# DVAIA: Docker setup — Ollama (optional) + Qdrant + Flask app
#
# IMPORTANT: Run ./run_docker.sh (not plain "docker compose up") for an interactive
# setup that asks whether to use local Ollama models or cloud (Gemini / OpenAI),
# with requirements and .env guidance for each option.
#
# Quick reference:
#   ./run_docker.sh              Interactive prompt (recommended)
#   ./run_docker.sh --local        Local Ollama + model downloads (~9–10 GB disk)
#   ./run_docker.sh --gemini-only  Cloud Gemini (GOOGLE_API_KEY in .env)
#   ./run_docker.sh --openai-only  Cloud OpenAI (OPENAI_API_KEY in .env)
#   ./run_docker.sh --skip-prompt  Use GEMINI_ONLY / OPENAI_ONLY from .env; else local
#
# Direct compose (no prompt):
#   docker compose --profile ollama up --build   # local
#   docker compose up --build                    # cloud-only if GEMINI_ONLY/OPENAI_ONLY in .env
services:
  ollama:
    profiles:
      - ollama
    networks:
      - dvaia
    image: ollama/ollama:latest
    container_name: dvaia-ollama
    dns:
      - 8.8.8.8
      - 1.1.1.1
    volumes:
      - ollama_data:/root/.ollama
    ports:
      - "11480:11434"
    restart: "no"
    entrypoint: ["/bin/sh", "-c"]
    command:
      - |
        set -e
        rm -f /tmp/dvaia-ollama-ready
        /bin/ollama serve &
        sleep 5
        # Pin concrete tags (not floating :latest) so lab models stay stable.
        # Stage lines are parsed by run_docker.sh for progress UI.
        echo "DVAIA: pulling 1/4 llama3.2:3b (~2 GB)"
        ollama pull llama3.2:3b
        echo "DVAIA: done 1/4 llama3.2:3b"
        echo "DVAIA: pulling 2/4 nomic-embed-text (~275 MB)"
        ollama pull nomic-embed-text
        echo "DVAIA: done 2/4 nomic-embed-text"
        echo "DVAIA: pulling 3/4 qwen3:0.6b (~400 MB)"
        ollama pull qwen3:0.6b
        echo "DVAIA: done 3/4 qwen3:0.6b"
        echo "DVAIA: pulling 4/4 qwen2.5vl:7b (~6 GB — largest; can take a long time)"
        ollama pull qwen2.5vl:7b
        echo "DVAIA: done 4/4 qwen2.5vl:7b"
        touch /tmp/dvaia-ollama-ready
        echo "DVAIA: all Ollama models ready"
        wait
    healthcheck:
      test: ["CMD", "test", "-f", "/tmp/dvaia-ollama-ready"]
      interval: 10s
      timeout: 5s
      retries: 360
      start_period: 20s

  qdrant:
    image: qdrant/qdrant:latest
    container_name: dvaia-qdrant
    networks:
      - dvaia
    volumes:
      - qdrant_data:/qdrant/storage
    ports:
      - "6333:6333"
    restart: "no"

  dvaia:
    build: .
    container_name: dvaia-app
    networks:
      - dvaia
    volumes:
      - .:/app
    env_file:
      - .env
    environment:
      - PORT=${PORT:-5000}
      - GEMINI_ONLY=${GEMINI_ONLY:-false}
      - OPENAI_ONLY=${OPENAI_ONLY:-false}
      - RESET_DATA_ON_START=${RESET_DATA_ON_START:-false}
      # Always use Compose DNS. Do not pass through .env OLLAMA_HOST=localhost:11480
      # (that value is for host-native runs only and causes Connection refused in-container).
      - OLLAMA_HOST=http://ollama:11434
      - QDRANT_HOST=qdrant
      - QDRANT_PORT=6333
      - DEFAULT_MODEL=${DEFAULT_MODEL:-ollama:llama3.2:3b}
      - VISION_MODEL=${VISION_MODEL:-ollama:qwen2.5vl:7b}
      - AGENTIC_MODEL=${AGENTIC_MODEL:-qwen3:0.6b}
      - EMBEDDING_BACKEND=${EMBEDDING_BACKEND:-ollama}
      - WHISPER_MODEL=${WHISPER_MODEL:-base}
      - DATABASE_URI=${DATABASE_URI:-data/app.db}
      - UPLOAD_DIR=${UPLOAD_DIR:-data/uploads}
      - PAYLOADS_OUTPUT_DIR=${PAYLOADS_OUTPUT_DIR:-data/payloads/generate}
    ports:
      - "${PORT:-5000}:5000"
    depends_on:
      qdrant:
        condition: service_started
      # required: false so cloud-only (no ollama profile) still starts.
      ollama:
        condition: service_healthy
        required: false
    extra_hosts:
      - "host.docker.internal:host-gateway"
    restart: "no"

networks:
  dvaia:
    driver: bridge

volumes:
  ollama_data:
  qdrant_data: