A practical guide to Kubernetes probes

The three probe types answer three different questions, and using one to answer another question is how a healthy deployment turns into a restart loop.

What each probe controls

• Startup: has the process finished booting? Suspends the other probes while it runs.
• Readiness: should this pod receive traffic right now?
• Liveness: is this process unrecoverable and worth restarting?

The common failure

Pointing liveness at a handler that checks dependencies. When the database slows down, every pod fails liveness, restarts simultaneously, and turns a degraded dependency into a full outage.

Sensible defaults

startupProbe:
  httpGet: { path: /healthz, port: 8080 }
  failureThreshold: 30
  periodSeconds: 2
readinessProbe:
  httpGet: { path: /readyz, port: 8080 }
  periodSeconds: 5
livenessProbe:
  httpGet: { path: /healthz, port: 8080 }
  periodSeconds: 10
  failureThreshold: 3

Tune the startup threshold to your slowest cold start, and leave liveness generous. Restarting is the most destructive action the platform can take on your behalf.