test: gate final replay on 25 focused historical comparisons #9
This file contains hidden or 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
| name: Design diff frozen benchmark | |
| on: | |
| push: | |
| branches: [codex/design-diff-benchmark-v3] | |
| permissions: | |
| contents: read | |
| concurrency: | |
| group: design-diff-benchmark-v3 | |
| cancel-in-progress: true | |
| env: | |
| ENGINE_SHA: e62bc4c33c4163bd18f96e6f15d30308daf16615 | |
| jobs: | |
| replay: | |
| runs-on: ${{ (vars.CI_PROVIDER == '' || vars.CI_PROVIDER == 'blacksmith') && 'blacksmith-2vcpu-ubuntu-2404' || 'ubuntu-latest' }} | |
| timeout-minutes: 10 | |
| strategy: | |
| fail-fast: false | |
| max-parallel: 20 | |
| matrix: | |
| shard: [3, 4, 10, 14, 18, 26, 36, 39, 47, 53, 56, 63, 64, 70, 91, 94, 97, 102, 103, 114, 120, 123, 150, 169, 175] | |
| steps: | |
| - name: Checkout immutable engine | |
| uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 | |
| with: | |
| ref: ${{ env.ENGINE_SHA }} | |
| path: engine | |
| fetch-depth: 0 | |
| persist-credentials: false | |
| - name: Checkout benchmark driver | |
| uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 | |
| with: | |
| ref: ${{ github.sha }} | |
| path: driver | |
| persist-credentials: false | |
| - name: Setup Bun | |
| uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2 | |
| with: | |
| bun-version: 1.4.1 | |
| - name: Install trusted dependencies | |
| working-directory: engine | |
| run: bun install --frozen-lockfile --ignore-scripts | |
| - name: Freeze shard and fetch source as data | |
| env: | |
| SHARD: ${{ matrix.shard }} | |
| GH_TOKEN: ${{ github.token }} | |
| run: bun --no-env-file driver/scripts/design-diff/cloud-benchmark.ts | |
| - name: Replay frozen comparisons | |
| working-directory: engine | |
| run: | | |
| bun --no-env-file scripts/design-diff/benchmark.ts \ | |
| --engine "$PWD" --sha "$ENGINE_SHA" --workers 1 --timeout-seconds 180 \ | |
| --manifest "$RUNNER_TEMP/comparisons.json" --output "$RUNNER_TEMP/results" | |
| - name: Retain available results | |
| if: always() | |
| uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4 | |
| with: | |
| name: design-benchmark-${{ env.ENGINE_SHA }}-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.shard }} | |
| path: ${{ runner.temp }}/results/ | |
| retention-days: 7 | |
| if-no-files-found: error |