docker-compose.yml

services:
  # 1. FastAPI 后端服务
  api:
    build:
      context: .
      dockerfile: Dockerfile
    container_name: coderag-api
    ports:
      - "8000:8000"
    environment:
      # LLM 提供方 & 模型，默认走 OpenAI，具体用哪家由 .env 控制
      - RAG_LLM_PROVIDER=${RAG_LLM_PROVIDER:-openai}
      - RAG_LLM_MODEL=${RAG_LLM_MODEL:-gpt-4o-mini}
      # 本地 llama 时用到的 base URL（Windows + GPU 场景）
      - LOCAL_LLM_BASE=${LOCAL_LLM_BASE:-http://llama:8080/v1}
      # OpenAI key（在 .env 里填）
      - OPENAI_API_KEY=${OPENAI_API_KEY:-dummy}
      # Redis 连接（缓存 / RQ 都用得到）
      - REDIS_URL=redis://redis:6379/0
      - HF_ENDPOINT=https://hf-mirror.com
    volumes:
      # 挂载本地 data 目录，保证你的向量库持久化
      - ./data:/app/data
      # 挂载 logs 目录，方便在宿主机看日志
      - ./logs:/app/logs
      # 挂载 chroma 缓存目录，避免每次重启都重新下载 ONNX 模型
      - ./cache/chroma:/root/.cache/chroma
    depends_on:
      - redis
      # llama 是可选服务，不强依赖
      # - llama

  # 2. 本地 LLM 推理服务 (llama.cpp) —— 仅在 Windows + GPU 场景使用
  # llama:
  #   image: ghcr.io/ggml-org/llama.cpp:server-cuda
  #   container_name: coderag-llama
  #   ports:
  #     - "8081:8080" # 宿主机 8081 -> 容器 8080
  #   environment:
  #     - LLAMA_ARG_MODEL=/models/qwen2.5-coder-7b-instruct-q6_k.gguf
  #     - LLAMA_ARG_CTX_SIZE=8192
  #     - LLAMA_ARG_N_GPU_LAYERS=35
  #   volumes:
  #     - ./models:/models
  #   deploy:
  #     resources:
  #       reservations:
  #         devices:
  #           - driver: nvidia
  #             count: all
  #             capabilities: [gpu]

  # 3. Redis（缓存 & RQ）
  redis:
    image: redis:7
    container_name: coderag-redis
    ports:
      - "6379:6379"

  # 4. RQ Worker
  worker:
    build:
      context: .
      dockerfile: Dockerfile
    container_name: coderag-worker
    # 开发阶段挂代码目录，worker 能 import 最新代码
    volumes:
      - .:/app
      - ./data:/app/data
      - ./logs:/app/logs
      # 和 api 一样挂 chroma 缓存，保证 ONNX 模型只下一次
      - ./cache/chroma:/root/.cache/chroma
    environment:
      - RAG_LLM_PROVIDER=${RAG_LLM_PROVIDER:-openai}
      - RAG_LLM_MODEL=${RAG_LLM_MODEL:-gpt-4o-mini}
      - LOCAL_LLM_BASE=${LOCAL_LLM_BASE:-http://llama:8080/v1}
      - OPENAI_API_KEY=${OPENAI_API_KEY:-dummy}
      - REDIS_URL=redis://redis:6379/0
      - HF_ENDPOINT=https://hf-mirror.com
    depends_on:
      - redis
      # 同样不强依赖 llama
      # - llama
    command: ["rq", "worker", "coderag"]