Xccl process group for Pytorch #15
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Lint | |
| on: | |
| pull_request: | |
| branches-ignore: | |
| - nightly | |
| push: | |
| branches: | |
| - main | |
| - release/* | |
| - landchecks/* | |
| workflow_dispatch: | |
| permissions: read-all | |
| # The names of steps that actually test the code should be suffixed with `(nonretryable)`. | |
| # When any other step fails, it's job will be retried once by retryBot. | |
| jobs: | |
| get-label-type: | |
| name: get-label-type | |
| uses: ./.github/workflows/_runner-determinator.yml | |
| with: | |
| triggering_actor: ${{ github.triggering_actor }} | |
| issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }} | |
| curr_branch: ${{ github.head_ref || github.ref_name }} | |
| lintrunner-clang: | |
| uses: pytorch/test-infra/.github/workflows/linux_job.yml@main | |
| needs: get-label-type | |
| with: | |
| timeout: 120 | |
| runner: "${{ needs.get-label-type.outputs.label-type }}amz2023.linux.2xlarge" | |
| docker-image: pytorch-linux-jammy-cuda11.8-cudnn9-py3.9-linter | |
| # NB: A shallow checkout won't work here because calculate-docker-image requires a full checkout | |
| # to run git rev-parse HEAD~:.ci/docker when a new image is needed | |
| fetch-depth: 0 | |
| submodules: true | |
| ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} | |
| script: | | |
| export ADDITIONAL_LINTRUNNER_ARGS="--take CLANGTIDY,CLANGFORMAT" | |
| export CLANG=1 | |
| .github/scripts/lintrunner.sh | |
| lintrunner-noclang: | |
| uses: pytorch/test-infra/.github/workflows/linux_job.yml@main | |
| needs: get-label-type | |
| with: | |
| timeout: 120 | |
| runner: "${{ needs.get-label-type.outputs.label-type }}amz2023.linux.2xlarge" | |
| docker-image: pytorch-linux-jammy-cuda11.8-cudnn9-py3.9-linter | |
| # NB: A shallow checkout won't work here because calculate-docker-image requires a full checkout | |
| # to run git rev-parse HEAD~:.ci/docker when a new image is needed | |
| fetch-depth: 0 | |
| submodules: true | |
| ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} | |
| script: | | |
| export ADDITIONAL_LINTRUNNER_ARGS="--skip CLANGTIDY,CLANGFORMAT" | |
| .github/scripts/lintrunner.sh | |
| quick-checks: | |
| uses: pytorch/test-infra/.github/workflows/linux_job.yml@main | |
| needs: get-label-type | |
| with: | |
| runner: "${{ needs.get-label-type.outputs.label-type }}amz2023.linux.2xlarge" | |
| docker-image: pytorch-linux-focal-linter | |
| fetch-depth: 0 | |
| ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} | |
| script: | | |
| # The generic Linux job chooses to use base env, not the one setup by the image | |
| CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]") | |
| conda activate "${CONDA_ENV}" | |
| # Ensure no non-breaking spaces | |
| # NB: We use 'printf' below rather than '\u000a' since bash pre-4.2 | |
| # does not support the '\u000a' syntax (which is relevant for local linters) | |
| (! git --no-pager grep -In "$(printf '\xC2\xA0')" -- . || (echo "The above lines have non-breaking spaces (U+00A0); please convert them to spaces (U+0020)"; false)) | |
| # Ensure cross-OS compatible file names | |
| (! git ls-files | grep -E '([<>:"|?*]|[ .]$)' || (echo "The above file names are not valid across all operating systems. Please ensure they don't contain the characters '<>:""|?*' and don't end with a white space or a '.' "; false)) | |
| # Ensure no versionless Python shebangs | |
| (! git --no-pager grep -In '#!.*python$' -- . || (echo "The above lines have versionless Python shebangs; please specify either python2 or python3"; false)) | |
| # Ensure ciflow tags mentioned in config | |
| python3 .github/scripts/collect_ciflow_labels.py --validate-tags | |
| # C++ docs check | |
| pushd docs/cpp/source | |
| ./check-doxygen.sh | |
| popd | |
| # CUDA kernel launch check | |
| set -eux | |
| python3 torch/testing/_internal/check_kernel_launches.py |& tee cuda_kernel_launch_checks.txt | |
| pr-sanity-checks: | |
| name: pr-sanity-checks | |
| needs: get-label-type | |
| runs-on: [self-hosted, "${{ needs.get-label-type.outputs.label-type }}amz2023.linux.large"] | |
| # Only run this on pull requests. This check is simple enough to be done without a Docker image | |
| if: github.event_name == 'pull_request' && !contains(github.event.pull_request.labels.*.name, 'skip-pr-sanity-checks') | |
| steps: | |
| - name: Checkout PyTorch | |
| uses: pytorch/pytorch/.github/actions/checkout-pytorch@main | |
| with: | |
| submodules: false | |
| fetch-depth: -1 | |
| - name: PR size check (nonretryable) | |
| env: | |
| BASE: ${{ github.event.pull_request.base.sha }} | |
| HEAD: ${{ github.event.pull_request.head.sha }} | |
| run: | | |
| bash .github/scripts/pr-sanity-check.sh | |
| workflow-checks: | |
| uses: pytorch/test-infra/.github/workflows/linux_job.yml@main | |
| needs: get-label-type | |
| with: | |
| runner: "${{ needs.get-label-type.outputs.label-type }}amz2023.linux.2xlarge" | |
| docker-image: pytorch-linux-focal-linter | |
| fetch-depth: -1 | |
| submodules: true | |
| ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} | |
| script: | | |
| # The generic Linux job chooses to use base env, not the one setup by the image | |
| CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]") | |
| conda activate "${CONDA_ENV}" | |
| # Regenerate workflows | |
| .github/scripts/generate_ci_workflows.py | |
| RC=0 | |
| # Assert that regenerating the workflows didn't change them | |
| if ! .github/scripts/report_git_status.sh .github/workflows; then | |
| echo | |
| echo 'As shown by the above diff, the committed .github/workflows' | |
| echo 'are not up to date according to .github/templates.' | |
| echo 'Please run this command, commit, and push again to your PR:' | |
| echo | |
| echo ' .github/scripts/generate_ci_workflows.py' | |
| echo | |
| echo 'If running that command does nothing, you may need to rebase' | |
| echo 'onto a more recent commit from the PyTorch main branch.' | |
| RC=1 | |
| fi | |
| # Check that jobs will be cancelled | |
| .github/scripts/ensure_actions_will_cancel.py | |
| exit $RC | |
| toc: | |
| uses: pytorch/test-infra/.github/workflows/linux_job.yml@main | |
| needs: get-label-type | |
| with: | |
| runner: "${{ needs.get-label-type.outputs.label-type }}amz2023.linux.2xlarge" | |
| docker-image: pytorch-linux-focal-linter | |
| fetch-depth: 0 | |
| ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} | |
| script: | | |
| # The generic Linux job chooses to use base env, not the one setup by the image | |
| CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]") | |
| conda activate "${CONDA_ENV}" | |
| # Regenerate ToCs and check that they didn't change | |
| set -eu | |
| export PATH=~/.npm-global/bin:"$PATH" | |
| for FILE in $(git grep -Il '<!-- toc -->' -- '**.md'); do | |
| markdown-toc --bullets='-' -i "$FILE" | |
| done | |
| if ! .github/scripts/report_git_status.sh .; then | |
| echo | |
| echo 'As shown by the above diff, the table of contents in one or' | |
| echo 'more Markdown files is not up to date with the file contents.' | |
| echo 'You can either apply that Git diff directly to correct the' | |
| echo 'table of contents, or if you have npm installed, you can' | |
| echo 'install the npm package markdown-toc and run the following' | |
| # shellcheck disable=SC2016 | |
| echo 'command (replacing $FILE with the filename for which you want' | |
| echo 'to regenerate the table of contents):' | |
| echo | |
| # shellcheck disable=SC2016 | |
| echo " markdown-toc --bullets='-' -i \"\$FILE\"" | |
| false | |
| fi | |
| test-tools: | |
| name: Test tools | |
| if: ${{ github.repository == 'pytorch/pytorch' }} | |
| uses: pytorch/test-infra/.github/workflows/linux_job.yml@main | |
| needs: get-label-type | |
| with: | |
| runner: "${{ needs.get-label-type.outputs.label-type }}amz2023.linux.2xlarge" | |
| docker-image: pytorch-linux-focal-linter | |
| fetch-depth: 0 | |
| ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }} | |
| script: | | |
| # The generic Linux job chooses to use base env, not the one setup by the image | |
| CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]") | |
| conda activate "${CONDA_ENV}" | |
| # Test tools | |
| python3 -m unittest discover -vs tools/test -p 'test_*.py' | |
| python3 -m unittest discover -vs .github/scripts -p 'test_*.py' | |
| test_run_test: | |
| name: Test `run_test.py` is usable without boto3/rockset | |
| if: ${{ github.repository == 'pytorch/pytorch' }} | |
| runs-on: linux.20_04.4x | |
| steps: | |
| - name: Checkout PyTorch | |
| uses: pytorch/pytorch/.github/actions/checkout-pytorch@main | |
| with: | |
| submodules: false | |
| fetch-depth: 1 | |
| - name: Setup Python 3.8 | |
| uses: actions/setup-python@v4 | |
| with: | |
| python-version: '3.8' | |
| architecture: x64 | |
| cache: pip | |
| - name: Install dependencies | |
| run: | | |
| pip install pytest-rerunfailures==11.1.* pytest-flakefinder==1.1.* pytest-xdist==3.3.* expecttest==0.1.* numpy==1.24.* | |
| pip install torch --pre --index-url https://download.pytorch.org/whl/nightly/cpu/ | |
| - name: Run run_test.py (nonretryable) | |
| run: | | |
| # Run test_vulkan, which is a fast noop on Linux | |
| python3 test/run_test.py --include test_vulkan --verbose | |
| test_collect_env: | |
| if: ${{ github.repository == 'pytorch/pytorch' }} | |
| name: Test collect_env | |
| runs-on: linux.20_04.4x | |
| strategy: | |
| matrix: | |
| test_type: [with_torch, without_torch, older_python_version] | |
| steps: | |
| # [see note: pytorch repo ref] | |
| # deep clone (fetch-depth 0) required, to allow us to use git log | |
| - name: Checkout PyTorch | |
| uses: pytorch/pytorch/.github/actions/checkout-pytorch@main | |
| with: | |
| submodules: false | |
| fetch-depth: 1 | |
| - name: Setup Python 3.6 | |
| if: matrix.test_type == 'older_python_version' | |
| uses: actions/setup-python@v4 | |
| with: | |
| python-version: '3.6' | |
| architecture: x64 | |
| check-latest: false | |
| cache: pip | |
| cache-dependency-path: | | |
| **/requirements.txt | |
| - name: Setup Python 3.8 | |
| if: matrix.test_type != 'older_python_version' | |
| uses: actions/setup-python@v4 | |
| with: | |
| python-version: '3.8' | |
| architecture: x64 | |
| check-latest: false | |
| cache: pip | |
| cache-dependency-path: | | |
| **/requirements.txt | |
| - name: Install torch | |
| if: matrix.test_type == 'with_torch' | |
| run: | | |
| pip install -r requirements.txt | |
| # Doesn't really matter what torch version, we just need ANY torch installed | |
| pip install 'torch==1.*' | |
| - name: Run collect_env.py (nonretryable) | |
| run: | | |
| # All we need to see is that it passes | |
| python3 torch/utils/collect_env.py | |
| concurrency: | |
| group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }} | |
| cancel-in-progress: true |