ci(frontend): refresh the bundle-size baseline nightly, not on every push

Persisting a new baseline on every push to master (which happens many times a day) burns a full production build each time for no benefit a once-a-day refresh doesn't already cover -- a day-old baseline is fine for catching relative regressions on PRs. Moves baseline persistence to a new frontend-bundle-size-nightly.yml (schedule + workflow_dispatch), and restricts superset-frontend.yml's bundle-size job to pull_request only: compare-and-alert against the last nightly baseline, never persist. Also caches webpack's persistent filesystem build cache (superset-frontend/.temp_cache) across CI runs via actions/cache, keyed on the same files webpack's own `buildDependencies` invalidates on. Without this, every PR run would pay the full cold-build cost (several minutes) instead of the ~20s warm-build cost the cache is supposed to buy -- GH-hosted runners are fresh VMs with nothing carried over between jobs, so the cache has to be restored explicitly or it does nothing.
2026-08-05 21:42:48 +00:00 · 2026-07-28 09:38:31 -07:00
parent e1d7e61514
commit f7e29c15d5
2 changed files with 130 additions and 14 deletions
--- a/.github/workflows/frontend-bundle-size-nightly.yml
+++ b/.github/workflows/frontend-bundle-size-nightly.yml
@@ -0,0 +1,100 @@
+name: Frontend bundle size (nightly baseline)
+
+# Refreshes the bundle-size baseline that superset-frontend.yml's `bundle-size`
+# job compares PRs against. Deliberately NOT triggered on every push to
+# master: a day-old baseline is fine for catching relative regressions on
+# PRs, and building the production bundle on every one of the many pushes
+# master gets per day would burn CI time for no benefit a nightly refresh
+# doesn't already cover.
+on:
+  schedule:
+    - cron: "0 6 * * *"
+  workflow_dispatch: {}
+
+concurrency:
+  group: ${{ github.workflow }}
+  cancel-in-progress: true
+
+env:
+  TAG: apache/superset:bundle-size-nightly-${{ github.run_id }}
+
+permissions:
+  contents: read
+
+jobs:
+  refresh-baseline:
+    runs-on: ubuntu-26.04
+    timeout-minutes: 30
+    steps:
+      - name: "Checkout master"
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+          ref: master
+
+      - name: Build Docker Image
+        run: |
+          docker buildx build \
+            -t $TAG \
+            --cache-from=type=registry,ref=apache/superset-cache:3.11-slim-trixie \
+            --target superset-node-ci \
+            .
+
+      # Same cache the PR-time bundle-size job restores/writes -- webpack's
+      # persistent filesystem cache turns a warm production build into ~20s
+      # instead of several minutes. See superset-frontend.yml for the
+      # matching restore step and why it's keyed this way.
+      - name: Restore webpack build cache
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: superset-frontend/.temp_cache
+          key: >-
+            webpack-prod-cache-${{ hashFiles('superset-frontend/package-lock.json',
+            'superset-frontend/babel.config.js', 'superset-frontend/tsconfig.json',
+            'superset-frontend/webpack.config.js') }}
+
+      # Only ever pull the last recorded data point off the cache, keyed by
+      # run ID -- `restore-keys` prefix-matches the most recently created
+      # entry. Absent on the very first run ever; benchmark-action starts a
+      # fresh history in that case.
+      - name: Restore bundle size history
+        uses: actions/cache/restore@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: bundle-size-history.json
+          key: bundle-size-history-${{ github.run_id }}
+          restore-keys: |
+            bundle-size-history-
+
+      - name: Build production bundle with stats
+        run: |
+          mkdir -p ${{ github.workspace }}/superset-frontend/bundle-stats
+          mkdir -p ${{ github.workspace }}/superset-frontend/.temp_cache
+          docker run \
+          -v ${{ github.workspace }}/superset-frontend/bundle-stats:/app/superset-frontend/bundle-stats \
+          -v ${{ github.workspace }}/superset-frontend/.temp_cache:/app/superset-frontend/.temp_cache \
+          --rm $TAG \
+          bash -c \
+          "npm i && BUNDLE_SIZE_STATS=true npm run build -- --json=bundle-stats/stats.json"
+
+      - name: Summarize bundle size
+        run: |
+          node superset-frontend/scripts/bundle-size-summary.js \
+            superset-frontend/bundle-stats/stats.json > bundle-size-summary.json
+          rm -rf superset-frontend/bundle-stats
+
+      # No PR to comment on here, so comment-on-alert is off -- the job
+      # summary (summary-always) is the only surface for this run.
+      - name: Update bundle size baseline
+        uses: benchmark-action/github-action-benchmark@52576c92bccf6ac60c8223ec7eb2565637cae9ba # v1.22.1
+        with:
+          tool: customSmallerIsBetter
+          output-file-path: bundle-size-summary.json
+          external-data-json-path: bundle-size-history.json
+          fail-on-alert: false
+          summary-always: true
+
+      - name: Save bundle size history
+        uses: actions/cache/save@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: bundle-size-history.json
+          key: bundle-size-history-${{ github.run_id }}
--- a/.github/workflows/superset-frontend.yml
+++ b/.github/workflows/superset-frontend.yml
@@ -196,9 +196,15 @@ jobs:
          docker run --rm $TAG bash -c \
          "npm run build-storybook && npx playwright install-deps && npx playwright install chromium && npm run test-storybook:ci"

+  # Compares a PR's own bundle size against the last nightly-recorded
+  # baseline (see frontend-bundle-size-nightly.yml, which owns actually
+  # persisting new baselines). PR-only: a push to master doesn't need this
+  # check re-run against itself, and re-persisting the baseline on every
+  # push to master -- which happens many times a day -- would burn a full
+  # production build for no benefit nightly refresh doesn't already cover.
  bundle-size:
    needs: frontend-build
-    if: needs.frontend-build.outputs.should-run == 'true'
+    if: needs.frontend-build.outputs.should-run == 'true' && github.event_name == 'pull_request'
    runs-on: ubuntu-26.04
    timeout-minutes: 15
    permissions:
@@ -219,10 +225,26 @@ jobs:
        run: |
          zstd -d < docker-image.tar.zst | docker load

+      # webpack's persistent filesystem cache (superset-frontend/webpack.config.js)
+      # turns a warm production build into ~20s instead of several minutes,
+      # but GH-hosted runners are fresh VMs with nothing carried over between
+      # jobs -- without restoring it explicitly, every single PR would pay
+      # the full cold-build cost. Keyed on the same files webpack's own
+      # `buildDependencies` invalidates on, so a stale cache is never used.
+      - name: Restore webpack build cache
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: superset-frontend/.temp_cache
+          key: >-
+            webpack-prod-cache-${{ hashFiles('superset-frontend/package-lock.json',
+            'superset-frontend/babel.config.js', 'superset-frontend/tsconfig.json',
+            'superset-frontend/webpack.config.js') }}
+
      # Only ever pull the last recorded data point off the cache, keyed by
      # run ID -- `restore-keys` prefix-matches the most recently created
-      # entry. Absent on the very first run ever; benchmark-action starts a
-      # fresh history in that case.
+      # entry, which is always the latest nightly run. Absent before the
+      # first nightly run ever happens; benchmark-action starts a fresh
+      # history in that case.
      - name: Restore bundle size history
        uses: actions/cache/restore@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
        with:
@@ -234,8 +256,10 @@ jobs:
      - name: Build production bundle with stats
        run: |
          mkdir -p ${{ github.workspace }}/superset-frontend/bundle-stats
+          mkdir -p ${{ github.workspace }}/superset-frontend/.temp_cache
          docker run \
          -v ${{ github.workspace }}/superset-frontend/bundle-stats:/app/superset-frontend/bundle-stats \
+          -v ${{ github.workspace }}/superset-frontend/.temp_cache:/app/superset-frontend/.temp_cache \
          --rm $TAG \
          bash -c \
          "npm i && BUNDLE_SIZE_STATS=true npm run build -- --json=bundle-stats/stats.json"
@@ -246,7 +270,9 @@ jobs:
            superset-frontend/bundle-stats/stats.json > bundle-size-summary.json
          rm -rf superset-frontend/bundle-stats

-      - name: Track bundle size
+      # Comparison + alert only -- this job never persists. See
+      # frontend-bundle-size-nightly.yml for why.
+      - name: Compare bundle size against nightly baseline
        uses: benchmark-action/github-action-benchmark@52576c92bccf6ac60c8223ec7eb2565637cae9ba # v1.22.1
        with:
          tool: customSmallerIsBetter
@@ -257,13 +283,3 @@ jobs:
          alert-threshold: "110%"
          fail-on-alert: false
          summary-always: true
-
-      # Only the trunk's own bundle size becomes the baseline everyone else
-      # is compared against -- an unmerged PR's numbers (including ones that
-      # regress on purpose to test something) never get persisted.
-      - name: Save bundle size history
-        if: github.event_name == 'push' && github.ref == 'refs/heads/master'
-        uses: actions/cache/save@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
-        with:
-          path: bundle-size-history.json
-          key: bundle-size-history-${{ github.run_id }}