AgentSkillsCN

health-checks

实施存活、就绪和依赖健康检查

SKILL.md
--- frontmatter
name: health-checks
description: "Implement liveness, readiness, and dependency health checks"
triggers:
  - "health check endpoints"
  - "liveness probe"
  - "readiness probe"
  - "kubernetes health checks"
priority: 1

Health Checks

Different checks serve different purposes. Don't conflate them.

Check Types

EndpointPurposeOn FailureShould Check
/health/liveProcess alive?K8s restarts podOnly process responsiveness
/health/readyCan handle traffic?K8s removes from LBDB, cache, critical deps
/health/startupInit complete?K8s waitsInitialization status

Liveness (Simple)

code
Return 200 OK immediately. Never check dependencies.

Checking DB in liveness = pod restarts when DB is down = cascading failure.

Readiness (Dependency Checks)

code
For each dependency:
  → Check with timeout (1-2s)
  → Record healthy/unhealthy status
Return 503 if any critical dependency unhealthy

Response Format

json
{
  "status": "healthy|unhealthy",
  "version": "1.2.3",
  "dependencies": {
    "database": "healthy",
    "cache": "unhealthy: connection refused"
  }
}

Kubernetes Config

yaml
livenessProbe:
  httpGet: { path: /health/live, port: 8080 }
  periodSeconds: 10
  failureThreshold: 3

readinessProbe:
  httpGet: { path: /health/ready, port: 8080 }
  periodSeconds: 5
  failureThreshold: 3

startupProbe:
  httpGet: { path: /health/startup, port: 8080 }
  periodSeconds: 5
  failureThreshold: 30  # 150s max startup

Anti-Patterns

  • •Liveness checks dependencies → Cascading restarts
  • •No timeout on checks → Health endpoint hangs
  • •No caching → Thundering herd on health endpoints

References

  • •references/platforms/{platform}/health-checks.md