Collect tracelane replays across a sharded Playwright run

What you’ll end up with

A single downloadable artifact containing every shard’s tracelane-reports/*.html. Because each failure is already a standalone file, there’s nothing to merge — you just collect them.

Prerequisites

• A Playwright suite already running sharded in CI (--shard)
• @tracelane/playwright wired in (see the 5-minute setup)

Steps

1. Upload each shard’s reports under a per-shard name

actions/upload-artifact@v4 forbids two jobs writing the same artifact name, so namespace by shard:

  test:
    strategy:
      fail-fast: false
      matrix:
        shard: [1, 2, 3]
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - run: npx playwright test --shard=${{ matrix.shard }}/3
      - name: Upload this shard's replays
        if: ${{ !cancelled() }}
        uses: actions/upload-artifact@v4
        with:
          name: tracelane-reports-${{ matrix.shard }}
          path: tracelane-reports/
          if-no-files-found: ignore

2. Gather them in a final job

  collect:
    needs: test
    if: ${{ !cancelled() }}
    runs-on: ubuntu-latest
    steps:
      - name: Download every shard's replays into one folder
        uses: actions/download-artifact@v4
        with:
          pattern: tracelane-reports-*
          merge-multiple: true
          path: tracelane-reports
      - name: Re-upload the combined set
        uses: actions/upload-artifact@v4
        with:
          name: tracelane-reports
          path: tracelane-reports/
          if-no-files-found: ignore

3. Open any failure

Download tracelane-reports, open any .html — each is a complete, offline replay on its own.

Why this works

tracelane’s artifact is per-failure, not a single aggregate report, and filenames carry a per-project id + a millisecond timestamp — so shards and parallel workers stay distinct and no merge step is required. download-artifact’s merge-multiple just flattens the per-shard folders into one.

Next steps

• Attach an rrweb replay to every Playwright failure on a PR
• Triage a CI run with 200+ failures