ray.serve.llm.LLMServer.record_routing_stats

async LLMServer.record_routing_stats() → Dict[str, Any]

Serve request-router hook, polled by the controller.

Surfaces this replica’s routing stats (the engine’s KV-events endpoint for KV-aware routing); the deployment’s KVRouterActor reads them off the LongPoll replica snapshot to register the worker.