# =============================================================================
# Finance-DeepSeek — Docker Compose Stack
# =============================================================================
# Usage:
#   docker compose up --build        # build & start all services
#   docker compose up api            # start only API service
#   docker compose run --rm trainer  # run training job once
#   docker compose run --rm data-gen # run data generation once
#   docker compose run --rm indexer  # build FAISS index from corpus
# =============================================================================

services:
  # ---------------------------------------------------------------------------
  # API Service — FastAPI + Uvicorn (GPU-enabled)
  # ---------------------------------------------------------------------------
  api:
    build:
      context: .
      dockerfile: Dockerfile
      target: production
    image: finance-deepseek:latest
    container_name: finance-deepseek-api
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    ports:
      - "8000:8000"
    volumes:
      - ./models:/app/models:ro
      - ./data:/app/data
      - ./outputs:/app/outputs
      - ./logs:/app/logs
      - ./config.yaml:/app/config.yaml:ro
    environment:
      - CUDA_VISIBLE_DEVICES=0
      - TRANSFORMERS_CACHE=/app/models/.cache
      - HF_HOME=/app/models/.cache/huggingface
      - PYTHONPATH=/app
      - LOG_LEVEL=INFO
    env_file:
      - .env
    healthcheck:
      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s
    restart: unless-stopped
    command: >
      uvicorn api.main:app
      --host 0.0.0.0
      --port 8000
      --workers 1
      --log-level info

  # ---------------------------------------------------------------------------
  # Training Service — QLoRA fine-tuning (one-shot, GPU required)
  # ---------------------------------------------------------------------------
  trainer:
    build:
      context: .
      dockerfile: Dockerfile
      target: production
    image: finance-deepseek:latest
    container_name: finance-deepseek-trainer
    profiles: ["training"]
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    volumes:
      - ./models:/app/models:ro
      - ./data:/app/data
      - ./outputs:/app/outputs
      - ./logs:/app/logs
      - ./config.yaml:/app/config.yaml:ro
    environment:
      - CUDA_VISIBLE_DEVICES=0
      - TRANSFORMERS_CACHE=/app/models/.cache
      - HF_HOME=/app/models/.cache/huggingface
      - PYTHONPATH=/app
      - LOG_LEVEL=INFO
    env_file:
      - .env
    command: >
      python -m training.qlora_trainer

  # ---------------------------------------------------------------------------
  # Data Generation Service — SFT dataset generation (one-shot)
  # ---------------------------------------------------------------------------
  data-gen:
    build:
      context: .
      dockerfile: Dockerfile
      target: production
    image: finance-deepseek:latest
    container_name: finance-deepseek-data-gen
    profiles: ["data-gen"]
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    volumes:
      - ./models:/app/models:ro
      - ./data:/app/data
      - ./outputs:/app/outputs
      - ./config.yaml:/app/config.yaml:ro
    environment:
      - CUDA_VISIBLE_DEVICES=0
      - TRANSFORMERS_CACHE=/app/models/.cache
      - HF_HOME=/app/models/.cache/huggingface
      - PYTHONPATH=/app
    env_file:
      - .env
    command: >
      python -m training.data_generator

  # ---------------------------------------------------------------------------
  # Indexer Service — FAISS index build from corpus (one-shot)
  # ---------------------------------------------------------------------------
  indexer:
    build:
      context: .
      dockerfile: Dockerfile
      target: production
    image: finance-deepseek:latest
    container_name: finance-deepseek-indexer
    profiles: ["indexer"]
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    volumes:
      - ./models:/app/models:ro
      - ./data:/app/data
      - ./outputs:/app/outputs
      - ./config.yaml:/app/config.yaml:ro
    environment:
      - CUDA_VISIBLE_DEVICES=0
      - TRANSFORMERS_CACHE=/app/models/.cache
      - HF_HOME=/app/models/.cache/huggingface
      - PYTHONPATH=/app
    env_file:
      - .env
    command: >
      python -c "from rag.indexer import build_from_corpus_dir; build_from_corpus_dir('/app/data/corpus')"