hvd-tests.yml

name: Run Horovod tests
on:
  push:
    branches:
      - master
      - "*.*.*"
    paths:
      - "ignite/**"
      - "tests/ignite/**"
      - "tests/run_cpu_tests.sh"
      - ".github/workflows/hvd-tests.yml"
  pull_request:
    paths:
      - "ignite/**"
      - "tests/ignite/**"
      - "tests/run_cpu_tests.sh"
      - ".github/workflows/hvd-tests.yml"
  workflow_dispatch:

concurrency:
  # <workflow_name>-<branch_name>-<true || commit_sha (if branch is protected)>
  group: hvd-tests-${{ github.ref_name }}-${{ !(github.ref_protected) || github.sha }}
  cancel-in-progress: true

jobs:
  horovod-tests:
    runs-on: ubuntu-latest
    timeout-minutes: 60
    strategy:
      matrix:
        python-version: ["3.11"]
        pytorch-channel: [pytorch]

    steps:
      - uses: actions/checkout@v4

      - name: Get year & week number
        id: get-date
        run: echo "date=$(/bin/date "+%Y-%U")" >> $GITHUB_OUTPUT
        shell: bash -l {0}

      - name: Get pip cache dir
        id: pip-cache
        run: |
          python3 -m pip install -U pip
          echo "pip_cache=$(python3 -m pip cache dir)" >> $GITHUB_OUTPUT
        shell: bash -l {0}

      - uses: actions/cache@v3
        with:
          path: |
            ~/conda_pkgs_dir
            ${{ steps.pip-cache.outputs.pip_cache }}
          key: ${{ steps.get-date.outputs.date }}-horovod-${{ hashFiles('requirements-dev.txt') }}

      - uses: actions/setup-python@v4
        with:
          python-version: ${{ matrix.python-version }}

      - name: Install dependencies
        shell: bash -l {0}
        run: |

          #install other dependencies
          pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
          pip install -r requirements-dev.txt
          
          # Install Horovod from source and apply a patch to build with recent pytorch
          # We can't use pip install <whatever> as build-env can't find pytorch and 
          # `--no-build-isolation` does not work with horovod setup.py
          git clone --recursive https://github.com/horovod/horovod.git /tmp/horovod
          cd /tmp/horovod
          sed -i "s/CMAKE_CXX_STANDARD 14/CMAKE_CXX_STANDARD 17/g" CMakeLists.txt
          sed -i "s/CMAKE_CXX_STANDARD 14/CMAKE_CXX_STANDARD 17/g" horovod/torch/CMakeLists.txt          
          HOROVOD_WITH_PYTORCH=1 python setup.py install                
          cd -
          # test the installation:
          python -c "import horovod.torch as hvd; hvd.mpi_ops.Sum"

          # Install ignite
          python setup.py install

      # Download MNIST: https://github.com/pytorch/ignite/issues/1737
      # to "/tmp" for cpu tests
      - name: Download MNIST
        uses: pytorch-ignite/download-mnist-github-action@master
        with:
          target_dir: /tmp

      - name: Run Tests
        uses: nick-fields/retry@v3
        with:
          max_attempts: 5
          timeout_minutes: 15
          shell: bash
          command: bash tests/run_cpu_tests.sh
          new_command_on_retry: USE_LAST_FAILED=1 bash tests/run_cpu_tests.sh

      - name: Upload coverage to Codecov
        uses: codecov/codecov-action@v3
        with:
          file: ./coverage.xml
          flags: hvd-cpu
          fail_ci_if_error: false