services:
  # Embedding server service
  embedding_server:
    build:
      context: .
      dockerfile: Dockerfile.embedding_server
    image: bertrend-embedding-server:latest
    container_name: bertrend-embedding-server
    user: "${HOST_UID:-1000}:${HOST_GID:-1000}"
    ports:
      - "6464:6464"  # Embedding server API
    volumes:
      # Mount HF_HOME directory to share with host
      - ${HF_HOME:-.cache/huggingface}:/app/.cache/huggingface
    environment:
      - PYTHONDONTWRITEBYTECODE=1
      - PYTHONUNBUFFERED=1
      - CUDA_VISIBLE_DEVICES=0
      # Ensure these are not empty strings
      - DEFAULT_RATE_LIMIT=${DEFAULT_RATE_LIMIT:-50}
      - DEFAULT_RATE_WINDOW=${DEFAULT_RATE_WINDOW:-60}
      # Set HF_HOME inside container
      - HF_HOME=/app/.cache/huggingface
      # Set UV_CACHE_DIR to a location where the user has write permissions
      - UV_CACHE_DIR=/app/.cache/uv
    healthcheck:
      test: curl --fail --insecure https://localhost:6464/health || exit 1
      interval: 120s
      timeout: 30s
      retries: 3
      start_period: 120s  # Longer start period as model loading can take time
    restart: unless-stopped
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1  # Use 1 GPU for the embedding server
              capabilities: [gpu]

  # Main BERTrend application
  bertrend:
    build:
      context: .
      dockerfile: Dockerfile
    image: bertrend:latest
    container_name: bertrend
    user: "${HOST_UID:-1000}:${HOST_GID:-1000}"
    depends_on:
      - embedding_server
    ports:
      - "8083:8083"  # Topic Analysis demo
      - "8084:8084"  # Weak Signals demo
      - "8081:8081"  # Prospective demo
    volumes:
      - ${BERTREND_BASE_DIR:-.bertrend}:/bertrend
      - ${HF_HOME:-.cache/huggingface}:/app/.cache/huggingface
    environment:
      - NO_PROXY=$NO_PROXY,embedding_server,bertrend # to avoid communication issues between containers
      - no_proxy=$no_proxy,embedding_server,bertrend # to avoid communication issues between containers
      - PYTHONDONTWRITEBYTECODE=1
      - PYTHONUNBUFFERED=1
      - CUDA_VISIBLE_DEVICES=0
      # Set HF_HOME inside container
      - HF_HOME=/app/.cache/huggingface
      # Set UV_CACHE_DIR to a location where the user has write permissions
      - UV_CACHE_DIR=/app/.cache/uv
      # Set NLTK_DATA to a location where the user has write permissions
      - NLTK_DATA=/app/nltk_data
      # BERTrend base dir
      - BERTREND_BASE_DIR=/bertrend/
      - BERTREND_CLIENT_SECRET=bd76aa472dd91aed4a56bf1935dbb802583c119824380d8567086579c0ef3324
      # Configure to use the embedding server
      - EMBEDDING_SERVICE_URL=https://embedding_server:6464
      - EMBEDDING_SERVICE_USE_LOCAL=false
      # OpenAI environment variables (replace with your own values)
      - OPENAI_API_KEY=${OPENAI_API_KEY:-}
      - OPENAI_BASE_URL=${OPENAI_BASE_URL:-}
      - OPENAI_DEFAULT_MODEL=${OPENAI_DEFAULT_MODEL:-gpt-5-mini}
    healthcheck:
      test: curl -f http://localhost:8081/_stcore/health && curl -f http://localhost:8083/_stcore/health && curl -f http://localhost:8084/_stcore/health || exit 1
      interval: 120s
      timeout: 30s
      retries: 3
      start_period: 30s
    restart: unless-stopped
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"
    deploy:
       resources:
         reservations:
           devices:
             - driver: nvidia
               count: 1  # Use 1 GPU for the bertrend server
               capabilities: [gpu]