o پin@sXdZddlZddlZddlmZddlZddlmZmZe e Z Gdddej Z dS)z Standard gRPC health check service implementation for Kubernetes probes. This module implements the grpc.health.v1.Health service protocol, enabling native Kubernetes gRPC health probes for liveness and readiness checks. N) AsyncIterator) health_pb2health_pb2_grpcc@sveZdZdZdZdZdefddZddZd d Z d e j d e j jd e jfddZd e j d e j jd ee jfddZdS)SGLangHealthServicera Standard gRPC health check service implementation for Kubernetes probes. Implements grpc.health.v1.Health protocol. Supports two service levels: 1. Overall server health (service="") - for liveness probes 2. SGLang service health (service="sglang.grpc.scheduler.SglangScheduler") - for readiness probes Health status lifecycle: - NOT_SERVING: Initial state, model loading, or shutting down - SERVING: Model loaded and ready to serve requests z%sglang.grpc.scheduler.SglangSchedulerscheduler_infocCs@||_||_i|_tjj|j|j<tjj|j|j<t ddS)z Initialize health servicer. Args: request_manager: GrpcRequestManager instance for checking server state scheduler_info: Dict containing scheduler metadata z(Standard gRPC health service initializedN) request_managerr_serving_statusrHealthCheckResponse NOT_SERVINGOVERALL_SERVERSGLANG_SERVICEloggerinfo)selfrrrS/home/ubuntu/.local/lib/python3.10/site-packages/sglang/srt/grpc/health_servicer.py__init__$s  zSGLangHealthServicer.__init__cC.tjj|j|j<tjj|j|j<tddS)z;Mark services as SERVING - call this after model is loaded.z$Health service status set to SERVINGN)rr SERVINGr r r rrrrrr set_serving:   z SGLangHealthServicer.set_servingcCr)z9Mark services as NOT_SERVING - call this during shutdown.z(Health service status set to NOT_SERVINGN)rr r r r r rrrrrrset_not_servingDrz$SGLangHealthServicer.set_not_servingrequestcontextreturncst|j}td|d|jjrtdtjtjjdS||jkr@|j |jtjj}tdtjj |tj|dS||j kr|j |j tjj}|tjjkr`tdtj|dSt|jj}|dkrt|jjdkrtd |d d t|jjd tjtjjdStd tjtjjdStd|d|tjj|d|tjtjjdS)a" Standard health check for Kubernetes probes. Args: request: Contains service name ("" for overall, or specific service) context: gRPC context Returns: HealthCheckResponse with SERVING/NOT_SERVING/SERVICE_UNKNOWN status z#Health check request for service: ''z%Health check: Server is shutting down)statuszOverall health check: z/Service health check: NOT_SERVING (base status)rz0Service health check: Scheduler not responsive (z.1fzs since last receive, z pending requests)zService health check: SERVINGz#Health check for unknown service: 'zUnknown service: )servicerdebugrgracefully_exitrr r r r get ServingStatusNamer rtimelast_receive_tstamplen rid_to_statewarningset_codegrpc StatusCode NOT_FOUND set_detailsSERVICE_UNKNOWN)rrr service_namer base_statustime_since_last_receiverrrCheckNs^            zSGLangHealthServicer.CheckcCs6|j}td|d|||IdH}|VdS)a Streaming health check - sends updates when status changes. For now, just send current status once (Kubernetes doesn't use Watch). A full implementation would monitor status changes and stream updates. Args: request: Contains service name context: gRPC context Yields: HealthCheckResponse messages when status changes z#Health watch request for service: 'rN)r rr!r4)rrrr1responserrrWatchs  zSGLangHealthServicer.WatchN)__name__ __module__ __qualname____doc__r r dictrrrrHealthCheckRequestr,aioServicerContextr r4rr6rrrrrs*   Ur)r:loggingr&typingrr,grpc_health.v1rr getLoggerr7rHealthServicerrrrrrs