deflake_tests.yaml

name: Deflake Tests
description: Re-triggers test jobs if they failed.
on:
  workflow_run:
    workflows: [linux]
    types:
      - completed

jobs:
  rerun-failing-jobs:
    runs-on: ubuntu-latest
    # if: github.event.workflow_run.conclusion == 'failure' || github.event.workflow_run.conclusion == 'cancelled'
    permissions:
      actions: write
      checks: read
    steps:
      - name: Download CI Essentials
        uses: actions/download-artifact@v4
        with:
          pattern: ci-essentials-*
          merge-multiple: true
          run-id: ${{ github.event.workflow_run.id }}
          github-token: ${{ secrets.GITHUB_TOKEN }}

      - name: Check Retry Count
        id: retry_check
        env:
          DEFLAKE_ATTEMPTS: 3
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          HEAD_REF: ${{ github.event.workflow_run.pull_requests[0].head.ref }}
          PR_NUMBER: ${{ github.event.workflow_run.pull_requests[0].number }}
          WORKFLOW: ${{ github.event.workflow_run.id }}
        shell: bash
        run: |
          set +x
          # Attempt to access the last attempt of the workflow that triggereed this one.
          if gh run view ${WORKFLOW} --attempt ${DEFLAKE_ATTEMPTS}; then
            echo "continue=false" >> ${GITHUB_OUTPUT}
          else
            echo "continue=true" >> ${GITHUB_OUTPUT}
          fi

      - name: Rerun Failing Checks
        if: steps.retry_check.outputs.continue == 'true'
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          PR_NUMBER: ${{ github.event.workflow_run.pull_requests[0].number }}
          REPO: ${{ github.repository }}
        shell: bash
        run: |
          set +x
          echo "Fetching failing checks for PR #${PR_NUMBER}"
          failing_runs=$(gh pr checks "${PR_NUMBER}" --repo "${REPO}" | grep -E 'fail|cancel')

          if [ -z "${failing_runs}" ]; then
            echo "No failing checks found to rerun."
            exit 0
          fi

          run_ids_to_rerun=$(echo "${failing_runs}" | python3 ./.github/scripts/extract_run_ids.py)

          if [ -z "${run_ids_to_rerun}" ]; then
            echo "Could not extract run IDs from failing checks."
            exit 0
          fi

          for run_id in ${run_ids_to_rerun}; do
            gh run rerun "${run_id}" --repo "${REPO}"
          done