Dockerfile.gpu

# 1. Use NVIDIA CUDA base image for GPU support
FROM nvidia/cuda:12.1.0-cudnn8-runtime-ubuntu22.04

# 2. Build arguments for region-specific configuration
ARG UV_INSTALL_URL=https://astral.sh/uv/install.sh
ARG UV_INDEX_URL=https://pypi.org/simple
ARG APT_GET_URL=
ARG PYTORCH_INDEX_URL=https://download.pytorch.org/whl/cu121

# 3. Set environment variables
ENV PYTHONUNBUFFERED=1 \
    DEBIAN_FRONTEND=noninteractive \
    NVIDIA_VISIBLE_DEVICES=all \
    NVIDIA_DRIVER_CAPABILITIES=compute,utility \
    UV_HTTP_TIMEOUT=300 \
    UV_INDEX_URL=${UV_INDEX_URL}

# 4. Install Python 3.11 and system dependencies
# Fix GPG signature issues when using mirror by using http instead of https
RUN MIRROR_URL="${APT_GET_URL:-http://archive.ubuntu.com}" && \
    # Force convert https:// to http:// to avoid GPG signature issues
    MIRROR_URL=$(echo "$MIRROR_URL" | sed 's|^https://|http://|') && \
    echo "Configuring apt mirror: $MIRROR_URL..." && \
    # Use http instead of https for mirrors to avoid GPG issues
    sed -i "s|http://archive.ubuntu.com|${MIRROR_URL}|g" /etc/apt/sources.list && \
    sed -i "s|https://archive.ubuntu.com|${MIRROR_URL}|g" /etc/apt/sources.list || true && \
    sed -i "s|http://security.ubuntu.com|${MIRROR_URL}|g" /etc/apt/sources.list && \
    sed -i "s|https://security.ubuntu.com|${MIRROR_URL}|g" /etc/apt/sources.list || true && \
    # Configure apt timeout and retry settings
    echo 'Acquire::http::Timeout "600";' > /etc/apt/apt.conf.d/99timeout && \
    echo 'Acquire::ftp::Timeout "600";' >> /etc/apt/apt.conf.d/99timeout && \
    echo 'Acquire::Retries "10";' >> /etc/apt/apt.conf.d/99timeout && \
    echo 'Acquire::http::MaxParallelDownloads "4";' >> /etc/apt/apt.conf.d/99timeout && \
    apt-get update && \
    apt-get install -y --no-install-recommends \
    software-properties-common \
    ca-certificates \
    gnupg2 \
    && echo "Checking if Python 3.11 is available in official Ubuntu repository..." && \
    (apt-cache show python3.11 2>/dev/null | grep -q "Version:" && \
     echo "Python 3.11 found in official repository, using it instead of deadsnakes PPA" || \
     (echo "Python 3.11 not in official repo, adding deadsnakes PPA..." && \
      add-apt-repository -y ppa:deadsnakes/ppa && \
      if [ -n "$APT_GET_URL" ]; then \
          echo "Replacing PPA with USTC Launchpad proxy for faster downloads in Mainland China..." && \
          find /etc/apt/sources.list.d/ -type f -name "*deadsnakes*.list" \
              -exec sed -i.bak -r 's#deb(-src)?\s*https?://ppa\.launchpadcontent\.net#deb\1 http://launchpad.proxy.ustclug.org#ig' {} \; && \
          find /etc/apt/sources.list.d/ -type f -name "*deadsnakes*.list" \
              -exec sed -i.bak -r 's#deb(-src)?\s*https?://ppa\.launchpad\.net#deb\1 http://launchpad.proxy.ustclug.org#ig' {} \; || true; \
      fi && \
      echo "Updating package lists (this may take a while for PPA)..." && \
      apt-get update)) && \
    echo "Installing Python 3.11 and dependencies..." && \
    apt-get install -y --no-install-recommends \
    python3.11 \
    python3.11-dev \
    python3.11-distutils \
    build-essential \
    curl \
    git \
    libpq-dev \
    tzdata \
    && apt-get clean && \
    rm -rf /var/lib/apt/lists/*

# 5. Set Python 3.11 as default
RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.11 1 && \
    update-alternatives --install /usr/bin/python python /usr/bin/python3.11 1

# 6. Install uv
RUN curl -LsSf ${UV_INSTALL_URL} | sh && \
    mv /root/.local/bin/uv /usr/local/bin/uv && \
    chmod +x /usr/local/bin/uv

# 7. Create and set working directory
WORKDIR /rag_arc

# 8. Copy only dependency files first (for better caching)
COPY pyproject.toml uv.lock /rag_arc/

# 9. Install Python dependencies using uv sync with virtual environment
# Use --no-cache and configure index URL based on region
# First install all dependencies except PyTorch
RUN uv sync --no-cache --index-url ${UV_INDEX_URL} && \
    # Install PyTorch with CUDA support from PyTorch index
    # This is an exception as PyTorch CUDA packages are typically not on PyPI
    # Use uv pip install only for PyTorch as it requires a specific index
    uv pip install --no-cache --index-url ${PYTORCH_INDEX_URL} \
        torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 || \
    uv pip install --no-cache --index-url ${UV_INDEX_URL} \
        torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1

# 10. Copy the rest of the project files (after dependencies are installed)
COPY . /rag_arc/

# 11. Reinstall the package in editable mode using uv sync
RUN uv sync --no-cache --index-url ${UV_INDEX_URL}

# 12. Create necessary directories
RUN mkdir -p \
    /rag_arc/data/parsed_files \
    /rag_arc/data/file_store \
    /rag_arc/data/chunk_store \
    /rag_arc/data/faiss_index \
    /rag_arc/data/unified_faiss_index \
    /rag_arc/data/unified_bm25_index \
    /rag_arc/data/graph_index \
    /rag_arc/data/graph_index_neo4j \
    /rag_arc/local/files \
    /rag_arc/models

# 13. Expose application port
EXPOSE 8000

# 14. Set PATH to include virtual environment
ENV PATH="/rag_arc/.venv/bin:$PATH"

# 15. Start the application
CMD ["uv", "run", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]