CI-UnitTest #217
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
| # Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved. | |
| # SPDX-License-Identifier: Apache-2.0 | |
| name: CI-UnitTest | |
| on: | |
| # This workflow includes AWS credential to run jobs on AWS batch. | |
| # Thus, this workflow cannot checkout PRs and can only be triggered by other workflows. | |
| workflow_run: | |
| workflows: ["CI-Entry", "CI-Entry-Nightly"] | |
| types: | |
| - completed | |
| defaults: | |
| run: | |
| shell: bash | |
| jobs: | |
| check_status: | |
| if: github.repository == 'awslabs/ratex' | |
| runs-on: ubuntu-latest | |
| outputs: | |
| trigger: ${{ steps.job_info.outputs.trigger }} | |
| ci_image: "metaprojdev/ratex:ci_gpu-latest" | |
| skip_ci: ${{ steps.job_info.outputs.skip_ci }} | |
| ref: ${{ steps.job_info.outputs.ref }} | |
| repo: ${{ steps.job_info.outputs.repo }} | |
| pr: ${{ steps.pr_job_info.outputs.pr }} | |
| sha: ${{ steps.pr_job_info.outputs.sha }} | |
| job_tag: ${{ steps.gen_tag.outputs.tag }} | |
| steps: | |
| # The workflow triggered by workflow run will not show its status | |
| # in Github by default, so we have to use this action to enable it. | |
| # Note that when this action is used, "name" in job is unavailable: | |
| # https://github.com/haya14busa/action-workflow_run-status/issues/158 | |
| - uses: haya14busa/action-workflow_run-status@v1 | |
| - name: Whether linting was failed | |
| if: ${{ github.event.workflow_run.conclusion != 'success' }} | |
| run: | | |
| conclusion=${{ github.event.workflow_run.conclusion }} | |
| echo "Linting wasn't successed: ${conclusion}" | |
| exit 1 | |
| - name: Download artifact | |
| uses: dawidd6/action-download-artifact@v2 | |
| with: | |
| workflow: ci_lint.yml | |
| run_id: ${{ github.event.workflow_run.id }} | |
| - name: Parse common job info | |
| id: job_info | |
| run: | | |
| skip_ci=$(head -n 1 artifact/skip.txt) | |
| echo "::set-output name=skip_ci::${skip_ci}" | |
| ref=$(head -n 1 artifact/ref.txt) | |
| echo "::set-output name=ref::${ref}" | |
| repo=$(head -n 1 artifact/repo.txt) | |
| echo "::set-output name=repo::${repo}" | |
| trigger=$(head -n 1 artifact/trigger.txt) | |
| echo "::set-output name=trigger::${trigger}" | |
| - name: Parse PR job info | |
| id: pr_job_info | |
| continue-on-error: true | |
| # Note that pr and sha only available for pull request, and will be empty for push events. | |
| run: | | |
| pr=$(head -n 1 artifact/pr.txt) | |
| echo "::set-output name=pr::${pr}" | |
| sha=$(head -n 1 artifact/sha.txt) | |
| echo "::set-output name=sha::${sha}" | |
| - name: Generate tag | |
| id: gen_tag | |
| # This tag is PR-unique so it can be used to connect jobs for the same PR. | |
| # For example, we use it to share ccache caches and cancel previous runs. | |
| run: | | |
| tag=${{ steps.job_info.outputs.repo }}/${{ steps.pr_job_info.outputs.pr }} | |
| echo "::set-output name=tag::${tag}" | |
| test_on_gpu: | |
| # Run full tests with the latest docker image on GPU. | |
| needs: [check_status] | |
| if: github.repository == 'awslabs/ratex' | |
| runs-on: ubuntu-latest | |
| steps: | |
| - uses: haya14busa/action-workflow_run-status@v1 | |
| - name: List environments | |
| run: | | |
| echo "Trigger: ${{ needs.check_status.outputs.trigger }}" | |
| echo "Docker Image: ${{ needs.check_status.outputs.ci_image }}" | |
| echo "Job tag: ${{ needs.check_status.outputs.job_tag }}" | |
| echo "Skip CI? ${{ needs.check_status.outputs.skip_ci }}" | |
| echo "REF: ${{ needs.check_status.outputs.ref }}" | |
| echo "REPO: ${{ needs.check_status.outputs.repo }}" | |
| echo "PR: ${{ needs.check_status.outputs.pr }}" | |
| echo "SHA: ${{ needs.check_status.outputs.sha }}" | |
| - name: Configure AWS Credentials | |
| uses: aws-actions/configure-aws-credentials@v1 | |
| with: | |
| aws-access-key-id: ${{ secrets.AWS_BATCH_ACCESS_ID }} | |
| aws-secret-access-key: ${{ secrets.AWS_BATCH_SECRET_ACCESS_KEY }} | |
| aws-region: us-west-2 | |
| - name: Checkout repository | |
| # No need to checkout submodules because we only need the script. | |
| uses: actions/checkout@v2 | |
| - name: Lint-Build-Test | |
| run: | | |
| # env vars are unavailable in job.if so we have to implement it here. | |
| if [ "${{ needs.check_status.outputs.skip_ci }}" == "1" ]; then | |
| echo "Skip unit tests" | |
| exit 0 | |
| fi | |
| echo "Running tests on CPU and GPU" | |
| python3 -m pip install argparse boto3 | |
| python3 ./ci/batch/submit-job.py \ | |
| --job-type GPU \ | |
| --image ${{ needs.check_status.outputs.ci_image }} \ | |
| --name ci-ratex-gpu-${{ needs.check_status.outputs.job_tag }} \ | |
| --job-queue ci-gpu-queue \ | |
| --job-def-cfg ./ci/batch/job-def-cfg.json \ | |
| --entry-script /batch/entry.sh \ | |
| --source-ref ${{ needs.check_status.outputs.ref }} \ | |
| --repo ${{ needs.check_status.outputs.repo }} \ | |
| --wait \ | |
| --command "bash ./ci/batch/cli.sh set_pytorch ${{ needs.check_status.outputs.trigger }} && | |
| bash ./ci/batch/cli.sh compile ${{ needs.check_status.outputs.job_tag }} && | |
| bash ./ci/batch/cli.sh lint && bash ./ci/batch/cli.sh unit_test GPU" | |
| test_on_multi_gpu: | |
| needs: [check_status] | |
| if: github.repository == 'awslabs/ratex' | |
| runs-on: ubuntu-latest | |
| steps: | |
| - uses: haya14busa/action-workflow_run-status@v1 | |
| - name: List environments | |
| run: | | |
| echo "Trigger: ${{ needs.check_status.outputs.trigger }}" | |
| echo "Docker Image: ${{ needs.check_status.outputs.ci_image }}" | |
| echo "Job tag: ${{ needs.check_status.outputs.job_tag }}" | |
| echo "Skip CI? ${{ needs.check_status.outputs.skip_ci }}" | |
| echo "REF: ${{ needs.check_status.outputs.ref }}" | |
| echo "REPO: ${{ needs.check_status.outputs.repo }}" | |
| echo "PR: ${{ needs.check_status.outputs.pr }}" | |
| echo "SHA: ${{ needs.check_status.outputs.sha }}" | |
| - name: Configure AWS Credentials | |
| uses: aws-actions/configure-aws-credentials@v1 | |
| with: | |
| aws-access-key-id: ${{ secrets.AWS_BATCH_ACCESS_ID }} | |
| aws-secret-access-key: ${{ secrets.AWS_BATCH_SECRET_ACCESS_KEY }} | |
| aws-region: us-west-2 | |
| - name: Checkout repository | |
| # No need to checkout submodules because we only need the script. | |
| uses: actions/checkout@v2 | |
| - name: Test | |
| run: | | |
| # env vars are unavailable in job.if so we have to implement it here. | |
| if [ "${{ needs.check_status.outputs.skip_ci }}" == "1" ]; then | |
| echo "Skip multi-GPU tests" | |
| exit 0 | |
| fi | |
| echo "Running tests on Multiply GPUs" | |
| python3 -m pip install argparse boto3 | |
| python3 ./ci/batch/submit-job.py \ | |
| --job-type multi-GPU \ | |
| --image ${{ needs.check_status.outputs.ci_image }} \ | |
| --name ci-ratex-multi-gpu-${{ needs.check_status.outputs.job_tag }} \ | |
| --job-queue ci-gpu-queue \ | |
| --job-def-cfg ./ci/batch/job-def-cfg.json \ | |
| --entry-script /batch/entry.sh \ | |
| --source-ref ${{ needs.check_status.outputs.ref }} \ | |
| --repo ${{ needs.check_status.outputs.repo }} \ | |
| --wait \ | |
| --command "bash ./ci/batch/cli.sh set_pytorch ${{ needs.check_status.outputs.trigger }} && | |
| bash ./ci/batch/cli.sh compile multi-GPU-${{ needs.check_status.outputs.job_tag }} && | |
| bash ./ci/batch/cli.sh unit_test multi-GPU" | |
| update_ci_badge: | |
| # Run this job whatever the unit tests were success or not. | |
| needs: [test_on_gpu, test_on_multi_gpu] | |
| if: ${{ always() && github.repository == 'awslabs/ratex' }} | |
| runs-on: ubuntu-latest | |
| steps: | |
| - uses: haya14busa/action-workflow_run-status@v1 | |
| - name: Checkout repository | |
| # No need to checkout submodules because we only need to get the HEAD commit hash. | |
| uses: actions/checkout@v2 | |
| - name: Download artifact | |
| uses: dawidd6/action-download-artifact@v2 | |
| with: | |
| workflow: ci_lint.yml | |
| run_id: ${{ github.event.workflow_run.id }} | |
| - name: Parse PR job info | |
| id: pr_job_info | |
| continue-on-error: true | |
| # Note that pr and sha only available for pull request, and will be empty for push events. | |
| run: | | |
| pr=$(head -n 1 artifact/pr.txt) | |
| trigger=$(head -n 1 artifact/trigger.txt) | |
| echo "::set-output name=pr::${pr}" | |
| echo "::set-output name=trigger::${trigger}" | |
| - name: Generate CI badge | |
| id: badge | |
| run: | | |
| # env vars are unavailable in job.if so we have to implement it here. | |
| echo "Trigger: ${{ steps.pr_job_info.outputs.trigger }}" | |
| echo "PR: ${{ steps.pr_job_info.outputs.pr }}" | |
| if [ "${{ steps.pr_job_info.outputs.pr }}" != "" ]; then | |
| echo "No need to update badge for PR CI. Skip." | |
| exit 0 | |
| fi | |
| if [ "${{ steps.pr_job_info.outputs.trigger }}" == "nightly" ]; then | |
| echo "No need to update badge for nightly PR CI. Skip." | |
| exit 0 | |
| fi | |
| echo "::set-output name=gist_id::aeb41ce3096bc1aaeb671f2c58836a3f" | |
| head_commit=$(git rev-parse --short HEAD) | |
| if [[ "${{ needs.test_on_gpu.result }}" == "success" && | |
| "${{ needs.test_on_multi_gpu.result }}" == "success" ]]; then | |
| echo "::set-output name=message::passing (${head_commit})" | |
| echo "::set-output name=color::success" | |
| else | |
| echo "::set-output name=message::failing (${head_commit})" | |
| echo "::set-output name=color::critical" | |
| fi | |
| - name: Update CI badge | |
| # Intentionally fail this step with empty gist_id. | |
| uses: schneegans/[email protected] | |
| continue-on-error: true | |
| with: | |
| auth: ${{ secrets.BOT_TOKEN }} | |
| gistID: ${{ steps.badge.outputs.gist_id }} | |
| filename: awslabs-ratex-ci-badge-last-pass.json | |
| label: CI-UnitTests | |
| message: ${{ steps.badge.outputs.message }} | |
| color: ${{ steps.badge.outputs.color }} |