A health check endpoint reports whether a service process is running or ready to receive traffic. Production deployments commonly separate these responsibilities into liveness and readiness probes. section 9.3.3
Use the probes for different lifecycle decisions:
- The liveness endpoint returns HTTP 200 while the server process is responsive. Failure indicates conditions such as deadlock, out-of-memory failure, or another unrecoverable state, prompting a container restart. section 9.3.3
- The readiness endpoint returns HTTP 200 only when startup is complete and the container can accept traffic. During startup or graceful shutdown, it returns 503 so routing can be withheld. section 9.3.3
- Keep both probes shallow: checking process responsiveness avoids coupling container health to gateway or service dependencies. section 9.3.3
Deployment and shutdown
A runtime server may expose /health/live and /health/ready. While it initializes its module, GenAIPlatform instance, and gRPC connections, it remains live but unready. On termination, it reports 503, allowing in-flight requests to finish before removal from load balancing. section 9.3.3






