Question

A team reports that requests to their ClusterIP Service named 'web-svc' in the 'shop' namespace are timing out. The backing Deployment has 3 Running and Ready pods. Running 'kubectl get endpoints web-svc -n shop' shows the ENDPOINTS column is '<none>'. What is the MOST likely cause?

Accepted Answer

The Service's spec.selector labels do not match the labels on the pods, so no endpoints are populated.

Answer

CoreDNS is down, preventing the Service name from resolving to a ClusterIP.

Answer

kube-proxy is not running on any node, so iptables rules for the Service were never created.

Answer

The pods are failing their readiness probes and have been removed from the endpoint list.