TheRock/rocm-systems/.github/workflows/rdc-ci.yml

# RDC keeps this standalone workflow temporarily to validate project-specific
# packaging and test coverage while integration with TheRock CI is completed.
name: RDC CI

on:
  pull_request:
    branches: [ 'develop', 'amd-staging', 'amd-mainline' ]
    paths:
      - 'projects/rdc/**'
      - '.github/workflows/rdc-ci.yml'
  workflow_dispatch:

permissions:
  contents: read

concurrency:
  group: ${{ github.workflow }}-${{ github.event.number || github.sha }}
  cancel-in-progress: true

env:
  DEBIAN_FRONTEND: noninteractive
  DEBCONF_NONINTERACTIVE_SEEN: true
  BUILD_TYPE: RelWithDebInfo
  ROCM_DIR: /opt/rocm
  GRPC_ROOT: /opt/grpc
  GRPC_VERSION: "1.67.1"
  GPU_TARGETS: ${{ vars.GPU_TARGETS || 'gfx942' }}

defaults:
  run:
    working-directory: projects/rdc

jobs:
  # Resolve the current GPU runner label from ROCm/therock-ci-config instead of
  # hard-coding one (labels get decommissioned). Reuses the shared resolver action.
  ci_config:
    if: ${{ vars.RDC_GPU_TESTS_ENABLED == 'true' && (github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork) }}
    name: Resolve GPU runner label
    runs-on: ubuntu-latest
    permissions:
      contents: read
    outputs:
      gfx94x_runner: ${{ steps.runner.outputs.gfx94x_runner }}
    steps:
    - name: Checkout resolve-ci-config action
      uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
      with:
        sparse-checkout: .github/actions/rocprofiler-sdk-util/resolve-ci-config
        set-safe-directory: true

    - name: Resolve runner labels
      id: runner
      uses: ./.github/actions/rocprofiler-sdk-util/resolve-ci-config

  rdc_build:
    if: ${{ (github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork) }}
    runs-on: lstt_ci_runners
    timeout-minutes: 90
    container: rocm/dev-ubuntu-22.04:7.2
    outputs:
      TODAY: ${{ steps.metadata.outputs.TODAY }}

    steps:
    - uses: actions/checkout@v4
      with:
        sparse-checkout: |
          projects/amdsmi
          projects/rdc

    - name: Build metadata
      id: metadata
      working-directory: .
      run: |
        TODAY="$(date +%Y_%m_%d)"
        echo "TODAY=$TODAY" >> "$GITHUB_ENV"
        echo "TODAY=$TODAY" >> "$GITHUB_OUTPUT"

    - name: Install build tools
      working-directory: .
      run: |
        apt update -y
        apt install -y \
          autoconf \
          automake \
          build-essential \
          ccache \
          clang \
          cmake \
          curl \
          git \
          gzip \
          hip-dev \
          hip-runtime-amd \
          hipcc \
          hsa-rocr-dev \
          jq \
          libcap-dev \
          libc++-dev \
          libdrm-dev \
          libgflags-dev \
          libgtest-dev \
          libtool \
          libmnl-dev \
          libnl-3-dev \
          libnl-genl-3-dev \
          pkg-config \
          rocm-core \
          rocm-developer-tools \
          rocm-device-libs \
          rocm-smi-lib \
          rocprofiler-dev \
          rocprofiler-plugins \
          rocprofiler-register \
          rocprofiler-sdk \
          tar \
          unzip \
          wget \
          zip \
          zstd

    - name: Fetch gRPC source
      working-directory: .
      shell: bash
      run: |
        set -euo pipefail
        git clone --branch "v${GRPC_VERSION}" --depth 1 \
          https://github.com/grpc/grpc.git /tmp/grpc
        git -C /tmp/grpc submodule update --init --depth 1 \
          third_party/abseil-cpp \
          third_party/boringssl-with-bazel \
          third_party/cares/cares \
          third_party/protobuf \
          third_party/re2 \
          third_party/zlib

    - name: Build and install gRPC
      working-directory: .
      shell: bash
      run: |
        set -euo pipefail
        cmake -S /tmp/grpc -B /tmp/grpc/build \
          -DgRPC_INSTALL=ON \
          -DgRPC_BUILD_TESTS=OFF \
          -DBUILD_SHARED_LIBS=ON \
          -DCMAKE_INSTALL_PREFIX="${GRPC_ROOT}" \
          -DCMAKE_INSTALL_LIBDIR=lib \
          -DCMAKE_BUILD_TYPE=Release
        cmake --build /tmp/grpc/build --parallel "$(nproc)"
        cmake --install /tmp/grpc/build
        echo "${GRPC_ROOT}/lib" > /etc/ld.so.conf.d/grpc.conf
        ldconfig

    - name: Cache .ccache
      uses: actions/cache@v4
      with:
        path: ~/.cache/ccache
        key: ${{ runner.os }}-rdc-ccache-${{ github.ref_name }}-${{ env.TODAY }}
        restore-keys: |
          ${{ runner.os }}-rdc-ccache-${{ github.ref_name }}-
          ${{ runner.os }}-rdc-ccache-

    - name: Build and install amd-smi
      working-directory: projects/amdsmi
      run: |
        # RDC's metric fetcher links amdsmi_get_cpu_* (ESMI/CPU), so amd-smi
        # must be built with the E-SMI library enabled or librdc fails to link.
        cmake -B build \
          -DCMAKE_C_COMPILER_LAUNCHER=ccache \
          -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
          -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
          -DCMAKE_INSTALL_PREFIX=${ROCM_DIR} \
          -DCPACK_GENERATOR=DEB \
          -DENABLE_ESMI_LIB=ON
        make -C build -j $(nproc)
        make -C build install
        make -C build -j $(nproc) package

    - name: Build RDC
      run: |
        pwd
        cmake \
            -B build \
            -DCMAKE_C_COMPILER_LAUNCHER=ccache \
            -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
            -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
            -DGRPC_DESIRED_VERSION="${GRPC_VERSION}" \
            -DGRPC_ROOT="${GRPC_ROOT}" \
            -DBUILD_RUNTIME=ON \
            -DGPU_TARGETS="${GPU_TARGETS}" \
            -DBUILD_PROFILER=ON \
            -DBUILD_RVS=OFF \
            -DBUILD_TESTS=ON \
            -DCPACK_GENERATOR="DEB" \
            -DCMAKE_INSTALL_PREFIX=${ROCM_DIR}
        make -C build -j $(nproc)
        make -C build -j $(nproc) package

    - name: Install RDC
      run: |
        make -C build -j $(nproc) install

    - name: Upload RDC packages
      uses: actions/upload-artifact@v4
      with:
        name: rdc-packages
        path: |
          projects/amdsmi/build/amd-smi-lib*.deb
          projects/rdc/build/rdc*.deb
        if-no-files-found: error
        retention-days: 5

  # =================================================================
  #  Test job — non-GPU tests (installation, CLI, Python, parsers)
  # =================================================================
  rdc_test:
    if: ${{ (github.event_name != 'pull_request' || !github.event.pull_request.head.repo.fork) }}
    needs: rdc_build
    runs-on: lstt_ci_runners
    timeout-minutes: 30
    container: rocm/dev-ubuntu-22.04:7.2

    steps:
    - uses: actions/checkout@v4
      with:
        sparse-checkout: projects/rdc

    - name: Install test dependencies
      working-directory: .
      run: |
        apt update -y
        # rdctst pulls in libamd_smi's netlink (NIC) code, so the runtime
        # libnl stack must be present or the binary fails to load.
        apt install -y python3 python3-pip libnl-3-200 libnl-genl-3-200 libmnl0
        pip3 install pytest prometheus_client

    - name: Download RDC packages
      uses: actions/download-artifact@v4
      with:
        name: rdc-packages
        path: /opt/

    - name: Install RDC packages
      shell: bash
      working-directory: .
      run: |
        mapfile -t AMD_SMI_DEBS < <(find /opt/ -iname 'amd-smi-lib*.deb' ! -iname '*tests*' | sort)
        mapfile -t RDC_DEBS < <(find /opt/ -iname 'rdc*.deb' | sort)
        test "${#AMD_SMI_DEBS[@]}" -eq '1'
        test "${#RDC_DEBS[@]}" -eq '2'
        apt-get install -y "${AMD_SMI_DEBS[@]}" "${RDC_DEBS[@]}"

    - name: Configure RDC runtime library path
      shell: bash
      working-directory: .
      run: |
        echo "LD_LIBRARY_PATH=${ROCM_DIR}/lib:${ROCM_DIR}/lib/rdc:${ROCM_DIR}/lib/rdc/grpc/lib:${LD_LIBRARY_PATH:-}" >> "$GITHUB_ENV"

    - name: Create test results directory
      run: mkdir -p build

    - name: Verify RDC installation
      shell: bash
      working-directory: .
      run: |
        # confirm binaries are installed
        find $ROCM_DIR/bin -maxdepth 1 -iname rdcd
        find $ROCM_DIR/bin -maxdepth 1 -iname rdci
        find $ROCM_DIR/share/rdc -iname rdctst
        # confirm that libraries are installed
        MISSING_LIBS=()
        for lib in librdc.so librdc_bootstrap.so librdc_client.so; do
          test -e "$ROCM_DIR/lib/$lib" || MISSING_LIBS+=("$lib")
        done
        for lib in librdc_rocr.so librdc_rocp.so; do
          test -e "$ROCM_DIR/lib/rdc/$lib" || MISSING_LIBS+=("$lib")
        done
        if test "${#MISSING_LIBS[@]}" != "0"; then
          echo "Missing libs found!"
          for lib in "${MISSING_LIBS[@]}"; do
            echo "- $lib"
          done
          exit 1
        else
          echo "No missing libs found!"
        fi

    - name: Run RDC test suite (no GPU)
      shell: bash
      run: |
        python3 -m pytest tests/ci \
          -v --tb=short \
          --junitxml=build/test-results-nogpu.xml
      env:
        RDC_SRC_DIR: ${{ github.workspace }}/projects/rdc/python_binding

    - name: Publish test summary
      if: always()
      run: python3 tests/ci/junit_summary.py >> "$GITHUB_STEP_SUMMARY"

    - name: Upload test results
      if: always()
      uses: actions/upload-artifact@v4
      with:
        name: rdc-test-results
        path: projects/rdc/build/test-results-*.xml
        retention-days: 30

  # =================================================================
  #  GPU test job — requires AMD GPU hardware
  #  Enable by setting repo variable: RDC_GPU_TESTS_ENABLED=true
  # =================================================================
  rdc_test_gpu:
    needs: [rdc_build, ci_config]
    runs-on: ${{ needs.ci_config.outputs.gfx94x_runner }}
    timeout-minutes: 90
    if: ${{ vars.RDC_GPU_TESTS_ENABLED == 'true' }}
    container:
      image: rocm/dev-ubuntu-22.04:7.2
      options: --device=/dev/kfd --device=/dev/dri --group-add video

    steps:
    - uses: actions/checkout@v4
      with:
        sparse-checkout: projects/rdc

    - name: Set up environment
      working-directory: .
      run: |
        apt update -y
        # rdctst pulls in libamd_smi's netlink (NIC) code, so the runtime
        # libnl stack must be present or the binary fails to load.
        apt install -y python3 python3-pip libnl-3-200 libnl-genl-3-200 libmnl0
        pip3 install pytest prometheus_client

    - name: Download RDC packages
      uses: actions/download-artifact@v4
      with:
        name: rdc-packages
        path: /opt/

    - name: Install RDC packages
      shell: bash
      working-directory: .
      run: |
        mapfile -t AMD_SMI_DEBS < <(find /opt/ -iname 'amd-smi-lib*.deb' ! -iname '*tests*' | sort)
        mapfile -t RDC_DEBS < <(find /opt/ -iname 'rdc*.deb' | sort)
        test "${#AMD_SMI_DEBS[@]}" -eq '1'
        test "${#RDC_DEBS[@]}" -eq '2'
        apt-get install -y "${AMD_SMI_DEBS[@]}" "${RDC_DEBS[@]}"

    - name: Configure RDC runtime library path
      shell: bash
      working-directory: .
      run: |
        echo "LD_LIBRARY_PATH=${ROCM_DIR}/lib:${ROCM_DIR}/lib/rdc:${ROCM_DIR}/lib/rdc/grpc/lib:${LD_LIBRARY_PATH:-}" >> "$GITHUB_ENV"

    - name: Create test results directory
      run: mkdir -p build

    - name: Run RDC test suite (with GPU)
      shell: bash
      run: |
        python3 -m pytest tests/ci \
          -v --tb=short \
          --junitxml=build/test-results-gpu.xml
      env:
        RDC_SRC_DIR: ${{ github.workspace }}/projects/rdc/python_binding

    - name: Publish test summary
      if: always()
      run: python3 tests/ci/junit_summary.py >> "$GITHUB_STEP_SUMMARY"

    - name: Upload GPU test results
      if: always()
      uses: actions/upload-artifact@v4
      with:
        name: rdc-test-results-gpu
        path: projects/rdc/build/test-results-*.xml
        retention-days: 30