the issue:

In EKS add-on versions v1.9.3-eksbuild.3 and later and v1.10.1-eksbuild.6 and later, the CoreDNS Deployment sets the readinessProbe to use the /ready endpoint. This endpoint is enabled in the Corefile configuration file for CoreDNS.

If you use a custom Corefile, you must add the ready plugin to the config, so that the /ready endpoint is active in CoreDNS for the probe to use.

In EKS add-on versions v1.9.3-eksbuild.7 and later and v1.10.1-eksbuild.4 and later, you can change the PodDisruptionBudget. You can edit the add-on and change these settings in the Optional configuration settings using the fields in the following example. This example shows the default PodDisruptionBudget.

{
    "podDisruptionBudget": {
        "enabled": true,
        "maxUnavailable": 1
        }
}

You can set maxUnavailable or minAvailable, but you can’t set both in a single PodDisruptionBudget. For more information about PodDisruptionBudgets, see Specifying a PodDisruptionBudget in the Kubernetes documentation.

Note that if you set enabled to false, the PodDisruptionBudget isn’t removed. After you set this field to false, you must delete the PodDisruptionBudget object. Similarly, if you edit the add-on to use an older version of the add-on (downgrade the add-on) after upgrading to a version with a PodDisruptionBudget, the PodDisruptionBudget isn’t removed. To delete the PodDisruptionBudget, you can run the following command:

kubectl delete poddisruptionbudget coredns -n kube-system

In EKS add-on versions v1.10.1-eksbuild.5 and later, change the default toleration from node-role.kubernetes.io/master:NoSchedule to node-role.kubernetes.io/control-plane:NoSchedule to comply with KEP 2067. For more information about KEP 2067, see KEP-2067: Rename the kubeadm "master" label and taint in the Kubernetes Enhancement Proposals (KEPs) on GitHub.

In EKS add-on versions v1.8.7-eksbuild.8 and later and v1.9.3-eksbuild.9 and later, both tolerations are set to be compatible with every Kubernetes version.

In EKS add-on versions v1.9.3-eksbuild.11 and v1.10.1-eksbuild.7 and later, the CoreDNS Deployment sets a default value for topologySpreadConstraints. The default value ensures that the CoreDNS Pods are spread across the Availability Zones if there are nodes in multiple Availability Zones available. You can set a custom value that will be used instead of the default value. The default value follows:

topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: topology.kubernetes.io/zone
    whenUnsatisfiable: ScheduleAnyway
    labelSelector:
      matchLabels:
        k8s-app: kube-dns

In EKS add-on versions v1.11.1-eksbuild.4 and later, the container image is based on a minimal base image maintained by Amazon EKS Distro, which contains minimal packages and doesn’t have shells. For more information, see Amazon EKS Distro. The usage and troubleshooting of the CoreDNS image remains the same.