Question

A web application backed by a Deployment with 3 replicas is intermittently returning connection errors from clients. All 3 pods show STATUS 'Running' but 'kubectl get pods' shows READY as '0/1' for all three. The Service has a correct selector matching the pods, yet 'kubectl get endpoints web-svc' shows no addresses in the ENDPOINTS column. What is the MOST likely cause?

Accepted Answer

The pods are failing their readiness probe, so the endpoints controller excludes them from the Service endpoints.

Answer

The Service is of type ClusterIP and therefore cannot register endpoints until an Ingress is created.

Answer

kube-proxy is not running on the nodes, which prevents endpoints from being populated.

Answer

The pods are missing a liveness probe, so kubelet refuses to add them to the endpoint list.