Dockerfile.worker.gpu

# VLog Remote Transcoding Worker with GPU Support (Rocky Linux 10)
# Containerized worker with NVIDIA NVENC and Intel VAAPI hardware encoding
#
# Build: docker build -f Dockerfile.worker.gpu -t vlog-worker-gpu .
# Run (NVIDIA): docker run --gpus all -e VLOG_WORKER_API_KEY=<key> vlog-worker-gpu
# Run (Intel):  docker run --device /dev/dri -e VLOG_WORKER_API_KEY=<key> vlog-worker-gpu
# Run (Both):   docker run --gpus all --device /dev/dri -e VLOG_WORKER_API_KEY=<key> vlog-worker-gpu
#
# Base: Rocky Linux 10 with RPM Fusion packages
# - FFmpeg 7.1.2 with nvenc, vaapi, qsv encoders (from RPM Fusion)
# - intel-media-driver 25.2.6 for Battlemage/Arc support (from RPM Fusion)
# - NVIDIA libraries injected at runtime via nvidia-container-toolkit

# Pin base image version for reproducible builds
# Update this periodically and test before deploying
ARG ROCKY_VERSION=10

# Code version for worker compatibility checking
# Set this during build: docker build --build-arg CODE_VERSION=$(git rev-parse --short HEAD) ...
ARG CODE_VERSION=dev
ARG BUILD_TIMESTAMP=""

# =============================================================================
# Stage 1: Builder - Install build dependencies and Python packages
# =============================================================================
FROM rockylinux/rockylinux:${ROCKY_VERSION} AS builder

# Install EPEL and RPM Fusion repositories
RUN dnf install -y epel-release && \
    dnf install -y --nogpgcheck \
        https://mirrors.rpmfusion.org/free/el/rpmfusion-free-release-10.noarch.rpm \
        https://mirrors.rpmfusion.org/nonfree/el/rpmfusion-nonfree-release-10.noarch.rpm && \
    dnf clean all

# Install Python 3.12 and build dependencies for packages with C extensions (argon2-cffi)
RUN dnf install -y \
    python3.12 \
    python3.12-pip \
    python3.12-devel \
    gcc \
    libffi-devel \
    && dnf clean all && \
    alternatives --install /usr/bin/python3 python3 /usr/bin/python3.12 1 && \
    alternatives --install /usr/bin/pip3 pip3 /usr/bin/pip3.12 1

WORKDIR /build

# Copy package files for dependency installation
COPY pyproject.toml ./

# Install Python dependencies to a specific location for copying
# Note: filelock and jaraco-context are security patches for transitive dependencies
RUN pip3 install --no-cache-dir --prefix=/install \
    fastapi>=0.100.0 \
    uvicorn>=0.23.0 \
    python-multipart>=0.0.6 \
    asyncpg>=0.29.0 \
    "databases[postgresql]>=0.8.0" \
    psycopg2-binary>=2.9.0 \
    sqlalchemy>=2.0.0 \
    alembic>=1.13.0 \
    python-slugify>=8.0.0 \
    aiofiles>=23.0.0 \
    httpx>=0.25.0 \
    watchdog>=3.0.0 \
    slowapi>=0.1.9 \
    limits>=3.0.0 \
    "filelock>=3.20.3" \
    "jaraco-context>=6.1.0"

# Copy source code and install the package
COPY config.py ./
COPY code_version.py ./
COPY api/ api/
COPY worker/ worker/
COPY cli/ cli/

# Create constraints file for security-patched transitive dependencies
RUN echo "filelock>=3.20.3" > /tmp/constraints.txt && \
    echo "jaraco-context>=6.1.0" >> /tmp/constraints.txt && \
    pip3 install --no-cache-dir --prefix=/install -c /tmp/constraints.txt .

# =============================================================================
# Stage 2: Runtime - Minimal production image
# =============================================================================
FROM rockylinux/rockylinux:${ROCKY_VERSION} AS runtime

# Install EPEL and RPM Fusion repositories
RUN dnf install -y epel-release && \
    dnf install -y --nogpgcheck \
        https://mirrors.rpmfusion.org/free/el/rpmfusion-free-release-10.noarch.rpm \
        https://mirrors.rpmfusion.org/nonfree/el/rpmfusion-nonfree-release-10.noarch.rpm && \
    dnf clean all

# Install FFmpeg with hardware acceleration support and curl for health checks
# RPM Fusion's FFmpeg 7.1.2 includes nvenc, vaapi, and qsv encoders
RUN dnf install -y \
    ffmpeg \
    ffmpeg-libs \
    curl \
    && dnf update -y \
    && dnf clean all

# Install Intel VAAPI driver for Arc/Battlemage support
# intel-media-driver 25.2.6 from RPM Fusion supports Battlemage (Arc B580)
# Note: libva and mesa-dri-drivers are pulled in as ffmpeg dependencies
RUN dnf install -y \
    intel-media-driver \
    && dnf clean all

# Install Python 3.12 runtime (no devel packages needed)
RUN dnf install -y \
    python3.12 \
    python3.12-pip \
    && dnf clean all && \
    alternatives --install /usr/bin/python3 python3 /usr/bin/python3.12 1 && \
    alternatives --install /usr/bin/pip3 pip3 /usr/bin/pip3.12 1

# Create non-root user for security
RUN useradd -m -u 1000 -s /bin/bash vlog && \
    usermod -aG video vlog

WORKDIR /app

# Copy installed Python packages from builder
COPY --from=builder /install /usr/local

# Ensure security-patched packages are at correct versions
# (dnf-installed Python packages may have brought in older transitive deps)
RUN pip3 install --no-cache-dir --upgrade \
    "filelock>=3.20.3" \
    "jaraco-context>=6.1.0" \
    "wheel>=0.46.2"

# Copy source code
COPY --chown=vlog:vlog config.py ./
COPY --chown=vlog:vlog code_version.py ./
COPY --chown=vlog:vlog api/ api/
COPY --chown=vlog:vlog worker/ worker/
COPY --chown=vlog:vlog cli/ cli/
COPY --chown=vlog:vlog pyproject.toml ./

# Re-declare build args for use in this stage (args don't persist across stages)
ARG CODE_VERSION=dev
ARG BUILD_TIMESTAMP=""

# Switch to non-root user
USER vlog

# Create work directory
RUN mkdir -p /tmp/vlog-worker

# Environment variables
# VLOG_CODE_VERSION and VLOG_BUILD_TIMESTAMP are used for worker compatibility checking
ENV VLOG_WORKER_API_URL=http://vlog-worker-api:9002 \
    VLOG_WORKER_WORK_DIR=/tmp/vlog-worker \
    VLOG_WORKER_HEARTBEAT_INTERVAL=30 \
    VLOG_WORKER_POLL_INTERVAL=10 \
    # Code version for compatibility checking
    VLOG_CODE_VERSION=${CODE_VERSION} \
    VLOG_BUILD_TIMESTAMP=${BUILD_TIMESTAMP} \
    # GPU detection (auto-detect by default)
    VLOG_HWACCEL_TYPE=auto \
    VLOG_HWACCEL_PREFERRED_CODEC=av1 \
    # NVIDIA runtime (nvidia-container-toolkit injects libraries)
    NVIDIA_VISIBLE_DEVICES=all \
    NVIDIA_DRIVER_CAPABILITIES=compute,video,utility \
    # Intel VAAPI - use renderD128 by default (can be overridden)
    LIBVA_DRIVER_NAME=iHD \
    # Python settings
    PYTHONUNBUFFERED=1

# Health check - verify worker's health server is responding
# The worker exposes /health on port 8080 which verifies:
# - Worker process is alive
# - API connection is working
# - FFmpeg is available (including GPU codecs)
HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
    CMD curl -f http://localhost:8080/health || exit 1

# Run the remote transcoder
CMD ["python3", "-m", "worker.remote_transcoder"]