Readiness probe failed (Kubernetes) — how to fix
Warning Unhealthy 12s (x6 over 62s) kubelet Readiness probe failed: Get "http://10.244.1.17:8080/health": dial tcp 10.244.1.17:8080: connect: connection refused
Warning Unhealthy 5s kubelet Readiness probe failed: HTTP probe failed with statuscode: 503
Warning Unhealthy 3s kubelet Readiness probe failed: Get "http://10.244.1.17:8080/ready": context deadline exceeded (Client.Timeout exceeded while awaiting headers)
Quick fix: Run kubectl describe pod <pod> to see the exact probe failure, then test the endpoint from inside: kubectl exec <pod> -- wget -qO- localhost:8080/health. Fix the probe's port/path to match the app, make the app bind 0.0.0.0, and raise initialDelaySeconds/timeoutSeconds or add a startupProbe for slow starts.
What it means
A readiness probe decides whether a pod should receive traffic. When it fails, the container keeps running but shows READY 0/1, and the pod is taken out of the Service's endpoints — so requests through the Service or Ingress get 502/503 or "no endpoints".
Unlike a liveness probe, a failing readiness probe never restarts the container. If you also see restarts, the liveness probe is failing too (or the app crashes — see CrashLoopBackOff).
kubectl get pods
kubectl describe pod <pod> # Events show the exact probe error
kubectl get endpoints <service> # empty = no ready pods
kubectl logs <pod>
The text after Readiness probe failed: tells you which cause applies.
Cause 1: connection refused — wrong port, or app not listening yet
The probe connected to the pod IP and nothing was listening on that port.
- Check the port the app actually uses (
kubectl logs, orkubectl exec <pod> -- netstat -tlnp/ss -tlnpif available). - The app must listen on
0.0.0.0(all interfaces), not127.0.0.1. The kubelet probes the pod IP, so a localhost-only server refuses it. E.g. Flaskapp.run(host="0.0.0.0"), Nodeserver.listen(8080, "0.0.0.0"). - A named port (
port: http) must match acontainerPortname.
readinessProbe:
httpGet:
path: /health
port: 8080 # must match what the app listens on
periodSeconds: 5
Cause 2: statuscode: 503 / 404 — endpoint returns failure
HTTP probes succeed only with status 200–399. A 404 means the path is wrong; a 5xx means your health handler reports unhealthy (often a dependency such as the database is unreachable). Test it from inside the pod:
kubectl exec -it <pod> -- wget -qO- http://localhost:8080/health
# or: curl -i http://localhost:8080/health
Keep readiness checks cheap and decide carefully which dependencies should make the pod unready — if every pod checks the same database, a DB blip takes all of them out of service at once.
Spring Boot exposes /actuator/health/readiness when management.endpoint.health.probes.enabled=true (on by default in Kubernetes).
Cause 3: context deadline exceeded — too slow
The default timeoutSeconds is 1 second. A slow health endpoint, CPU throttling (low resources.limits.cpu) or a JVM warming up will exceed it.
readinessProbe:
httpGet: { path: /ready, port: 8080 }
initialDelaySeconds: 10
periodSeconds: 10
timeoutSeconds: 5
failureThreshold: 3
Cause 4: slow startup
For apps that take a long time to boot, add a startupProbe; readiness and liveness checks wait until it succeeds:
startupProbe:
httpGet: { path: /health, port: 8080 }
periodSeconds: 5
failureThreshold: 30 # up to 150 s to start
Readiness failures during the first seconds after start are normal and harmless; only persistent ones matter.
Cause 5: exec probe returns non-zero
For exec probes, the command's exit code must be 0. Run the same command manually:
kubectl exec <pod> -- sh -c 'pg_isready -U postgres; echo exit=$?'
Common problems: the binary isn't in the image (distroless/scratch images have no shell), or the command needs env vars it doesn't get.
Validate the YAML
Probe fields are easy to mis-indent (they belong under the container, not the pod spec). Check with the YAML validator and kubectl apply --dry-run=server -f deploy.yaml.
Last updated 2026-11-19.