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

name: Test component

on:
  workflow_call:
    inputs:
      artifact_run_id:
        type: string
        default: ""
      therock_ref:
        type: string
        required: true
      amdgpu_families:
        type: string
      test_runs_on:
        type: string
      platform:
        type: string
      component:
        type: string


permissions:
  contents: read

jobs:
  test_component:
    name: 'Test ${{ fromJSON(inputs.component).job_name }} (shard ${{ matrix.shard }} of ${{ fromJSON(inputs.component).total_shards }})'
    runs-on: ${{ inputs.test_runs_on }}
    timeout-minutes: 210
    container:
      image: ${{ inputs.platform == 'linux' && 'ghcr.io/rocm/no_rocm_image_ubuntu24_04@sha256:4150afe4759d14822f0e3f8930e1124f26e11f68b5c7b91ec9a02b20b1ebbb98' || null }}
      options: --ipc host
        --group-add video
        --device /dev/kfd
        --device /dev/dri
        --group-add 993
        --group-add 992
        --group-add 110
        --env-file /etc/podinfo/gha-gpu-isolation-settings
        -e ROCR_VISIBLE_DEVICES
        -e KUBE_CPU_REQUEST
    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 }}
      ARTIFACT_GROUP: ${{ inputs.amdgpu_families }}
      TEST_COMPONENT: ${{ fromJSON(inputs.component).job_name }}
    steps:
      - name: "Fetch 'build_tools' from repository"
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          repository: "ROCm/TheRock"
          sparse-checkout: build_tools
          path: "prejob"

      - name: Pre-job cleanup processes on Windows
        if: ${{ runner.os == 'Windows' }}
        timeout-minutes: 5
        shell: powershell
        run: . '${{ 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@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          repository: "ROCm/TheRock"
          ref: ${{ inputs.therock_ref }}

      - name: Configure git for long paths on Windows
        if: ${{ runner.os == 'Windows' }}
        run: git config --global core.longpaths true

      - name: Checkout rocm-libraries repository for scripts
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
        with:
          path: rocm-libraries

      - name: Run setup test environment workflow
        timeout-minutes: 15
        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

      # Set thread limits from KUBE_CPU_REQUEST (Kubernetes-injected) to avoid oversubscription.
      # Must be done at runtime since container env vars aren't in GitHub's env context.
      - name: Set thread limits
        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 }}
        run: |
          # Create logs directory
          mkdir -p ./test_logs

          # Run test and capture output (still display to console)
          python ./build_tools/memory_monitor.py --phase "Test ${{ fromJSON(inputs.component).job_name }}" -- \
            bash -c '${{ fromJSON(inputs.component).test_script }}' 2>&1 | tee ./test_logs/test_output.log

      - name: Notify Teams Channel on Failure
        if: ${{ failure() && !github.event.pull_request.head.repo.fork }}
        env:
          PR_NUMBER: ${{ github.event.pull_request.number }}
          PR_TITLE: ${{ github.event.pull_request.title }}
          MIOPEN_WEBHOOK_URL: ${{ secrets.MIOPEN_CI_WEBHOOK_URL }}
          HIPDNN_WEBHOOK_URL: ${{ secrets.HIPDNN_CI_WEBHOOK_URL }}
          COMPONENT_NAME: ${{ fromJSON(inputs.component).job_name }}
          JOB_NAME: "${{ fromJSON(inputs.component).job_name }} (shard ${{ matrix.shard }}/${{ fromJSON(inputs.component).total_shards }})"
        run: |
          python3 rocm-libraries/.github/scripts/notify_teams.py \
            --failure-stage test \
            --log-path ./test_logs/test_output.log \
            --webhook-urls "{\"miopen\":\"${MIOPEN_WEBHOOK_URL}\",\"hipdnn\":\"${HIPDNN_WEBHOOK_URL}\"}" \
            --pr-number "${PR_NUMBER}" \
            --pr-title "${PR_TITLE}" \
            --component-name "${COMPONENT_NAME}" \
            --job-name "${JOB_NAME}"

      # 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: ${{ always() && runner.os == 'Windows' }}
        timeout-minutes: 5
        shell: powershell
        run: . '${{ 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