A readiness check determines whether a service instance can receive traffic. The runtime server’s /health/ready endpoint returns HTTP 200 when ready and 503 while initialization is incomplete or the instance is draining. An orchestrator keeps an unready instance out of the traffic pool. section 9.3.3 section 9.8
Readiness differs from liveness. A process can be live while it imports the developer’s module, initializes the platform instance, and establishes gateway connections. A failed readiness check withholds traffic; a failed liveness check signals an unrecoverable process problem and prompts a restart. section 9.3.3
On graceful shutdown, the runtime server returns 503 for readiness so the orchestrator stops sending new requests. The container then completes in-flight requests before exiting, reducing dropped requests during deployments and scaling. section 9.3.3
Keep the probe shallow rather than making it depend on a gateway ping or a test call to another service. Coupling readiness to dependency health can turn a dependency restart into a cascading outage. section 9.3.3 section 9.8
Gating a workflow deployment
A workflow deployment creates replicas and waits for readiness probes to pass. The Workflow Service then registers healthy endpoint addresses with the API Gateway; newly scaled replicas are added after they pass, and removed replicas leave the routing table. Deployment status becomes healthy once enough replicas pass their probes. section 9.6.2 section 9.5.2
A generated Kubernetes Deployment points its liveness and readiness probes at the runtime health endpoints. During a rolling update, new containers start while old containers are terminated, making readiness and draining part of the traffic transition. section 9.7 section 9.3.3







