docker-compose.yaml

networks:
  openlpr-network:
    driver: bridge

volumes:
  lpr_data:
    driver: local
  lpr_media:
    driver: local
  lpr_static:
    driver: local
  model_files:
    driver: local
  prometheus_data:
    driver: local
  grafana_data:
    driver: local
  lpr_metrics:
    driver: local

services:
  # Core Services
  lpr-app:
    env_file:
      - .env.llamacpp
    image: ghcr.io/faisalthaheem/open-lpr:latest
    container_name: lpr-app
    volumes:
      # SQLite database persistence
      - ./container-data:/app/data
      # Media files persistence (uploaded and processed images)
      - ./container-media:/app/media
      # Static files (optional - for development)
      - ./staticfiles:/app/staticfiles
      # Metrics persistence
      - ./container-metrics:/app/metrics
    # Run as root initially for setup, then entrypoint switches to django user
    user: "0:0"
    environment:
      # Django Settings
      - SECRET_KEY=${SECRET_KEY:-django-insecure-change-me-in-production}
      - DEBUG=${DEBUG:-False}
      - ALLOWED_HOSTS=${ALLOWED_HOSTS:-localhost,127.0.0.1,0.0.0.0,lpr-app}
      - CORS_ALLOWED_ORIGINS=${CORS_ALLOWED_ORIGINS:-http://localhost:3000}
      - CORS_ALLOW_PRIVATE_NETWORK=${CORS_ALLOW_PRIVATE_NETWORK:-False}
      
      # Rate Limiting
      - RATE_LIMIT_ENABLE=${RATE_LIMIT_ENABLE:-True}
      - RATE_LIMIT_RATE=${RATE_LIMIT_RATE:-2/min}
      - RATE_LIMIT_EXCLUDE_PATHS=${RATE_LIMIT_EXCLUDE_PATHS:-/health/,/api/v1/health-light/}
      - RATE_LIMIT_INCLUDE_PATHS=${RATE_LIMIT_INCLUDE_PATHS:-/api/v1/ocr/}
      # Qwen3-VL API Configuration - Updated to use local LlamaCpp service
      - QWEN_API_KEY=${QWEN_API_KEY:-sk-llamacpp-local}
      - QWEN_BASE_URL=${QWEN_BASE_URL:-http://llamacpp-cpu:8000/v1}
      - QWEN_MODEL=${QWEN_MODEL:-gpt-4-vision-preview}  # Model alias for compatibility
      
      # File Upload Settings
      - UPLOAD_FILE_MAX_SIZE=${UPLOAD_FILE_MAX_SIZE:-10485760}
      - MAX_BATCH_SIZE=${MAX_BATCH_SIZE:-10}
      
      # Database Configuration
      - DATABASE_PATH=/app/data/db.sqlite3
      
      # Metrics persistence
      - METRICS_FILE_PATH=/app/metrics/metrics_state.json
      
      # Prometheus URL for availability proxy
      - PROMETHEUS_URL=${PROMETHEUS_URL:-http://prometheus:9090}
      
      # Scheduler - auto-detected under Gunicorn, set explicitly if needed
      # - RUN_SCHEDULER=true
      
      # Canary Configuration
      - CANARY_HEADER_NAME=${CANARY_HEADER_NAME:-X-Canary-Request}
      - CANARY_HEADER_VALUE=${CANARY_HEADER_VALUE:-true}
      - CANARY_ENABLED=${CANARY_ENABLED:-true}
      
      # Optional: Superuser creation
      - DJANGO_SUPERUSER_USERNAME=${DJANGO_SUPERUSER_USERNAME:-}
      - DJANGO_SUPERUSER_EMAIL=${DJANGO_SUPERUSER_EMAIL:-}
      - DJANGO_SUPERUSER_PASSWORD=${DJANGO_SUPERUSER_PASSWORD:-}
    networks:
      - openlpr-network
    # Port is configurable via environment variable
    ports:
      - "${LPR_APP_PORT:-8000}:8000"
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "python", "manage.py", "check"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    entrypoint: ["./docker-entrypoint.sh"]
    command: ["gunicorn", "--bind", "0.0.0.0:8000", "--workers", "1", "--timeout", "120", "lpr_project.wsgi:application"]
    profiles:
      - core

  spa:
    image: ghcr.io/faisalthaheem/open-lpr-spa:latest
    container_name: spa
    environment:
      - BACKEND_API_URL=${BACKEND_API_URL:-http://lpr-app:8000}
      - PORT=${SPA_PORT:-3000}
    ports:
      - "${SPA_PORT:-3000}:${SPA_PORT:-3000}"
    networks:
      - openlpr-network
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "wget --spider --quiet http://127.0.0.1:${SPA_PORT:-3000} || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 30s
    profiles:
      - core

  prometheus:
    # Use custom image with config baked in
    image: ghcr.io/faisalthaheem/open-lpr-prometheus:latest
    container_name: prometheus
    # Port is configurable via environment variable
    ports:
      - "${PROMETHEUS_PORT:-9090}:9090"
    volumes:
      # Only need data volume, no config mount needed
      - prometheus_data:/prometheus
    networks:
      - openlpr-network
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "wget --spider --quiet http://127.0.0.1:9090/-/healthy || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    profiles:
      - core

  grafana:
    # Use custom image with provisioning baked in
    image: ghcr.io/faisalthaheem/open-lpr-grafana:latest
    container_name: grafana
    # Port is configurable via environment variable
    ports:
      - "${GRAFANA_PORT:-3000}:3000"
    environment:
      - GF_SECURITY_ADMIN_USER=${GRAFANA_USER:-admin}
      - GF_SECURITY_ADMIN_PASSWORD=${GRAFANA_PASSWORD:-admin}
      - GF_USERS_ALLOW_SIGN_UP=false
    volumes:
      # Only need data volume, no provisioning mount needed
      - grafana_data:/var/lib/grafana
    networks:
      - openlpr-network
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "wget --spider --quiet http://127.0.0.1:3000/api/health || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    profiles:
      - core

  blackbox-exporter:
    # Use custom image with config baked in
    image: ghcr.io/faisalthaheem/open-lpr-blackbox:latest
    container_name: blackbox-exporter
    # Port is configurable via environment variable
    ports:
      - "${BLACKBOX_PORT:-9115}:9115"
    networks:
      - openlpr-network
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "wget --spider --quiet http://127.0.0.1:9115 || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    profiles:
      - core

  lpr-canary:
    env_file:
      - .env.llamacpp
    image: ghcr.io/faisalthaheem/open-lpr-canary:latest
    container_name: lpr-canary
    # Port is configurable via environment variable
    ports:
      - "${CANARY_PORT:-9100}:9100"
    environment:
      - LPR_API_URL=${LPR_API_URL:-http://lpr-app:8000/api/v1/ocr/}
      - CANARY_HEADER_NAME=${CANARY_HEADER_NAME:-X-Canary-Request}
      - CANARY_HEADER_VALUE=${CANARY_HEADER_VALUE:-true}
      - CANARY_INTERVAL=${CANARY_INTERVAL:-900}
      - PROMETHEUS_PORT=9100
    networks:
      - openlpr-network
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "wget --spider --quiet http://localhost:9100/metrics || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    profiles:
      - core

  # Inference Services
  llamacpp-cpu:
    env_file:
      - .env.llamacpp
    image: ghcr.io/ggml-org/llama.cpp:server
    container_name: llamacpp-cpu
    ports:
      - "8001:8000"  # Use 8001 to avoid conflict with OpenLPR on 8000
    volumes:
      # Model storage
      - ./model_files:/models
      - ./model_files_cache:/root/.cache/
    environment:
      # Model configuration
      - LLAMA_ARG_MODEL=/models/${MODEL_FILE:-Qwen3-VL-4B-Instruct-Q5_K_M.gguf}
      - LLAMA_ARG_HF_REPO=${MODEL_REPO:-unsloth/Qwen3-VL-4B-Instruct-GGUF}
      - LLAMA_ARG_HF_FILE=${MODEL_FILE:-Qwen3-VL-4B-Instruct-Q5_K_M.gguf}
      - LLAMA_ARG_MMPROJ_URL=${MMPROJ_URL:-https://huggingface.co/unsloth/Qwen3-VL-4B-Instruct-GGUF/resolve/main/mmproj-BF16.gguf}
      - HF_TOKEN=${HF_TOKEN}
      - LLAMA_ARG_JINJA=1
      - LLAMA_ARG_CHAT_TEMPLATE=chatml
      - LLAMA_ARG_HOST=0.0.0.0
      - LLAMA_ARG_PORT=8000
    networks:
      - openlpr-network
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    profiles:
      - cpu

  llamacpp-amd-vulkan:
    env_file:
      - .env.llamacpp
    image: ghcr.io/ggml-org/llama.cpp:server-vulkan
    container_name: llamacpp-amd-vulkan
    ports:
      - "8001:8000"  # Use 8001 to avoid conflict with OpenLPR on 8000
    volumes:
      # Model storage
      - ./model_files:/models
      - ./model_files_cache:/root/.cache/
    environment:
      # Model configuration
      - LLAMA_ARG_MODEL=/models/${MODEL_FILE:-Qwen3-VL-4B-Instruct-Q5_K_M.gguf}
      - LLAMA_ARG_HF_REPO=${MODEL_REPO:-unsloth/Qwen3-VL-4B-Instruct-GGUF}
      - LLAMA_ARG_HF_FILE=${MODEL_FILE:-Qwen3-VL-4B-Instruct-Q5_K_M.gguf}
      - LLAMA_ARG_MMPROJ_URL=${MMPROJ_URL:-https://huggingface.co/unsloth/Qwen3-VL-4B-Instruct-GGUF/resolve/main/mmproj-BF16.gguf}
      - HF_TOKEN=${HF_TOKEN}
      - LLAMA_ARG_JINJA=1
      - LLAMA_ARG_CHAT_TEMPLATE=chatml
      - LLAMA_ARG_HOST=0.0.0.0
      - LLAMA_ARG_PORT=8000
      - LLAMA_ARG_DEVICE=vulkan0
    networks:
      - openlpr-network
    devices:
      - /dev/kfd
      - /dev/dri
    security_opt:
      - seccomp=unconfined
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    profiles:
      - amd-vulkan

  llamacpp-nvidia-cuda:
    env_file:
      - .env.llamacpp
    image: ghcr.io/ggml-org/llama.cpp:server-cuda
    container_name: llamacpp-nvidia-cuda
    ports:
      - "8001:8000"  # Use 8001 to avoid conflict with OpenLPR on 8000
    volumes:
      # Model storage
      - ./model_files:/models
      - ./model_files_cache:/root/.cache/
    environment:
      # Model configuration
      - LLAMA_ARG_MODEL=/models/${MODEL_FILE:-Qwen3-VL-4B-Instruct-Q5_K_M.gguf}
      - LLAMA_ARG_HF_REPO=${MODEL_REPO:-unsloth/Qwen3-VL-4B-Instruct-GGUF}
      - LLAMA_ARG_HF_FILE=${MODEL_FILE:-Qwen3-VL-4B-Instruct-Q5_K_M.gguf}
      - LLAMA_ARG_MMPROJ_URL=${MMPROJ_URL:-https://huggingface.co/unsloth/Qwen3-VL-4B-Instruct-GGUF/resolve/main/mmproj-BF16.gguf}
      - HF_TOKEN=${HF_TOKEN}
      - LLAMA_ARG_JINJA=1
      - LLAMA_ARG_CHAT_TEMPLATE=chatml
      - LLAMA_ARG_HOST=0.0.0.0
      - LLAMA_ARG_PORT=8000
      - LLAMA_ARG_N_GPUS=1
      - LLAMA_ARG_GPU_LAYERS=99
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    networks:
      - openlpr-network
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s
    profiles:
      - nvidia-cuda