[rocprof-compute] Merge CDash Nightly and Continuous workflow files (#2279)

* merged code-coverage and continuous workflow files * fixed runner typos and added build mode * add actor name to Continuous build * improve error handling and remove redundant verbose * fixed workflow file log output * revert logs output in run_ci.py * ruff format
2025-12-12 17:04:56 -05:00
@@ -1,157 +0,0 @@
-name: rocprofiler-compute Code Coverage
-run-name: rocprofiler-compute-code-coverage
-
-on:
-  schedule:
-    - cron: '0 2 * * *'
-  workflow_dispatch:
-    inputs:
-      runner_matrix:
-        description: 'Runner configuration'
-        required: false
-        type: string
-
-env:
-  ROCM_PATH: "/opt/rocm"
-  ROCM_VERSION: "7.0.2"
-
-jobs:
-  prepare_matrix:
-    runs-on: ubuntu-latest
-    outputs:
-      matrix: ${{ steps.generate_matrix.outputs.matrix }}
-
-    steps:
-      - name: Generate matrix
-        id: generate_matrix
-        run: |
-          MATRIX='[{"os-release":"22.04","gpu":"mi355","arch":"gfx950","runner":"linux-mi355-1gpu-ossci-rocm","code-name":"jammy"},{"os-release":"24.04","gpu":"mi355","arch":"gfx950","runner":"linux-mi355-1gpu-ossci-rocm","code-name":"noble"},{"os-release":"22.04","gpu":"mi325","arch":"gfx94X","runner":"linux-mi325-1gpu-ossci-rocm","code-name":"jammy"},{"os-release":"24.04","gpu":"mi325","arch":"gfx94X","runner":"linux-mi325-1gpu-ossci-rocm","code-name":"noble"}]'
-
-          if [ -n "${{ github.event.inputs.runner_matrix }}" ]; then
-            MATRIX='${{ github.event.inputs.runner_matrix }}'
-          fi
-
-          echo "matrix=${MATRIX}" >> $GITHUB_OUTPUT
-
-  coverage:
-    name: Coverage • ${{ matrix.system.gpu }} • Ubuntu ${{ matrix.system.os-release }}
-    needs: prepare_matrix
-    strategy:
-      fail-fast: false
-      matrix:
-        system: ${{ fromJSON(needs.prepare_matrix.outputs.matrix) }}
-    runs-on: ${{ matrix.system.runner }}
-    env:
-      HIP_PLATFORM: "amd"
-      OMPI_ALLOW_RUN_AS_ROOT: 1
-      OMPI_ALLOW_RUN_AS_ROOT_CONFIRM: 1
-    permissions:
-      packages: read
-    container:
-      image: ghcr.io/rocm/rocprofiler-ubuntu:${{ matrix.system.os-release }}-systems-ci-${{ matrix.system.arch }}
-      options:
-        --privileged
-        --ipc host
-        --group-add video
-        --device /dev/kfd
-        --device /dev/dri
-        --cap-add CAP_SYS_ADMIN
-
-    steps:
-      - uses: actions/checkout@v5
-        with:
-          sparse-checkout: projects/rocprofiler-compute/
-
-      - name: Update system packages
-        run: |
-          apt-get update
-          apt-get dist-upgrade -y
-          echo "✅ System packages updated!"
-
-      - name: Install amdgpu and dependencies
-        shell: bash
-        run: |
-          ROCM_MAJOR=$(echo ${{ env.ROCM_VERSION }} | sed 's/\./ /g' | awk '{print $1}')
-          ROCM_MINOR=$(echo ${{ env.ROCM_VERSION }} | sed 's/\./ /g' | awk '{print $2}')
-          ROCM_VERSN=$(( (${ROCM_MAJOR}*10000)+(${ROCM_MINOR}*100) ))
-          wget -N -P /tmp/ https://repo.radeon.com/amdgpu-install/${ROCM_MAJOR}.${ROCM_MINOR}/ubuntu/${{ matrix.system.code-name }}/amdgpu-install_${ROCM_MAJOR}.${ROCM_MINOR}.${ROCM_VERSN}-1_all.deb
-          apt-get install -y /tmp/amdgpu-install_${ROCM_MAJOR}.${ROCM_MINOR}.${ROCM_VERSN}-1_all.deb
-          apt-get update
-          apt install -y amd-smi-lib libdw-dev
-          echo "✅ amdgpu and dependencies Installed!"
-
-      - name: Install MPI
-        run: |
-          apt install -y libopenmpi-dev
-          echo "✅ MPI Installed!"
-
-      - name: Install ROCm
-        run: |
-          apt install -y rocm-dev
-          echo "✅ ROCm Installation Complete!"
-
-      - name: Install Python Dependencies
-        working-directory: projects/rocprofiler-compute
-        run: |
-          pip install -r requirements.txt --break-system-packages --ignore-installed
-          pip install -r requirements-test.txt --break-system-packages --ignore-installed
-          echo "✅ Python dependencies installed!"
-
-      - name: Configure, Build, Test, and Upload Coverage
-        id: run_coverage
-        timeout-minutes: 120
-        working-directory: projects/rocprofiler-compute
-        continue-on-error: true
-        run: |
-          set -e
-          git config --global --add safe.directory ${GITHUB_WORKSPACE}
-          git config --global --add safe.directory ${PWD}
-
-          BUILD_NAME="ROCm/rocprofiler-compute-${{ github.ref_name }}-ubuntu-${{ matrix.system.os-release }}-${{ matrix.system.gpu }}"
-          SITE_NAME="${{ matrix.system.runner }}"
-
-          PATH=${{ env.ROCM_PATH }}/bin:${{ env.ROCM_PATH }}/llvm/bin:$PATH \
-          LD_LIBRARY_PATH=${{ env.ROCM_PATH }}/lib:${{ env.ROCM_PATH }}/llvm/lib:$LD_LIBRARY_PATH \
-          python3 ./tools/run-code-coverage.py \
-            --build-name "${BUILD_NAME}" \
-            --site "${SITE_NAME}" \
-            --mode "Nightly" \
-            --build-jobs 16 \
-            --pytest-numprocs 8 \
-            --install
-
-      - name: Check Coverage Configuration
-        if: always()
-        working-directory: projects/rocprofiler-compute
-        run: |
-          echo "=== Checking Build Configuration ==="
-          if [ -f "build/CMakeCache.txt" ]; then
-            echo "CMakeCache.txt contents (relevant variables):"
-            grep -E "ENABLE_TESTS|ENABLE_COVERAGE|INSTALL_TESTS|PYTEST_NUMPROCS" build/CMakeCache.txt || echo "⚠️ Coverage variables not found!"
-            echo ""
-            echo "=== Checking for coverage.xml ==="
-            if [ -f "build/coverage.xml" ]; then
-              echo "✅ coverage.xml found!"
-              ls -lh build/coverage.xml
-            else
-              echo "❌ coverage.xml NOT found!"
-            fi
-            echo ""
-            echo "=== Checking CTest results ==="
-            if [ -f "build/Testing/Temporary/LastTest.log" ]; then
-              echo "Last 50 lines of test log:"
-              tail -n 50 build/Testing/Temporary/LastTest.log
-            fi
-          else
-            echo "❌ build/CMakeCache.txt not found - build may have failed!"
-          fi
-
-      - name: Upload Logs on Failure
-        if: failure()
-        uses: actions/upload-artifact@v4
-        with:
-          name: test-logs-${{ matrix.system.gpu }}-${{ matrix.system.os-release }}
-          path: |
-            projects/rocprofiler-compute/build/Testing/Temporary/LastTest*.log
-            projects/rocprofiler-compute/build/coverage.xml
-          retention-days: 7
@@ -5,6 +5,15 @@ on:
  schedule:
    - cron: '0 6 * * *'
  workflow_dispatch:
+    inputs:
+      mode:
+        description: 'Build mode'
+        required: false
+        default: 'continuous'
+        type: choice
+        options:
+          - continuous
+          - nightly
  push:
    branches: [ develop ]
    paths:
@@ -49,7 +58,7 @@ jobs:
        id: generate_matrix
        working-directory: projects/rocprofiler-compute/.github
        run: |
-          if [ '${{ github.event_name }}' = 'schedule' ]; then
+          if [ '${{ github.event_name }}' = 'schedule' ] || [ '${{ inputs.mode }}' = 'nightly' ]; then
            MATRIX_CONTENT=$(cat ci-matrix.yml | yq '.matrix-ubuntu-nightly' -I=0 -o=json)
          else
            MATRIX_CONTENT=$(cat ci-matrix.yml | yq '.matrix-ubuntu-ci' -I=0 -o=json)
@@ -88,17 +97,20 @@ jobs:
      - name: Setup Environment
        id: setup_env
        run: |
-          if [ '${{ github.event_name }}' = 'schedule' ]; then
+          if [ '${{ github.event_name }}' = 'schedule' ] || [ '${{ inputs.mode }}' = 'nightly' ]; then
            MODE=Nightly
            EXCLUDED_TESTS=""
+            ADD_COVERAGE="--coverage"
          else
            MODE=Continuous
            EXCLUDED_TESTS="test_profile_live_attach_detach"
+            ADD_COVERAGE=""
          fi
          echo "mode=${MODE}" >> $GITHUB_OUTPUT
          echo "excluded_tests=${EXCLUDED_TESTS}" >> $GITHUB_OUTPUT
+          echo "add_coverage=${ADD_COVERAGE}" >> $GITHUB_OUTPUT

-          if [ '${{ matrix.system.os-release }}' = '24.04']; then
+          if [ '${{ matrix.system.os-release }}' = '24.04' ]; then
            CODE_NAME=noble
          else
            CODE_NAME=jammy
@@ -129,13 +141,21 @@ jobs:
      - name: Install Python Requirements
        working-directory: projects/rocprofiler-compute
        run: |
-          pip install -r requirements.txt --break-system-packages --ignore-installed
-          pip install -r requirements-test.txt --break-system-packages --ignore-installed
+          for i in 1 2 3; do
+            pip install -r requirements.txt --break-system-packages --ignore-installed --timeout 60 && break
+            echo "⚠️ pip install attempt $i failed, retrying..."
+            sleep 10
+          done
+          for i in 1 2 3; do
+            pip install -r requirements-test.txt --break-system-packages --ignore-installed --timeout 60 && break
+            echo "⚠️ pip install attempt $i failed, retrying..."
+            sleep 10
+          done
          echo "✅ pip requirements installed!"
      
      - name: Configure, Build, and Test
        id: test
-        timeout-minutes: 90
+        timeout-minutes: ${{ steps.setup_env.outputs.mode == 'Nightly' && 120 || 90 }}
        working-directory: projects/rocprofiler-compute
        run: |
          set -e
@@ -145,9 +165,11 @@ jobs:
          LD_LIBRARY_PATH=${{ env.ROCM_PATH }}/lib:${{ env.ROCM_PATH }}/llvm/lib:${{ env.ROCM_PATH }}/lib/rocm_sysdeps/lib:$LD_LIBRARY_PATH \
          python3 ./tools/run-ci.py \
            --name "ROCm/rocprofiler-compute-${{ github.ref_name }}-ubuntu-${{ matrix.system.os-release }}-${{ matrix.system.gpu }}" \
+            --actor "${{ github.actor }}" \
            --site ${{ matrix.system.runner }} \
            --mode ${{ steps.setup_env.outputs.mode }} \
            --build-jobs 16 \
+            ${{ steps.setup_env.outputs.add_coverage }} \
            -- \
            -DCMAKE_BUILD_TYPE=Release \
            -DCMAKE_PREFIX_PATH=${{ env.ROCM_PATH }} \
@@ -158,8 +180,14 @@ jobs:
            -E "${{ steps.setup_env.outputs.excluded_tests }}"
      
      - name: Output Logs
-        if: failure() && steps.test.outcome == 'failure'
+        if: failure()
        working-directory: projects/rocprofiler-compute
        run: |
-          echo "❌ Run Failed: Now outputting LastTest.log files for detailed logs..."
-          cat build/Testing/Temporary/LastTest*.log
+          echo "❌ Run Failed: Outputting available log files..."
+          for log in build/Testing/Temporary/LastTest*.log build/Testing/Temporary/LastConfigure*.log; do
+            if [ -f "$log" ]; then
+              echo "=== $log ==="
+              cat "$log"
+            fi
+          done
+