ASV Benchmarks

.gitignore

@@ -18,3 +18,6 @@ benchmark_output
 
 # Generated Presto Config
 presto/docker/config/generated/
+velox/asv_benchmarks/results/
+velox/asv_benchmarks/.asv/
+**/.sccache-auth/

velox/asv_benchmarks/.dockerignore

@@ -0,0 +1,58 @@
+# ASV Benchmarks Docker Ignore File
+# Excludes unnecessary files from Docker build context
+
+# ASV generated files and directories
+.asv/
+*.html
+
+# Python cache and build artifacts
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+*.egg
+*.egg-info/
+dist/
+build/
+.eggs/
+
+# Virtual environments
+venv/
+env/
+ENV/
+
+# IDE and editor files
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+.DS_Store
+
+# Git
+.git/
+.gitignore
+
+# Documentation (already in image)
+*.md
+docs/
+
+# Test files (not needed for benchmark runs)
+tests/
+test_*.py
+*_test.py
+
+# Jupyter notebooks
+*.ipynb
+.ipynb_checkpoints/
+
+# Coverage reports
+htmlcov/
+.coverage
+.coverage.*
+coverage.xml
+*.cover
+
+# Logs
+*.log
+

velox/asv_benchmarks/Dockerfile

@@ -0,0 +1,108 @@
+# Dockerfile for Velox ASV Benchmarks with Cython Python Bindings
+# This dockerfile builds Cython-based Python bindings for TPC-H benchmarks and runs ASV
+#
+# Build from the velox-adapters-build image which already has Velox compiled
+# with CUDF support
+
+FROM velox-adapters-build:latest
+
+# Install Python development dependencies for Cython bindings
+RUN dnf install -y \
+    python3-devel \
+    python3-pip \
+    && dnf clean all
+
+# Upgrade pip and install Python packages
+RUN python3 -m pip install --upgrade pip setuptools wheel
+
+# Install Cython, ASV, and other Python dependencies
+RUN python3 -m pip install \
+    'Cython>=0.29' \
+    numpy \
+    asv \
+    && python3 -m pip cache purge
+
+# Get build configuration from environment
+ARG BUILD_BASE_DIR=/opt/velox-build
+ARG BUILD_TYPE=release
+ARG TPCH_DATA_PATH=/workspace/velox/velox-benchmark-data
+
+# Set environment variables for runtime
+ENV VELOX_ROOT=/workspace/velox \
+    VELOX_BUILD_DIR=${BUILD_BASE_DIR}/${BUILD_TYPE} \
+    BUILD_BASE_DIR=${BUILD_BASE_DIR} \
+    BUILD_TYPE=${BUILD_TYPE} \
+    TPCH_DATA_PATH=${TPCH_DATA_PATH} \
+    PYTHONUNBUFFERED=1
+
+# Update LD_LIBRARY_PATH to include Python bridge library location
+ENV LD_LIBRARY_PATH=${BUILD_BASE_DIR}/${BUILD_TYPE}/velox/experimental/cudf/benchmarks/python/cpp/tpch:${BUILD_BASE_DIR}/${BUILD_TYPE}/lib:${BUILD_BASE_DIR}/${BUILD_TYPE}/_deps/cudf-build:${BUILD_BASE_DIR}/${BUILD_TYPE}/_deps/rmm-build:${BUILD_BASE_DIR}/${BUILD_TYPE}/_deps/rapids_logger-build:${BUILD_BASE_DIR}/${BUILD_TYPE}/_deps/kvikio-build:${BUILD_BASE_DIR}/${BUILD_TYPE}/_deps/nvcomp_proprietary_binary-src/lib64:${LD_LIBRARY_PATH}
+
+# Set working directory
+WORKDIR /workspace/velox
+
+# Configure git to trust the repository (for ASV git operations)
+RUN git config --global --add safe.directory /workspace/velox
+
+# Build the Python bindings at image build time
+# This runs the CMake build for the C++ bridge and then builds the Cython extensions
+RUN --mount=type=bind,source=velox,target=/workspace/velox,ro \
+    <<EOF
+set -euxo pipefail
+
+# Copy Python bindings source to a writable location
+# We need this because build.sh builds "in-place" and the mounted source is read-only
+echo "Copying Python bindings source to writable location..."
+mkdir -p /tmp/python-bindings
+cp -r /workspace/velox/velox/experimental/cudf/benchmarks/python/* /tmp/python-bindings/
+
+# Build both C++ bridge and Cython bindings using the official build.sh script
+echo "Building C++ bridge and Cython bindings using build.sh..."
+cd /tmp/python-bindings
+
+# Set environment variables for build.sh
+export VELOX_ROOT=/workspace/velox
+export VELOX_BUILD_DIR=${VELOX_BUILD_DIR}
+
+# Run the official build script
+bash build.sh
+
+# Verify the C++ wrapper library was built
+WRAPPER_LIB="${VELOX_BUILD_DIR}/velox/experimental/cudf/benchmarks/python/cpp/tpch/libpython_benchmark_bridge_tpch.so"
+if [ ! -f "${WRAPPER_LIB}" ]; then
+    echo "ERROR: Failed to build C++ wrapper library: ${WRAPPER_LIB}"
+    exit 1
+fi
+
+# Verify .so files were created
+if ! ls cudf_tpch_benchmark*.so 1> /dev/null 2>&1; then
+    echo "ERROR: Failed to build Cython bindings - .so file not found"
+    exit 1
+fi
+
+echo "✓ C++ bridge and Cython bindings built successfully"
+
+# Install the Python package using pip
+# Use --no-build-isolation because extensions are already built by build.sh
+echo "Installing Python bindings package..."
+pip3 install --no-build-isolation --no-deps .
+
+# Verify installation
+python3 -c "import cudf_tpch_benchmark; print('✓ Package installed successfully')" || {
+    echo "ERROR: Failed to import installed package"
+    exit 1
+}
+
+echo "✓ Python bindings installed to system Python"
+
+EOF
+
+# Copy and setup entrypoint script
+COPY velox-testing/velox/asv_benchmarks/entrypoint.sh /entrypoint.sh
+RUN chmod +x /entrypoint.sh
+
+# Expose port for viewing results
+EXPOSE 8080
+
+# Set entrypoint
+ENTRYPOINT ["/entrypoint.sh"]