build(wheel): split libcuopt into per-component wheels

.github/workflows/build.yaml

@@ -241,8 +241,129 @@ jobs:
       package-name: libcuopt
       package-type: cpp
       publish-wheel-search-key: cuopt_wheel_cpp_libcuopt
+  wheel-build-libcuopt-client:
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    needs: [build-details]
+    secrets: inherit # zizmor: ignore[secrets-inherit]
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@release/26.10
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      build-datetime: ${{ needs.build-details.outputs.build-datetime }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      script: ci/build_wheel_libcuopt_client.sh
+      package-name: libcuopt_client
+      package-type: cpp
+      matrix_filter: group_by([.ARCH, (.CUDA_VER|split(".")|map(tonumber)|.[0])]) | map(max_by(.PY_VER|split(".")|map(tonumber)))
+  wheel-publish-libcuopt-client:
+    needs: wheel-build-libcuopt-client
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@release/26.10
+    secrets:
+      CONDA_RAPIDSAI_WHEELS_NIGHTLY_TOKEN: ${{ secrets.CONDA_RAPIDSAI_WHEELS_NIGHTLY_TOKEN }}
+      RAPIDSAI_PYPI_TOKEN: ${{ secrets.RAPIDSAI_PYPI_TOKEN }}
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: libcuopt_client
+      package-type: cpp
+      publish-wheel-search-key: cuopt_wheel_cpp_libcuopt_client
+  wheel-build-libcuopt-mathopt:
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    needs: [build-details]
+    secrets: inherit # zizmor: ignore[secrets-inherit]
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@release/26.10
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      build-datetime: ${{ needs.build-details.outputs.build-datetime }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      script: ci/build_wheel_libcuopt_mathopt.sh
+      package-name: libcuopt_mathopt
+      package-type: cpp
+      matrix_filter: group_by([.ARCH, (.CUDA_VER|split(".")|map(tonumber)|.[0])]) | map(max_by(.PY_VER|split(".")|map(tonumber)))
+  wheel-publish-libcuopt-mathopt:
+    needs: wheel-build-libcuopt-mathopt
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@release/26.10
+    secrets:
+      CONDA_RAPIDSAI_WHEELS_NIGHTLY_TOKEN: ${{ secrets.CONDA_RAPIDSAI_WHEELS_NIGHTLY_TOKEN }}
+      RAPIDSAI_PYPI_TOKEN: ${{ secrets.RAPIDSAI_PYPI_TOKEN }}
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: libcuopt_mathopt
+      package-type: cpp
+      publish-wheel-search-key: cuopt_wheel_cpp_libcuopt_mathopt
+  wheel-build-libcuopt-routing:
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    needs: [build-details]
+    secrets: inherit # zizmor: ignore[secrets-inherit]
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@release/26.10
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      build-datetime: ${{ needs.build-details.outputs.build-datetime }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      script: ci/build_wheel_libcuopt_routing.sh
+      package-name: libcuopt_routing
+      package-type: cpp
+      matrix_filter: group_by([.ARCH, (.CUDA_VER|split(".")|map(tonumber)|.[0])]) | map(max_by(.PY_VER|split(".")|map(tonumber)))
+  wheel-publish-libcuopt-routing:
+    needs: wheel-build-libcuopt-routing
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@release/26.10
+    secrets:
+      CONDA_RAPIDSAI_WHEELS_NIGHTLY_TOKEN: ${{ secrets.CONDA_RAPIDSAI_WHEELS_NIGHTLY_TOKEN }}
+      RAPIDSAI_PYPI_TOKEN: ${{ secrets.RAPIDSAI_PYPI_TOKEN }}
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: libcuopt_routing
+      package-type: cpp
+      publish-wheel-search-key: cuopt_wheel_cpp_libcuopt_routing
   wheel-build-cuopt:
-    needs: [build-details, wheel-build-libcuopt]
+    needs: [build-details, wheel-build-libcuopt, wheel-build-libcuopt-client,
+            wheel-build-libcuopt-mathopt, wheel-build-libcuopt-routing]
     permissions:
       actions: read
       contents: read

.github/workflows/pr.yaml

@@ -32,6 +32,9 @@ jobs:
       - docs-build
       - multi-gpu-cpp-tests
       - wheel-build-libcuopt
+      - wheel-build-libcuopt-client
+      - wheel-build-libcuopt-mathopt
+      - wheel-build-libcuopt-routing
       - wheel-build-cuopt
       - wheel-tests-cuopt
       - wheel-build-cuopt-server
@@ -608,8 +611,67 @@ jobs:
       build_type: pull-request
       build-datetime: ${{ needs.build-details.outputs.build-datetime }}
       script: ci/build_wheel_libcuopt.sh
+  wheel-build-libcuopt-client:
+    needs: [build-details, compute-matrix-filters, changed-files]
+    if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python_wheels
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    secrets: inherit # zizmor: ignore[secrets-inherit]
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@release/26.10
+    with:
+      # Same filter as libcuopt: every arch and CUDA version, latest Python only.
+      matrix_filter: ${{ needs.compute-matrix-filters.outputs.libcuopt_filter }}
+      package-type: cpp
+      package-name: libcuopt_client
+      build_type: pull-request
+      build-datetime: ${{ needs.build-details.outputs.build-datetime }}
+      script: ci/build_wheel_libcuopt_client.sh
+  wheel-build-libcuopt-mathopt:
+    needs: [build-details, compute-matrix-filters, changed-files]
+    if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python_wheels
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    secrets: inherit # zizmor: ignore[secrets-inherit]
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@release/26.10
+    with:
+      # Same filter as libcuopt: every arch and CUDA version, latest Python only.
+      matrix_filter: ${{ needs.compute-matrix-filters.outputs.libcuopt_filter }}
+      package-type: cpp
+      package-name: libcuopt_mathopt
+      build_type: pull-request
+      build-datetime: ${{ needs.build-details.outputs.build-datetime }}
+      script: ci/build_wheel_libcuopt_mathopt.sh
+  wheel-build-libcuopt-routing:
+    needs: [build-details, compute-matrix-filters, changed-files]
+    if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python_wheels
+    permissions:
+      actions: read
+      contents: read
+      id-token: write
+      packages: read
+      pull-requests: read
+    secrets: inherit # zizmor: ignore[secrets-inherit]
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@release/26.10
+    with:
+      # Same filter as libcuopt: every arch and CUDA version, latest Python only.
+      matrix_filter: ${{ needs.compute-matrix-filters.outputs.libcuopt_filter }}
+      package-type: cpp
+      package-name: libcuopt_routing
+      build_type: pull-request
+      build-datetime: ${{ needs.build-details.outputs.build-datetime }}
+      script: ci/build_wheel_libcuopt_routing.sh
   wheel-build-cuopt:
-    needs: [build-details, wheel-build-libcuopt, compute-matrix-filters, changed-files]
+    needs: [build-details, wheel-build-libcuopt, wheel-build-libcuopt-client,
+            wheel-build-libcuopt-mathopt, wheel-build-libcuopt-routing,
+            compute-matrix-filters, changed-files]
     if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python_wheels
     permissions:
       actions: read

ci/build_wheel_libcuopt.sh

@@ -103,6 +103,11 @@ EXCLUDE_ARGS=(
   --exclude "libnvJitLink.so*"
   --exclude "librapids_logger.so"
   --exclude "librmm.so"
+  # Provided by the sibling libcuopt-* wheels, so they must not be vendored here: doing so
+  # would ship several copies of each engine and defeat the split. RPATH resolves them.
+  --exclude "libcuopt_client.so"
+  --exclude "libcuopt_mathopt.so"
+  --exclude "libcuopt_routing.so"
   # OpenSSL 3 is intentionally NOT bundled. Resolving libssl.so.3 / libcrypto.so.3
   # at runtime via the host (or container image) keeps libcrypto and the FIPS
   # provider (system or mounted) byte-version-matched, which is required for

ci/build_wheel_libcuopt_client.sh

@@ -0,0 +1,133 @@
+#!/bin/bash
+# SPDX-FileCopyrightText: Copyright (c) 2023-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+set -euo pipefail
+
+# Same full C++ configure as every other cuOpt wheel; only the install components
+# staged into it differ, set by install.components in this package's pyproject.toml.
+# Kept in step with ci/build_wheel_libcuopt.sh -- regenerate all four together.
+
+source rapids-init-pip
+
+package_name="libcuopt_client"
+package_dir="python/libcuopt_client"
+
+# Install rockylinux repo
+if command -v dnf &> /dev/null; then
+    bash ci/utils/update_rockylinux_repo.sh
+fi
+
+# Install Boost and TBB
+bash ci/utils/install_boost_tbb.sh
+
+# Install libuuid (needed by cuopt_grpc_server)
+if command -v dnf &> /dev/null; then
+    dnf install -y libuuid-devel
+elif command -v apt-get &> /dev/null; then
+    apt-get update
+    apt-get install -y uuid-dev
+fi
+
+# Install Protobuf + gRPC (protoc + grpc_cpp_plugin)
+bash ci/utils/install_protobuf_grpc.sh
+
+# Compile against a modern GNU libgomp from conda-forge instead of bundled LLVM libomp, to
+# unify cuOpt's OpenMP runtime with the one cuDSS's threading layer needs (#1219).
+MODERN_LIBGOMP_DIR="$(pwd)/modern_libgomp"
+python -m pip install --quiet zstandard
+python ci/utils/install_modern_libgomp.py "${MODERN_LIBGOMP_DIR}"
+
+# Also build our own cuDSS threading layer against this same libgomp (see cpp/CMakeLists.txt,
+# cpp/src/barrier/cudss_mtlayer_cuopt.cpp), instead of cuDSS's prebuilt one, so both actually
+# share one instance, not just the same flavor. Conda keeps cuDSS's default (#1219 discussion).
+export SKBUILD_CMAKE_ARGS="-DOpenMP_gomp_LIBRARY:FILEPATH=${MODERN_LIBGOMP_DIR}/libgomp.so.1.0.0;-DCUOPT_BUILD_CUSTOM_CUDSS_MTLAYER=ON"
+
+# auditwheel repair does its own dependency resolution separately from the compiler; without
+# this it can't see our fetched copy and silently vendors the old Rocky 8 system one instead.
+export LD_LIBRARY_PATH="${MODERN_LIBGOMP_DIR}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}"
+
+# OpenSSL 3 hints for libcuopt's own find_package(OpenSSL).
+#
+# install_protobuf_grpc.sh links gRPC against OpenSSL 3 (see that script for
+# rationale). libcuopt then re-resolves OpenSSL via find_package because
+# gRPC's imported targets propagate it transitively. On Rocky/RHEL 8 the
+# EPEL openssl3-devel package installs in non-default paths, so we have to
+# point CMake at them; on Rocky/RHEL 9+ and Ubuntu 22.04+ the default
+# OpenSSL is already 3.x and no hints are needed.
+if [ -f /etc/os-release ]; then
+    . /etc/os-release
+    if [[ "$ID" == "rocky" || "$ID" == "centos" || "$ID" == "rhel" || "$ID" == "fedora" ]] && \
+       [[ "${VERSION_ID%%.*}" == "8" ]]; then
+        SKBUILD_CMAKE_ARGS="${SKBUILD_CMAKE_ARGS};-DOPENSSL_INCLUDE_DIR=/usr/include/openssl3;-DOPENSSL_SSL_LIBRARY=/usr/lib64/openssl3/libssl.so;-DOPENSSL_CRYPTO_LIBRARY=/usr/lib64/openssl3/libcrypto.so"
+    fi
+fi
+
+# For pull requests we are enabling assert mode.
+if [ "$RAPIDS_BUILD_TYPE" = "pull-request" ]; then
+    echo "Building in assert mode"
+    export SKBUILD_CMAKE_ARGS="${SKBUILD_CMAKE_ARGS};-DDEFINE_ASSERT=True"
+else
+    echo "Building in release mode"
+fi
+
+# Install cudss
+bash ci/utils/install_cudss.sh
+
+rapids-logger "Generating build requirements"
+
+rapids-dependency-file-generator \
+  --output requirements \
+  --file-key "py_build_${package_name}" \
+  --file-key "py_rapids_build_${package_name}" \
+  --matrix "cuda=${RAPIDS_CUDA_VERSION%.*};arch=$(arch);py=${RAPIDS_PY_VERSION};cuda_suffixed=true;use_cuda_wheels=true" \
+| tee /tmp/requirements-build.txt
+
+rapids-logger "Installing build requirements"
+rapids-pip-retry install \
+    -v \
+    --prefer-binary \
+    -r /tmp/requirements-build.txt
+
+# build with '--no-build-isolation', for better sccache hit rate
+# 0 really means "add --no-build-isolation" (ref: https://github.com/pypa/pip/issues/5735)
+export PIP_NO_BUILD_ISOLATION=0
+
+
+EXCLUDE_ARGS=(
+  --exclude "libraft.so"
+  --exclude "libcublas.so.*"
+  --exclude "libcublasLt.so.*"
+  --exclude "libcuda.so.1"
+  --exclude "libcudss.so.*"
+  --exclude "libcurand.so.*"
+  --exclude "libcusolver.so.*"
+  --exclude "libcusparse.so.*"
+  --exclude "libnccl.so.*"
+  --exclude "libnvJitLink.so*"
+  --exclude "librapids_logger.so"
+  --exclude "librmm.so"
+  # Provided by the sibling libcuopt-* wheels, so they must not be vendored here: doing so
+  # would ship several copies of each engine and defeat the split. RPATH resolves them.
+  --exclude "libcuopt_client.so"
+  --exclude "libcuopt_mathopt.so"
+  --exclude "libcuopt_routing.so"
+  # OpenSSL 3 is intentionally NOT bundled. Resolving libssl.so.3 / libcrypto.so.3
+  # at runtime via the host (or container image) keeps libcrypto and the FIPS
+  # provider (system or mounted) byte-version-matched, which is required for
+  # the FIPS provider's HMAC integrity check and avoids loading two libcrypto.so.3
+  # in the same process. Hosts must provide libssl.so.3 / libcrypto.so.3 (Ubuntu
+  # 22.04+, RHEL/Rocky 9+, manylinux_2_28+ with openssl3, Debian 12+).
+  --exclude "libssl.so.3"
+  --exclude "libcrypto.so.3"
+)
+
+ci/build_wheel.sh ${package_name} ${package_dir}
+
+mkdir -p final_dist
+python -m auditwheel repair "${EXCLUDE_ARGS[@]}" -w "${RAPIDS_WHEEL_BLD_OUTPUT_DIR}" ${package_dir}/dist/*
+
+ci/validate_wheel.sh ${package_dir} "${RAPIDS_WHEEL_BLD_OUTPUT_DIR}"
+
+RAPIDS_PACKAGE_NAME="$(rapids-artifact-name wheel_cpp libcuopt_client cuopt --cuda "$RAPIDS_CUDA_VERSION")"
+export RAPIDS_PACKAGE_NAME