Liveness vs Readiness: Health Check Endpoints That Wont Restart-Loop Your Service

Short summary

Splitting a single /health endpoint into liveness and readiness probes prevents orchestrators from restart-looping healthy pods during transient dependency failures. Liveness should only verify the process itself is alive — never check external dependencies — while readiness verifies whether the instance can serve traffic by checking databases, caches, and downstream APIs with short timeouts. Following patterns like the IETF health+json draft and avoiding cascading readiness checks keeps production services stable during partial outages.

• •Liveness checks must be internal-only; external dependency checks cause restart storms
• •Readiness checks verify dependencies with timeouts and return 503 when unready, letting load balancers route around degraded instances
• •Use structured health response formats and test probe endpoints like any other API

Generated with AI, which can make mistakes.