Health checks

Orchestrators use health endpoints to route traffic and restart unhealthy instances. Liveness answers "should this process restart?" Readiness answers "can this instance accept traffic?" Startup probes protect slow-booting apps. Dependency checks belong in readiness, not liveness, to avoid restart loops.

On interviews: distinguish liveness versus readiness for a service that cannot reach Kafka briefly.

Common pitfalls: checking downstream DB in liveness; readiness always true; health endpoint doing heavy work.

The trade-off is flexibility versus complexity—know when the simpler path is enough.

Checklist:

• Separate liveness, readiness, and startup probes.
• Keep liveness cheap and process-focused.
• Remove from load balancer when not ready.
• Version health responses for debugging.