World K9: Cluster Admin

Fix CoreDNS and kube-proxy

You will learn: to tell a phone book problem (CoreDNS) from a runner problem (kube-proxy), and fix the first.

Checked with Kubernetes 1.37 on 1 Nov 2026

The problem

From the client tray, calling the shop by its Service IP works. Calling it by the name shop fails. What is broken?

Watch

You are here: cluster kind-dev namespace default

~/shop $ kubectl exec client -- nslookup shopServer:                10.96.0.10Address:        10.96.0.10:53Name:        shop.default.svc.cluster.localAddress: 10.96.135.90~/shop $ kubectl get svc -n kube-systemNAME       TYPE        CLUSTER-IP   EXTERNAL-IP   PORT(S)                  AGEkube-dns   ClusterIP   10.96.0.10   <none>        53/UDP,53/TCP,9153/TCP   5m21s

Every tray asks the phone book at 10.96.0.10. That number is the kube-dns Service. Behind it run the trays.

Try it

Which part do you suspect? Sort each clue.

Drag a card into a box. Or tap the card, then tap the box.

Try it

Someone scaled CoreDNS to 0 “to save money”. Bring the phone book back: 2 ready trays. Note: this practice cluster still answers names meanwhile. A real one would not.

You are here: cluster kind-dev namespace default

shop3/3 ready

dev-control-planeOpenManager

No trays here.

dev-workerOpen

shop·5pldw
shop·r9q8j

dev-worker2Open

client
shop·j89pb

Time: +0s (sped up)

Hint

CoreDNS is a Deployment in kube-system. Scale it back to 2.

Watch

You are here: cluster kind-dev namespace default

~/shop $ kubectl get pods -n kube-system -l k8s-app=kube-proxy -o wideNAME               READY   STATUS    RESTARTS   AGE     IP           NODE                NOMINATED NODE   READINESS GATESkube-proxy-5wnvx   1/1     Running   0          5m20s   172.18.0.2   dev-control-plane   <none>           <none>kube-proxy-xwjpx   1/1     Running   0          5m20s   172.18.0.3   dev-worker          <none>           <none>kube-proxy-z25tj   1/1     Running   0          5m20s   172.18.0.4   dev-worker2         <none>           <none>

is the runner on every stall. It turns a Service IP into a real tray’s address. A DaemonSet keeps one per stall.

Predict

You delete the kube-proxy tray on dev-worker2. What happens next?

Check

Names fail. Put the DNS ladder in order, like the official guide.

First rung at the top. Put these in order. Use the arrows, or drag.

kubectl logs -n kube-system -l k8s-app=kube-dns
kubectl exec client -- cat /etc/resolv.conf
kubectl get endpointslice -l kubernetes.io/service-name=kube-dns -n kube-system
kubectl get pods -n kube-system -l k8s-app=kube-dns
kubectl exec client -- nslookup shop

You did it!

You can now tell a CoreDNS problem from a kube-proxy one.

Go deeper (optional)

CoreDNS reads its rules (the Corefile) from the ConfigMap coredns. Where the story breaks: a real food court is one building. Cluster stalls can sit in different rooms, or clouds. And this practice cluster answers names even with CoreDNS down.

Real mission

On a throwaway kind cluster, switch the phone book off.
kubectl -n kube-system scale deployment coredns --replicas=0
Ask for a name. Read how it fails.
kubectl run dnstest --image=busybox:1.37 --rm -it --restart=Never -- nslookup kubernetes.default
Switch it back on.
kubectl -n kube-system scale deployment coredns --replicas=2

0 of 3 steps done · Your ticks are saved.