k8s Flashcards
← Back to blog
ServicesNetworkingTroubleshootingCKAD

Service Has No Endpoints: Where Your Traffic Disappears

curl times out but pods look Running? An empty Endpoints list means the Service selector does not match any Ready pod — trace the chain before blaming the network.

1 min read

Users report: "The API is down."

kubectl get pods -n shop -l app=api
api-7f2a   1/1   Running
api-8b1c   1/1   Running

Pods look fine. You curl the Service:

kubectl run tmp --rm -it --image=curlimages/curl -- curl -sS --max-time 3 http://api.shop.svc:8080/health

Timeout.

Check Endpoints first

kubectl get endpoints api -n shop
# or
kubectl get endpointslice -n shop -l kubernetes.io/service-name=api
NAME   ENDPOINTS   AGE
api    <none>      3d

No endpoints means the Service has zero Ready backends. kube-proxy (or your dataplane) has nothing to route to — regardless of Running pods.

The usual mismatches

1. Selector ≠ pod labels

# Service
selector:
  app: api
# Pod labels
app: api-v2   # typo or rollout label drift

Fix labels or update the Service selector (carefully — selector is immutable on existing Services; often you fix the Deployment template labels).

2. Pods not Ready

Running but failing readiness → excluded from Endpoints. Back to describe pod and probe config.

3. Wrong port

Service targetPort: 8080 but container listens on 80. Endpoints may exist but connections fail — different symptom, still a Service spec issue.

4. Wrong namespace

Calling http://api from another namespace needs FQDN: api.shop.svc.cluster.local.

Trace order

  1. kubectl get svc,ep -n shop (or EndpointSlices)
  2. Compare Service .spec.selector to pod labels
  3. kubectl get pods -l <selector> -o wide — Ready column
  4. If not Ready → probe/path/port on pod

Practice

Service Connectivity on Decision Trainer walks this chain with graded first steps — before you reach for NetworkPolicy or node firewall rules.