ci_unit_test.yml

# 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 CI-Lint.
  workflow_run:
    workflows: ["CI-Lint"]
    types:
      - completed

defaults:
  run:
    shell: bash

jobs:
  check_status:
    if: github.repository == 'awslabs/slapo'
    runs-on: ubuntu-latest
    outputs:
      # FIXME: We have to make lots of changes on AWS batch to allow
      # private docker hub images, so we currently use a private ECR image.
      docker_image: "metaprojdev/slapo:ci-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: 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 "skip_ci=${skip_ci}" >> $GITHUB_OUTPUT
          ref=$(head -n 1 artifact/ref.txt)
          echo "ref=${ref}" >> $GITHUB_OUTPUT
          repo=$(head -n 1 artifact/repo.txt)
          echo "repo=${repo}" >> $GITHUB_OUTPUT
          lint=${{ github.event.workflow_run.conclusion }}
          echo "Linting result: ${lint}"
      - name: Parse PR job info
        id: pr_job_info
        continue-on-error: true
        # Note that pr and sha only available for pull request, and will be empty for push events.
        run: |
          pr=$(head -n 1 artifact/pr.txt)
          echo "pr=${pr}" >> $GITHUB_OUTPUT
          sha=$(head -n 1 artifact/sha.txt)
          echo "sha=${sha}" >> $GITHUB_OUTPUT
      - name: Generate tag
        id: gen_tag
        # This tag is PR-unique so it can be used to connect jobs for the same PR.
        # For example, we use it to share ccache caches and cancel previous runs.
        run: |
          tag=${{ steps.job_info.outputs.repo }}/${{ steps.pr_job_info.outputs.pr }}
          echo "tag=${tag}" >> $GITHUB_OUTPUT
      - name: Whether linting was failed
        if: ${{ github.event.workflow_run.conclusion != 'success' }}
        run: exit 1
  unit_test:
    needs: [check_status]
    if: github.repository == 'awslabs/slapo'
    runs-on: ubuntu-latest
    steps:
      - uses: haya14busa/action-workflow_run-status@v1
      - name: List environments
        run: |
          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 tests"
            exit 0
          fi
          echo "Running tests on GPUs"
          python3 -m pip install argparse boto3
          python3 ./ci/submit_job.py \
            --platform multi-GPU \
            --image ${{ needs.check_status.outputs.docker_image }} \
            --name ci-gpu-${{ needs.check_status.outputs.job_tag }} \
            --job-queue ci-gpu-queue \
            --job-def-cfg ./ci/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/install_test_pkgs.sh; bash ./ci/task_unit_test.sh"
  update_ci_badge:
    needs: [unit_test]
    # Run this job whatever the unit tests were success or not.
    if: ${{ always() && github.repository == 'awslabs/slapo' }}
    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)
          echo "pr=${pr}" >> $GITHUB_OUTPUT
      - name: Generate CI badge
        id: badge
        run: |
          # env vars are unavailable in job.if so we have to implement it here.
          if [ "${{ steps.pr_job_info.outputs.pr }}" != "" ]; then
            echo "No need to update badge for PR CI. Skip."
            exit 0
          fi
          echo "gist_id=4b5f48daff54fb5d1a862c1649632648" >> $GITHUB_OUTPUT
          head_commit=$(git rev-parse --short HEAD)
          if [[ "${{ needs.unit_test.result }}" == "success" ]]; then
            echo "message=passing (${head_commit})" >> $GITHUB_OUTPUT
            echo "color=success" >> $GITHUB_OUTPUT
          else
            echo "message=failing (${head_commit})" >> $GITHUB_OUTPUT
            echo "color=critical" >> $GITHUB_OUTPUT
          fi
      - name: Update CI badge
        # Intentionally fail this step with empty gist_id.
        uses: schneegans/dynamic-badges-action@v1.1.0
        continue-on-error: true
        with:
          auth: ${{ secrets.BOT_TOKEN }}
          gistID: ${{ steps.badge.outputs.gist_id }}
          filename: awslabs-slapo-ci-badge-last-pass.json
          label: CI-UnitTests
          message: ${{ steps.badge.outputs.message }}
          color: ${{ steps.badge.outputs.color }}