Skip to content

Kubernetes Probes

The three health endpoints map directly onto Kubernetes probes:

Endpoint Probe Semantics
/actuator/health/live livenessProbe Process is responding. Never checks dependencies — a dead database must not get your pod restarted.
/actuator/health/ready readinessProbe Dependencies are healthy; 503 takes the pod out of the Service until they recover.
/actuator/health dashboards / humans Full per-component detail. 
# deployment.yaml (container spec)
livenessProbe:
  httpGet:
    path: /actuator/health/live
    port: 8000
  initialDelaySeconds: 5
  periodSeconds: 10
readinessProbe:
  httpGet:
    path: /actuator/health/ready
    port: 8000
  initialDelaySeconds: 5
  periodSeconds: 5

Rules of thumb

  • Put only dependencies you can recover from behind readiness (database, message broker). A pod that is NotReady stops receiving traffic but keeps running — that is what you want during a database hiccup.
  • Never point the liveness probe at /health or /health/ready: a slow dependency would restart perfectly healthy pods in a loop.