TheRock/rocm-systems/.github/workflows/therock-test-component.yml

name: Test component

on:
  workflow_call:
    inputs:
      artifact_run_id:
        type: string
        default: ""
      amdgpu_families:
        type: string
      test_runs_on:
        type: string
      platform:
        type: string
      component:
        type: string
      default_container_image:
        type: string
        default: "ghcr.io/rocm/no_rocm_image_ubuntu24_04@sha256:4150afe4759d14822f0e3f8930e1124f26e11f68b5c7b91ec9a02b20b1ebbb98"


permissions:
  contents: read

jobs:
  test_component:
    name: >-
      Test ${{ fromJSON(inputs.component).job_name }}
      (shard ${{ matrix.shard }} of ${{ fromJSON(inputs.component).total_shards }})
      ${{ fromJSON(inputs.component).expect_failure == true && '(xfail)' || '' }}
    runs-on: ${{ inputs.test_runs_on }}
    continue-on-error: ${{ fromJSON(inputs.component).expect_failure == true }}
    container:
      # If the component has specified an alternate image, honor that option.
      # Otherwise use the default image.
      image: >-
        ${{ inputs.platform == 'linux' &&
            (fromJSON(inputs.component).container_image || inputs.default_container_image)
            || null
        }}
      # Container options are configured by TheRock's
      # build_tools/github_actions/fetch_test_configurations.py.
      options: ${{ fromJSON(inputs.component).container_options }}
    strategy:
      fail-fast: false
      matrix:
        # The shard array is based on "total_shards" from "fetch_test_configurations.py"
        # The test executable will shard based on the array. (ex: [1, 2, 3, 4] = four test shards)
        shard: ${{ fromJSON(inputs.component).shard_arr }}
    defaults:
      run:
        shell: bash
    env:
      VENV_DIR: ${{ github.workspace }}/.venv
      ARTIFACT_RUN_ID: "${{ inputs.artifact_run_id != '' && inputs.artifact_run_id || github.run_id }}"
      OUTPUT_ARTIFACTS_DIR: "./build"
      THEROCK_BIN_DIR: "./build/bin"
      AMDGPU_FAMILIES: ${{ inputs.amdgpu_families }}
    steps:
      - name: "Fetch 'build_tools' from repository"
        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
        with:
          repository: "ROCm/TheRock"
          sparse-checkout: build_tools
          path: "prejob"

      - name: Pre-job cleanup processes on Windows
        if: ${{ runner.os == 'Windows' }}
        shell: powershell
        run: . '${{ github.workspace }}\prejob\build_tools\github_actions\cleanup_processes.ps1'

      - name: Pre-job cleanup GPU processes on Linux
        if: ${{ runner.os == 'Linux' }}
        timeout-minutes: 5
        run: bash ./prejob/build_tools/github_actions/cleanup_processes.sh

      - name: Checkout Repository
        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
        with:
          repository: "ROCm/TheRock"
          ref: release/therock-10.0

      - name: Run setup test environment workflow
        uses: './.github/actions/setup_test_environment'
        with:
          ARTIFACT_RUN_ID: ${{ env.ARTIFACT_RUN_ID }}
          ARTIFACT_GROUP: ${{ inputs.amdgpu_families }}
          OUTPUT_ARTIFACTS_DIR: ${{ env.OUTPUT_ARTIFACTS_DIR }}
          VENV_DIR: ${{ env.VENV_DIR }}
          FETCH_ARTIFACT_ARGS: ${{ fromJSON(inputs.component).fetch_artifact_args }}
          IS_PR_FROM_FORK: ${{ github.event.pull_request.head.repo.fork }}

      - name: Driver / GPU sanity check
        timeout-minutes: 3
        run: |
          python ./build_tools/print_driver_gpu_info.py

      - name: Setup Additional Requirements
        if: ${{ fromJSON(inputs.component).additional_requirements_files != '' }}
        run: |
          python ./build_tools/install_additional_requirements.py \
          --requirements-files=${{ join(fromJSON(inputs.component).additional_requirements_files, ',') }}

      - name: Set thread limits
        if: ${{ runner.os == 'Linux' }}
        run: |
          if [[ -n "${KUBE_CPU_REQUEST:-}" ]]; then
            CPU_COUNT="${KUBE_CPU_REQUEST%.*}"
            echo "OPENBLAS_NUM_THREADS=${CPU_COUNT}" >> $GITHUB_ENV
            echo "OMP_NUM_THREADS=${CPU_COUNT}" >> $GITHUB_ENV
            echo "Thread limits set: OPENBLAS_NUM_THREADS=${CPU_COUNT}, OMP_NUM_THREADS=${CPU_COUNT} (from KUBE_CPU_REQUEST=${KUBE_CPU_REQUEST})"
          else
            echo "KUBE_CPU_REQUEST not set, thread limits not configured"
          fi

      - name: Test
        timeout-minutes: ${{ fromJSON(inputs.component).timeout_minutes }}
        env:
          SHARD_INDEX: ${{ matrix.shard }}
          TOTAL_SHARDS: ${{ fromJSON(inputs.component).total_shards }}
          TEST_TYPE: ${{ fromJSON(inputs.component).test_type }}
          TEST_COMPONENT: ${{ fromJSON(inputs.component).job_name }}
          # Windows hip-tests (PAL=1, ROCR=0) set this via matrix; other
          # components leave it empty. See: https://github.com/ROCm/TheRock/issues/3587
          GPU_ENABLE_PAL: ${{ fromJSON(inputs.component).gpu_enable_pal }}
        run: |
          ${{ fromJSON(inputs.component).test_script }}

      # GitHub's 'Complete job' step is unaware of launched executables
      # and will fail to clean up orphan processes.
      - name: Post-job cleanup processes on Windows
        if: ${{ always() && runner.os == 'Windows' }}
        shell: powershell
        run: . '${{ github.workspace }}\build_tools\github_actions\cleanup_processes.ps1'

      - name: Post-job cleanup GPU processes on Linux
        if: ${{ always() && runner.os == 'Linux' }}
        timeout-minutes: 5
        run: bash ./build_tools/github_actions/cleanup_processes.sh