Kuberneteskuberneteshealth-checksprobesreliabilityproduction-ready

Liveness and Readiness Probes

Comprehensive health probe configuration for Kubernetes with startup probes, graceful shutdown, and probe-driven traffic management.


title: "Liveness and Readiness Probes" description: "Comprehensive health probe configuration for Kubernetes with startup probes, graceful shutdown, and probe-driven traffic management." category: "Kubernetes" language: "yaml" tags: ["kubernetes", "health-checks", "probes", "reliability", "production-ready"] createdAt: "2025-03-25" updatedAt: "2025-04-10"

Overview

Kubernetes probe configuration that distinguishes between startup delays, readiness for traffic, and liveness for deadlock detection. Combined with graceful shutdown and PodDisruptionBudgets, this ensures zero-downtime operations.

Code

startupProbe:     # Am I done starting? (checked first)
  failureThreshold: 30
  periodSeconds: 2
readinessProbe:   # Can I receive traffic? (ongoing)
  successThreshold: 2
  failureThreshold: 3
livenessProbe:    # Am I deadlocked? (restart if failing)
  failureThreshold: 3

Notes

  • Startup probe prevents liveness probe from killing slow-starting containers (JVM, ML models)
  • Readiness failure removes the pod from Service endpoints (traffic stops, no restart)
  • Liveness failure triggers a container restart (use only for unrecoverable states like deadlocks)
  • preStop sleep gives ingress controllers time to remove the pod from their routing tables
  • terminationGracePeriodSeconds must exceed preStop sleep + application drain time
  • PodDisruptionBudget prevents voluntary evictions (node drain) from taking all pods offline