itsual Health Checks and Probes
Overview
Bad health checks cause either unnecessary restarts or traffic being sent to broken instances. Make them accurate and cheap.
When to Use
- Deploying services to Kubernetes or similar platforms
- Implementing /health, /ready, /live endpoints
- Debugging flapping or slow-to-start services
Probe Types
- Liveness – Is the process alive? (should it be restarted?)
- Readiness – Can it accept traffic right now?
- Startup – Has it finished initializing?
Practices
- Keep checks fast and free of heavy dependencies when possible
- Don’t make liveness depend on downstream systems (that causes cascading restarts)
- Make readiness reflect actual ability to serve useful traffic
- Include critical but cheap self-checks
- Log transitions for easier debugging
Verification
- Probes correctly distinguish “starting”, “ready”, “alive but not ready”, and “dead”
- False positives/negatives are rare under normal conditions