Running coredns in HA mode

[safderali5]

How to run coredns with 2 replcas on a multi-node k3s cluster? I would like to use the k3s managed coredns installation.

[bogdanr]

There's no replicas field in the packaged manifest at all, so scaling it is enough:

kubectl -n kube-system scale deployment coredns --replicas=2

That survives restarts. The deploy controller checksums the manifest file and compares it to the checksum stored on the Addon resource, and skips the apply completely when nothing changed (pkg/deploy/controller.go), so it never touches your replica count.

You don't need to add anti-affinity either, it's already in there. The deployment ships topologySpreadConstraints on kubernetes.io/hostname with maxSkew: 1 and whenUnsatisfiable: DoNotSchedule, so the second pod is forced onto a different node.

The one thing to watch is upgrades. A k3s upgrade bumps the coredns image in the manifest, the checksum changes and the apply runs, so check kubectl -n kube-system get deploy coredns afterwards.

If you'd rather have it declarative, drop an empty coredns.yaml.skip next to it in /var/lib/rancher/k3s/server/manifests and ship your own copy. k3s then ignores the file entirely without deleting anything, but you lose its templating for the cluster DNS IP and domain, so you'd have to hardcode both.