Startup, Liveness and Readiness Probes

Restart versus remove from traffic

Readiness probes ask "can this pod serve traffic now?"; failing removes the pod from Service endpoints without restarting it (useful during warm-up or temporary dependency issues). Liveness probes ask "is this container stuck?"; failing restarts the container. Startup probes protect slow-starting apps: liveness and readiness checks wait until the startup probe succeeds. Keep liveness checks simple and local (not dependent on databases), or a shared outage will cause restart storms.

Stay healthy, scale with load

Probes tell Kubernetes about health; autoscalers and disruption budgets keep capacity right.

How long before a probe acts? run

I ran this with Python 3. It is a simplified model of Kubernetes behaviour for learning, not the real controller code. With these settings, a startup probe allows up to 150 seconds, liveness restarts a container after about 30 seconds of failures, and readiness removes a pod from traffic after about 10 seconds.

probes = {
    "startupProbe":   {"periodSeconds": 5, "failureThreshold": 30},
    "livenessProbe":  {"periodSeconds": 10, "failureThreshold": 3},
    "readinessProbe": {"periodSeconds": 5, "failureThreshold": 2},
}
for name, p in probes.items():
    print(f"{name:<15} acts after about {p['periodSeconds'] * p['failureThreshold']:>3} s of consecutive failures")
print("startup budget lets a slow app take up to 150 s to start before liveness checks begin")
print("liveness failure -> container restarted; readiness failure -> removed from Service endpoints, not restarted")

Output:

startupProbe    acts after about 150 s of consecutive failures
livenessProbe   acts after about  30 s of consecutive failures
readinessProbe  acts after about  10 s of consecutive failures
startup budget lets a slow app take up to 150 s to start before liveness checks begin
liveness failure -> container restarted; readiness failure -> removed from Service endpoints, not restarted

Probe configuration

Not applied to a live cluster in this course; check field names against the API reference for your version.

startupProbe:
  httpGet: {path: /healthz, port: 8080}
  periodSeconds: 5
  failureThreshold: 30        # up to 150 s to start
livenessProbe:
  httpGet: {path: /healthz, port: 8080}   # cheap, local check
  periodSeconds: 10
  failureThreshold: 3
readinessProbe:
  httpGet: {path: /ready, port: 8080}     # may check dependencies
  periodSeconds: 5
  failureThreshold: 2

Quick check: What happens when a readiness probe fails?

• The node is drained
• The container is restarted
• The pod is removed from Service endpoints but not restarted
• The Deployment is deleted