Container Probes

May 20, 2026

Probe types

Probe	When fails	Action
Liveness	Unhealthy while running	Restart container
Readiness	Not ready for traffic	Removed from Service endpoints
Startup	Slow/failed start	Kill container; delays liveness until success

Order: startup runs first → then readiness + liveness concurrently
Readiness is most important in practice

Verification methods

Method	YAML	Success
Exec	`exec.command`	Exit code 0
HTTP	`httpGet` path/port	Status 200–399
TCP	`tcpSocket.port`	Connection opens
gRPC	`grpc`	gRPC health protocol

Use named ports in probes: port: nodejs-port

Timing (all probes)

Field	Default	Meaning
`initialDelaySeconds`	0	Wait before first check
`periodSeconds`	10	Interval between checks
`timeoutSeconds`	1	Max check duration
`failureThreshold`	3	Failures before action
`successThreshold`	1	Successes to recover after fail

YAML examples

readinessProbe:
  httpGet:
    path: /
    port: nodejs-port
  initialDelaySeconds: 2
  periodSeconds: 8

livenessProbe:
  exec:
    command: ["/bin/sh", "-c", "test -f /tmp/healthy"]

startupProbe:
  httpGet:
    path: /healthz
    port: 8080
  failureThreshold: 30
  periodSeconds: 10    # 30 * 10s max startup window

kubectl

kubectl get po <n>        # READY 0/1 if readiness failing
kubectl describe po <n>   # probe config + events

Failures

Symptom	Cause
`CrashLoopBackOff`	Wrong liveness → restarts
No traffic	Wrong readiness: Pod stays, removed from endpoints

Exam tips

Liveness too aggressive on slow apps → CrashLoop: use startupProbe
Readiness failure ≠ Pod restart (only removed from Service)
httpGet must use correct path and port (named port preferred)