openapi: 3.0.0 info: title: vLLM Health API version: 1.0.0 description: 作业 5 · 检查 GPU 机 vLLM 容器、GPU 指标、请求排队 servers: - url: http://192.168.110.207:8779 paths: /vllm/status: get: operationId: get_vllm_status summary: 检查 vLLM 容器是否运行 description: 通过 SSH 到 GPU 机执行 docker inspect,返回容器是否 running responses: "200": description: 成功 content: application/json: schema: type: object properties: host: type: string container: type: string running: type: boolean status: type: string started_at: type: string image: type: string healthy: type: boolean /gpu/info: get: operationId: get_gpu_info summary: 获取 GPU 利用率、显存、温度 description: 在 GPU 机执行 nvidia-smi 查询 responses: "200": description: 成功 content: application/json: schema: type: object properties: host: type: string gpus: type: array items: type: object /vllm/queue: get: operationId: get_request_queue summary: 获取当前排队请求数 description: 读取 vLLM /metrics 中的 num_requests_waiting responses: "200": description: 成功 content: application/json: schema: type: object properties: host: type: string model: type: string waiting: type: number running: type: number kv_cache_usage_perc: type: number source: type: string