# Open-ASR Model Explorer – Docker Compose
# Usage:
#   docker compose up --build          # start both services
#   docker compose up --build backend  # backend only
#   docker compose down -v             # stop and remove volumes
#
# The frontend is served by Nginx on http://localhost:3000.
# The backend FastAPI API is available directly on http://localhost:8000.
# Nginx proxies /api/* → http://backend:8000/* so the UI calls /api/… .

services:

  # ── FastAPI + vLLM backend ──────────────────────────────────────────────────
  backend:
    build:
      context: ./backend
      dockerfile: Dockerfile
    ports:
      - "8000:8000"
    volumes:
      # Persist downloaded HuggingFace model weights across container restarts.
      - hf_cache:/hf_cache
    environment:
      - HF_HOME=/hf_cache
      - TRANSFORMERS_CACHE=/hf_cache
      - HF_TOKEN=${HF_TOKEN}
      - GPU_MEMORY_UTILIZATION=${GPU_MEMORY_UTILIZATION:-0.65}
      - OPENASR_ENGINE_INIT_TIMEOUT_S=${OPENASR_ENGINE_INIT_TIMEOUT_S:-120}
      - OPENASR_VLLM_MAX_MODEL_LEN=${OPENASR_VLLM_MAX_MODEL_LEN:-8192}
      - OPENASR_VLLM_ENFORCE_EAGER=${OPENASR_VLLM_ENFORCE_EAGER:-true}
      - ALLOW_MOCK_FALLBACK=${ALLOW_MOCK_FALLBACK:-false}
    # GPU access – requires the NVIDIA Container Toolkit.
    # Remove the `deploy` block entirely if running CPU-only.
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"]
      interval: 30s
      timeout: 10s
      start_period: 60s
      retries: 3

  # ── React / Nginx frontend ──────────────────────────────────────────────────
  frontend:
    build:
      context: ./frontend
      dockerfile: Dockerfile
    ports:
      - "3000:80"
    depends_on:
      backend:
        condition: service_healthy
    restart: unless-stopped

volumes:
  # Named volume for HuggingFace model weight cache.
  hf_cache: