.github/workflows/check_failed_tests.yml

name: Process failed tests

on:
  workflow_call:
    inputs:
      docker:
        required: true
        type: string
      job:
        required: true
        type: string
      slack_report_channel:
        required: true
        type: string
      ci_event:
        required: true
        type: string
      report_repo_id:
        required: true
        type: string
      commit_sha:
        required: false
        type: string
      pr_number:
        required: false
        type: string
    outputs:
      report:
        description: "Content of the report of new failures"
        value: ${{ jobs.process_new_failures_with_commit_info.outputs.report }}

env:
  HF_HOME: /mnt/cache
  TRANSFORMERS_IS_CI: yes
  OMP_NUM_THREADS: 8
  MKL_NUM_THREADS: 8
  RUN_SLOW: yes
  # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access.
  # This token is created under the bot `hf-transformers-bot`.
  HF_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
  TF_FORCE_GPU_ALLOW_GROWTH: true
  CUDA_VISIBLE_DEVICES: 0,1


jobs:
  check_new_failures:
    name: "Find commits for new failing tests"
    strategy:
      matrix:
        run_idx: [1]
    runs-on:
      group: aws-g5-4xlarge-cache
    outputs:
      process: ${{ steps.check_file.outputs.process }}
    container:
      image: ${{ inputs.docker }}
      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
    steps:
      - uses: actions/download-artifact@v4
        with:
          name: ci_results_${{ inputs.job }}
          path: /transformers/ci_results_${{ inputs.job }}

      - name: Check file
        id: check_file
        working-directory: /transformers
        env:
          job: ${{ inputs.job }}
        run: |
          if [ -f "ci_results_${job}/new_failures.json" ]; then
            echo "\`ci_results_${job}/new_failures.json\` exists, continue ..."
            echo "process=true" >> $GITHUB_ENV
            echo "process=true" >> $GITHUB_OUTPUT
          else
            echo "\`ci_results_${job}/new_failures.json\` doesn't exist, abort."
            echo "process=false" >> $GITHUB_ENV
            echo "process=false" >> $GITHUB_OUTPUT
          fi

      - uses: actions/download-artifact@v4
        if: ${{ env.process == 'true' }}
        with:
          pattern: setup_values*
          path: setup_values
          merge-multiple: true

      - name: Prepare some setup values
        if: ${{ env.process == 'true' }}
        run: |
          if [ -f setup_values/prev_workflow_run_id.txt ]; then
            echo "PREV_WORKFLOW_RUN_ID=$(cat setup_values/prev_workflow_run_id.txt)" >> $GITHUB_ENV
          else
            echo "PREV_WORKFLOW_RUN_ID=" >> $GITHUB_ENV
          fi

      - name: Update clone
        working-directory: /transformers
        if: ${{ env.process == 'true' }}
        env:
          commit_sha: ${{ inputs.commit_sha || github.sha }}
        run: |
          git fetch origin "$commit_sha" && git checkout "$commit_sha"

      - name: Get `START_SHA`
        working-directory: /transformers/utils
        if: ${{ env.process == 'true' }}
        env:
          commit_sha: ${{ inputs.commit_sha || github.sha }}
        run: |
          echo "START_SHA=$commit_sha" >> $GITHUB_ENV

      # This is used if the CI is triggered from a pull request `self-comment-ci.yml` (after security check is verified)
      - name: Extract the base commit on `main` (of the merge commit created by Github) if it is a PR
        id: pr_info
        if: ${{ env.process == 'true' && inputs.pr_number != '' }}
        uses: actions/github-script@v6
        with:
          script: |            
            const { data: pr } = await github.rest.pulls.get({
              owner: context.repo.owner,
              repo: context.repo.repo,
              pull_number: ${{ inputs.pr_number }}
            });

            const { data: merge_commit }  = await github.rest.repos.getCommit({
              owner: pr.base.repo.owner.login,
              repo: pr.base.repo.name,
              ref: '${{ inputs.commit_sha }}',
            });

            core.setOutput('merge_commit_base_sha', merge_commit.parents[0].sha);

      # Usually, `END_SHA` should be the commit of the last previous workflow run of the **SAME** (scheduled) workflow.
      # (This is why we don't need to specify `workflow_id` which would be fetched automatically in the python script.)
      - name: Get `END_SHA` from previous CI runs of the same workflow
        working-directory: /transformers/utils
        if: ${{ env.process == 'true' && inputs.pr_number == '' }}
        env:
          ACCESS_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
        run: |
          echo "END_SHA=$(TOKEN="$ACCESS_TOKEN" python3 -c 'import os; from get_previous_daily_ci import get_last_daily_ci_run_commit; commit=get_last_daily_ci_run_commit(token=os.environ["TOKEN"], workflow_run_id=os.environ["PREV_WORKFLOW_RUN_ID"]); print(commit)')" >> $GITHUB_ENV

      # However, for workflow runs triggered by `issue_comment` (for pull requests), we want to check against the
      # parent commit (on `main`) of the `merge_commit` (dynamically created by GitHub). In this case, the goal is to
      # see if a reported failing test is actually ONLY failing on the `merge_commit`.
      - name: Set `END_SHA`
        if: ${{ env.process == 'true' && inputs.pr_number != '' }}
        env:
          merge_commit_base_sha: ${{ steps.pr_info.outputs.merge_commit_base_sha }}
        run: |
          echo "END_SHA=$merge_commit_base_sha" >> $GITHUB_ENV

      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
        working-directory: /transformers
        if: ${{ env.process == 'true' }}
        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .

      - name: NVIDIA-SMI
        if: ${{ env.process == 'true' }}
        run: |
          nvidia-smi

      - name: Environment
        working-directory: /transformers
        if: ${{ env.process == 'true' }}
        run: |
          python3 utils/print_env.py

      - name: Install pytest-flakefinder
        if: ${{ env.process == 'true' }}
        run: python3 -m pip install pytest-flakefinder

      - name: Show installed libraries and their versions
        working-directory: /transformers
        if: ${{ env.process == 'true' }}
        run: pip freeze

      - name: Check failed tests
        working-directory: /transformers
        if: ${{ env.process == 'true' }}
        env:
          job: ${{ inputs.job }}
          run_idx: ${{ matrix.run_idx }}
        run: python3 utils/check_bad_commit.py --start_commit "$START_SHA" --end_commit "$END_SHA" --file "ci_results_${job}/new_failures.json" --output_file "new_failures_with_bad_commit_${job}_${run_idx}.json"

      - name: Show results
        working-directory: /transformers
        if: ${{ env.process == 'true' }}
        env:
          job: ${{ inputs.job }}
          run_idx: ${{ matrix.run_idx }}
        run: |
          ls -l "new_failures_with_bad_commit_${job}_${run_idx}.json"
          cat "new_failures_with_bad_commit_${job}_${run_idx}.json"

      - name: Upload artifacts
        uses: actions/upload-artifact@v4
        with:
          name: new_failures_with_bad_commit_${{ inputs.job }}_${{ matrix.run_idx }}
          path: /transformers/new_failures_with_bad_commit_${{ inputs.job }}_${{ matrix.run_idx }}.json

  process_new_failures_with_commit_info:
    name: "process bad commit reports"
    needs: check_new_failures
    if: needs.check_new_failures.outputs.process == 'true'
    runs-on:
      group: aws-g5-4xlarge-cache
    outputs:
      report: ${{ steps.set_output.outputs.report }}
    container:
      image: ${{ inputs.docker }}
      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
    steps:
      - uses: actions/download-artifact@v4
        with:
          name: ci_results_${{ inputs.job }}
          path: /transformers/ci_results_${{ inputs.job }}

      - uses: actions/download-artifact@v4
        with:
          pattern: new_failures_with_bad_commit_${{ inputs.job }}*
          path: /transformers/new_failures_with_bad_commit_${{ inputs.job }}
          merge-multiple: true

      - name: Check files
        working-directory: /transformers
        env:
          job: ${{ inputs.job }}
        run: |
          ls -la /transformers
          ls -la "/transformers/new_failures_with_bad_commit_${job}"

      # Currently, we only run with a single runner by using `run_idx: [1]`. We might try to run with multiple runners
      # to further reduce the false positive caused by flaky tests, which requires further processing to merge reports.
      - name: Merge files
        shell: bash
        working-directory: /transformers
        env:
          job: ${{ inputs.job }}
        run: |
          cp "/transformers/new_failures_with_bad_commit_${job}/new_failures_with_bad_commit_${job}_1.json" new_failures_with_bad_commit.json

      - name: Update clone
        working-directory: /transformers
        env:
          commit_sha: ${{ inputs.commit_sha || github.sha }}
        run: |
          git fetch origin "$commit_sha" && git checkout "$commit_sha"

      - name: Process report
        shell: bash
        working-directory: /transformers
        env:
          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
          TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN: ${{ secrets.TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN }}
          JOB_NAME: ${{ inputs.job }}
          REPORT_REPO_ID: ${{ inputs.report_repo_id }}
        run: |
          {
            echo 'REPORT_TEXT<<EOF'
            python3 utils/process_bad_commit_report.py
            echo EOF
          } >> "$GITHUB_ENV"

      # The output is useful if a caller needs more processing, for example, we have a chain
      # self-comment-ci.yml -> self-scheduled.yml -> this one (check_failed_tests.yml),
      # and `self-comment-ci.yml` needs further processing before sending a GitHub comment to the pull request page.
      - name: Show results & Set outputs
        id: set_output
        working-directory: /transformers
        run: |
          ls -l new_failures_with_bad_commit.json
          cat new_failures_with_bad_commit.json

          {
            echo 'report<<EOF'
            cat new_failures_with_bad_commit.json
            echo ''  # Force a newline
            echo EOF
          } >> "$GITHUB_OUTPUT"

      - name: Upload artifacts
        uses: actions/upload-artifact@v4
        with:
          name: new_failures_with_bad_commit_${{ inputs.job }}
          path: /transformers/new_failures_with_bad_commit.json

      - name: Prepare Slack report title
        working-directory: /transformers
        env:
          ci_event: ${{ inputs.ci_event }}
          job: ${{ inputs.job }}
        run: |
          pip install slack_sdk
          echo "title=$(python3 -c 'import sys; import os; sys.path.append("utils"); from utils.notification_service import job_to_test_map; ci_event = os.environ["ci_event"]; job = os.environ["job"]; test_name = job_to_test_map[job]; title = f"New failed tests of {ci_event}" + ":" + f" {test_name}"; print(title)')" >> $GITHUB_ENV

      - name: Send processed report
        if: ${{ !endsWith(env.REPORT_TEXT, '{}') }}
        uses: slackapi/slack-github-action@6c661ce58804a1a20f6dc5fbee7f0381b469e001
        with:
          # Slack channel id, channel name, or user id to post message.
          # See also: https://api.slack.com/methods/chat.postMessage#channels
          channel-id: '#${{ inputs.slack_report_channel }}'
          # For posting a rich message using Block Kit
          payload: |
            {
              "blocks": [
                {
                  "type": "header",
                  "text": {
                    "type": "plain_text",
                    "text": "${{ env.title }}"
                  }
                },
                {
                  "type": "section",
                  "text": {
                    "type": "mrkdwn",
                    "text": "${{ env.REPORT_TEXT }}"
                  }
                }
              ]
            }
        env:
          SLACK_BOT_TOKEN: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}