networks: # Define a custom network
  internal_network:
    driver: bridge

services:
  nginx-proxy:
    image: nginx:latest
    container_name: nginx-proxy
    ports:
      - "80:80"
      - "443:443"
    volumes:
      - ./nginx.conf:/etc/nginx/conf.d/default.conf
    depends_on:
      vllm-server:
        condition: service_healthy # Wait for vLLM to be healthy
      open-webui:
        condition: service_started # Wait for Open WebUI to start
    networks: # Connect Nginx to the internal network
      - internal_network
    restart: always

  vllm-server:
    image: vllm/vllm-openai:nightly
    container_name: vllm-server
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: ["gpu"]
    volumes:
      - /srv/huggingface:/root/.cache/huggingface
    ipc: host
    environment:
      - HUGGING_FACE_HUB_TOKEN=${HF_TOKEN}
      - CUDA_VISIBLE_DEVICES=0
      - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
    command: --model ${HF_MODEL} --max-model-len 4096
    healthcheck:
      test: ["CMD-SHELL", "curl -f --connect-timeout 4 --max-time 9 http://127.0.0.1:8000/health || exit 1"]
      interval: 15s
      timeout: 10s
      retries: 5
      start_period: 1200s
    networks: # Connect vllm-server to the internal network
      - internal_network
    restart: always

  open-webui:
    image: ghcr.io/open-webui/open-webui:main
    container_name: open-webui
    depends_on:
      # For faster startup, don't wait for vllm-server to be fully healthy (service_healthy)
      vllm-server:
        condition: service_started
    volumes:
      - open-webui_data:/app/backend/data # Use a named volume for persistence
    environment:
      OPENAI_API_BASE_URL: http://vllm-server:8000/v1
    networks: # Connect open-webui to the internal network
      - internal_network
    restart: always

volumes:
  open-webui_data: # Define the named volume