openapi: 3.0.0
info:
  title: vLLM Self Heal API
  version: 1.0.0
  description: 作业 7 · 检测 OOM、重启 vLLM、钉钉恢复通知
servers:
  - url: http://192.168.110.207:8780
paths:
  /vllm/detect_anomaly:
    get:
      operationId: detect_vllm_anomaly
      summary: 检测 vLLM OOM 或容器异常
      description: 读取 simulate 标记或 GPU 机 docker logs 中的 OOM 关键字
      responses:
        "200":
          description: 成功
          content:
            application/json:
              schema:
                type: object
  /vllm/restart:
    post:
      operationId: restart_vllm_container
      summary: 重启 vLLM 容器
      description: 在 GPU 机执行 docker restart
      responses:
        "200":
          description: 成功
          content:
            application/json:
              schema:
                type: object
  /notify/recovery:
    post:
      operationId: send_recovery_notification
      summary: 发送恢复通知到钉钉
      requestBody:
        required: true
        content:
          application/json:
            schema:
              type: object
              required:
                - message
              properties:
                message:
                  type: string
                  description: 通知正文
                detail:
                  type: string
                  description: 附加细节
      responses:
        "200":
          description: 成功