🟡 Readiness Probe — Traffic Gate

The readiness probe controls whether a pod receives traffic from a Service. When it fails, the pod's IP is removed from the Service's Endpoints/EndpointSlice — no new connections are routed to it. The pod is NOT restarted; it just stops receiving traffic until the probe passes again.

Key use cases

readinessProbe:
  httpGet:
    path: /readyz
    port: 8080
  initialDelaySeconds: 5
  periodSeconds:      10
  timeoutSeconds:      2
  successThreshold:    1   # 1 success to re-enter rotation
  failureThreshold:    3   # 3 failures to leave rotation

🎛️ Tuning Probe Parameters

Misconfigured probe thresholds are one of the most common causes of production incidents. Here is a field-by-field guide:

Probe Anti-Patterns to Avoid

Observing Probe State

# See probe configuration and recent events
kubectl describe pod my-app-xyz

# Events section shows probe failures:
# Warning  Unhealthy  Liveness probe failed: Get "http://10.0.0.5:8080/healthz": context deadline exceeded
# Warning  Killing    Container my-app failed liveness probe, will be restarted

# Check restart count — tells you if liveness has been firing
kubectl get pod my-app-xyz -o jsonpath='{.status.containerStatuses[0].restartCount}'

# Watch readiness — pod is Ready only when readiness passes
kubectl get pod my-app-xyz -w
# NAME         READY   STATUS    RESTARTS
# my-app-xyz   0/1     Running   0        ← readiness failing
# my-app-xyz   1/1     Running   0        ← readiness passed

🔴 Liveness Probe — Restart on Deadlock

The liveness probe detects containers that are running but stuck — a goroutine leak causing deadlock, a thread pool exhausted, or a process that stopped responding but hasn't exited. Without it, the container sits in Running state forever while serving errors.

When liveness fails

1. kubelet increments the failure counter.
2. Once failureThreshold consecutive failures are reached, kubelet kills the container.
3. The container restarts according to the pod's restartPolicy (almost always Always for long-running services).
4. The pod's restartCount increments — visible in kubectl describe pod.

Complete liveness example with all fields

livenessProbe:
  httpGet:
    path: /healthz
    port: 8080
  initialDelaySeconds: 10   # wait before first probe
  periodSeconds:       15   # probe every 15s
  timeoutSeconds:       3   # fail if no response within 3s
  successThreshold:     1   # 1 success to go healthy (must be 1 for liveness)
  failureThreshold:     3   # 3 consecutive failures → restart

What a good /healthz endpoint looks like

// Go example — liveness checks ONLY internal state
http.HandleFunc("/healthz", func(w http.ResponseWriter, r *http.Request) {
    if isDeadlocked() || threadPoolExhausted() {
        w.WriteHeader(http.StatusServiceUnavailable)
        return
    }
    w.WriteHeader(http.StatusOK)
    w.Write([]byte("ok"))
})

// Readiness checks dependencies too
http.HandleFunc("/readyz", func(w http.ResponseWriter, r *http.Request) {
    if err := db.Ping(); err != nil {
        w.WriteHeader(http.StatusServiceUnavailable)
        return
    }
    w.WriteHeader(http.StatusOK)
})

🟢 Startup Probe — Protect Slow Starters

Before startup probes existed (added in 1.18, GA in 1.20), the only way to handle slow-starting apps was setting a large initialDelaySeconds on liveness — which meant waiting that long for every restart, slowing recovery. The startup probe solves this elegantly.

How startup probe works

• While the startup probe is pending or failing, liveness and readiness probes are disabled.
• Once the startup probe succeeds once, it hands off to liveness and readiness.
• If the startup probe fails failureThreshold × periodSeconds total time, the container is killed (same as liveness).

# Pattern: give a slow app up to 5 minutes to start
# then switch to a tight 15s liveness cycle
startupProbe:
  httpGet:
    path: /healthz
    port: 8080
  failureThreshold: 30    # 30 × 10s = 300s max startup window
  periodSeconds:    10

livenessProbe:
  httpGet:
    path: /healthz
    port: 8080
  periodSeconds:   15
  failureThreshold: 3   # tight once running

🧠 Knowledge Check