docker-compose.dev.yaml

# Docker Compose configuration for Ollama, Open-WebUI, ComfyUI, VRAM Manager, Watchtower, and Grafana
# DEVELOPMENT VERSION - All services including VRAM Manager and Watchtower
#
# Prerequisites:
# - Docker Engine with GPU support (nvidia-docker2)
# - NVIDIA Container Toolkit installed
#
# Usage:
#   cp .env.dev .env  # If you want to use dev settings
#   docker compose -f docker-compose.dev.yaml up -d

networks:
  ai-stack:
    driver: bridge
    name: ai-stack-network

services:
  # -------------------------------------------------
  # Ollama – LLM inference engine
  # -------------------------------------------------
  ollama:
    image: ${OLLAMA_IMAGE:-ollama/ollama:latest}
    container_name: ${OLLAMA_CONTAINER_NAME:-ollama}
    restart: ${RESTART_POLICY:-unless-stopped}
    networks:
      - ai-stack
    ports:
      - "${OLLAMA_PORT:-11434}:11434"
    volumes:
      - aistack-ollama_data:/root/.ollama
    environment:
      - OLLAMA_HOST=${OLLAMA_HOST:-0.0.0.0}
      - OLLAMA_ORIGINS=${OLLAMA_ORIGINS:-*}
      - OLLAMA_MAX_VRAM=${OLLAMA_MAX_VRAM:-10737418240}
      - OLLAMA_MAX_LOADED_MODELS=${OLLAMA_MAX_LOADED_MODELS:-2}
      - OLLAMA_NUM_PARALLEL=${OLLAMA_NUM_PARALLEL:-2}
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    healthcheck:
      test: ollama --version || exit 1
      interval: 300s
      timeout: 10s
      retries: 3
      start_period: 30s
    stop_grace_period: 30s
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  # -------------------------------------------------
  # Open‑WebUI – Front‑end for Ollama
  # -------------------------------------------------
  open-webui:
    image: ${OPENWEBUI_IMAGE:-ghcr.io/open-webui/open-webui:main}
    container_name: ${OPENWEBUI_CONTAINER_NAME:-open-webui}
    restart: ${RESTART_POLICY:-unless-stopped}
    networks:
      - ai-stack
    ports:
      - "${OPENWEBUI_PORT:-3000}:8080"
    volumes:
      - aistack-open_webui_data:/app/backend/data
    environment:
      - OLLAMA_BASE_URL=${OLLAMA_BASE_URL:-http://ollama:11434}
      - WEBUI_SECRET_KEY=${WEBUI_SECRET_KEY:-change-this-secret-key}
      - WEBUI_AUTH=${WEBUI_AUTH:-true}
      - WEBUI_NAME=${WEBUI_NAME:-Open WebUI}
    depends_on:
      ollama:
        condition: service_healthy
    healthcheck:
      test: ["CMD-SHELL", "wget -q --spider http://localhost:8080/health || exit 1"]
      interval: 300s
      timeout: 10s
      retries: 3
      start_period: 45s
    stop_grace_period: 30s
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  # -------------------------------------------------
  # ComfyUI – Stable Diffusion Web UI
  # -------------------------------------------------
  comfyui:
    # TODO: Update to latest runtime tag from https://github.com/SaladTechnologies/comfyui-api/pkgs/container/comfyui-api
    image: ${COMFYUI_IMAGE:-ghcr.io/saladtechnologies/comfyui-api:comfy0.3.67-api1.13.3-torch2.8.0-cuda12.8-runtime}
    container_name: ${COMFYUI_CONTAINER_NAME:-comfyui}
    restart: ${RESTART_POLICY:-unless-stopped}
    networks:
      - ai-stack
    ports:
      - "${COMFYUI_PORT:-8188}:8188"
    volumes:
      - aistack-comfyui_data:/root/.comfyui
      - ./data/comfyui/models:/opt/ComfyUI/models
      - ./data/comfyui/workflows:/opt/ComfyUI/user/default/workflows
      - ./data/comfyui/custom_nodes:/opt/ComfyUI/custom_nodes
    environment:
      - COMFYUI_PORT=${COMFYUI_INTERNAL_PORT:-8188}
      - CLI_ARGS=${COMFYUI_CLI_ARGS:---normalvram}
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    healthcheck:
      test: ["CMD-SHELL", "wget -q --spider http://localhost:8188/system_stats || exit 1"]
      interval: 300s
      timeout: 10s
      retries: 3
      start_period: 60s
    stop_grace_period: 30s
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  # -------------------------------------------------
  # Watchtower – Automatic image updates (ENABLED FOR DEVELOPMENT)
  #
  # ⚠️  Security Note: Watchtower requires docker.sock access
  # 📝 Development Use Only: Not recommended for production
  # -------------------------------------------------
  watchtower:
    image: ${WATCHTOWER_IMAGE:-containrrr/watchtower:latest}
    container_name: ${WATCHTOWER_CONTAINER_NAME:-watchtower}
    restart: ${RESTART_POLICY:-unless-stopped}
    networks:
      - ai-stack
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:ro
    environment:
      - WATCHTOWER_CLEANUP=${WATCHTOWER_CLEANUP:-true}
      - WATCHTOWER_INCLUDE_STOPPED=${WATCHTOWER_INCLUDE_STOPPED:-false}
      - WATCHTOWER_SCHEDULE=${WATCHTOWER_SCHEDULE:-0 0 3 * * *}
      - WATCHTOWER_NOTIFICATIONS=${WATCHTOWER_NOTIFICATIONS:-}
      - WATCHTOWER_NOTIFICATION_URL=${WATCHTOWER_NOTIFICATION_URL:-}
      - WATCHTOWER_DEBUG=${WATCHTOWER_DEBUG:-false}
    command:
      - --schedule
      - "${WATCHTOWER_SCHEDULE:-0 0 3 * * *}"
      - --cleanup
    deploy:
      resources:
        limits:
          memory: 256M
          cpus: '0.5'
    logging:
      driver: json-file
      options:
        max-size: "5m"
        max-file: "2"

  # -------------------------------------------------
  # Grafana – Metrics Visualization
  # -------------------------------------------------
  grafana:
    image: ${GRAFANA_IMAGE:-grafana/grafana:latest}
    container_name: ${GRAFANA_CONTAINER_NAME:-grafana}
    restart: ${RESTART_POLICY:-unless-stopped}
    networks:
      - ai-stack
    ports:
      - "${GRAFANA_PORT:-3001}:3000"
    volumes:
      - aistack-grafana_data:/var/lib/grafana
    environment:
      - GF_SECURITY_ADMIN_USER=${GF_ADMIN_USER:-admin}
      - GF_SECURITY_ADMIN_PASSWORD=${GF_ADMIN_PASSWORD:-admin}
      - GF_INSTALL_PLUGINS=${GF_INSTALL_PLUGINS:-}
      - GF_SERVER_ROOT_URL=${GF_SERVER_ROOT_URL:-http://localhost:3001}
      - GF_ANALYTICS_REPORTING_ENABLED=${GF_ANALYTICS_ENABLED:-false}
      - GF_ANALYTICS_CHECK_FOR_UPDATES=${GF_ANALYTICS_UPDATES:-false}
    healthcheck:
      test: ["CMD-SHELL", "wget -q --spider http://localhost:3000/api/health || exit 1"]
      interval: 300s
      timeout: 10s
      retries: 3
      start_period: 30s
    stop_grace_period: 10s
    deploy:
      resources:
        limits:
          memory: 4G
          cpus: '2.0'
    logging:
      driver: json-file
      options:
        max-size: "10m"
        max-file: "3"

  # -------------------------------------------------
  # VRAM Manager – Automatic VRAM optimization
  # Monitors Ollama model loading and frees ComfyUI VRAM
  # to prevent Ollama from falling back to CPU mode.
  # -------------------------------------------------
  vram-manager:
    image: python:3.11-slim
    container_name: ${VRAM_MANAGER_CONTAINER_NAME:-aistack-vram-manager}
    restart: ${RESTART_POLICY:-unless-stopped}
    networks:
      - ai-stack
    volumes:
      - ./scripts:/app/scripts:ro
    working_dir: /app
    entrypoint: ["/bin/sh", "-c"]
    command:
      - |
        pip install --no-cache-dir requests &&
        python /app/scripts/vram_manager.py \
          --ollama-url http://aistack-ollama:11434 \
          --comfyui-url http://aistack-comfyui:${COMFYUI_INTERNAL_PORT:-8188} \
          --check-interval ${VRAM_CHECK_INTERVAL:-5} \
          --vram-threshold ${VRAM_THRESHOLD:-75}
    depends_on:
      ollama:
        condition: service_healthy
      comfyui:
        condition: service_healthy
    deploy:
      resources:
        limits:
          memory: 128M
          cpus: '0.25'
    logging:
      driver: json-file
      options:
        max-size: "5m"
        max-file: "2"

volumes:
  aistack-ollama_data:
    driver: local
    name: aistack-ollama_data
  aistack-open_webui_data:
    driver: local
    name: aistack-open_webui_data
  aistack-comfyui_data:
    driver: local
    name: aistack-comfyui_data
  aistack-grafana_data:
    driver: local
    name: aistack-grafana_data