ci(e2e): the frontend suite runs in six shards, and one report is merged from them

.github/workflows/validate.yml

@@ -644,9 +644,18 @@ jobs:
           if-no-files-found: error
 
   fe-e2e:
-    name: Frontend end-to-end tests
+    # Split the suite across N parallel shards. Playwright's own `--shard`
+    # partitions by spec file, so a spec always lands on the same shard and a
+    # failure is easy to place. The build in front of the run is 38s against
+    # ~12m of tests, so paying for it six times still wins by a wide margin.
+    name: Frontend e2e (shard ${{ matrix.shard }}/6)
     needs: fe-static
     runs-on: ubuntu-latest
+    timeout-minutes: 20
+    strategy:
+      fail-fast: false
+      matrix:
+        shard: [ 1, 2, 3, 4, 5, 6 ]
     defaults:
       run:
         working-directory: services/frontend
@@ -663,17 +672,63 @@ jobs:
       - name: Install Playwright browser dependencies
         run: yarn playwright install --with-deps chromium
 
+      # --no-report leaves the raw coverage unconverted: a shard only ever holds
+      # part of the picture, so fe-e2e-coverage below merges the six.
       - name: Run frontend end-to-end tests
-        run: yarn test:e2e
+        run: yarn test:e2e --no-report --shard=${{ matrix.shard }}/6
 
       - name: Upload frontend e2e traces
         if: failure()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
         with:
-          name: frontend-e2e-traces
+          # Per-shard artifact name so parallel jobs don't collide.
+          name: frontend-e2e-traces-shard-${{ matrix.shard }}
           path: services/frontend/test-results
           if-no-files-found: ignore
 
+      - name: Upload frontend e2e raw coverage
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: frontend-e2e-coverage-raw-${{ matrix.shard }}
+          path: services/frontend/coverage/e2e/raw
+          if-no-files-found: error
+
+  fe-e2e-coverage:
+    # Fan-in for the sharded suite: one report over all six slices, under the
+    # name the single job used to publish. The converter walks its raw directory
+    # recursively and merges every JSON it finds, so the per-shard subdirectories
+    # the download leaves behind need no flattening.
+    name: Frontend e2e coverage report
+    needs: fe-e2e
+    if: always() && needs.fe-e2e.result != 'skipped'
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    defaults:
+      run:
+        working-directory: services/frontend
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Set up Node.js and Yarn
+        uses: ./.github/actions/setup-node-yarn
+        with:
+          node-version: ${{ env.NODE_VERSION }}
+          yarn-version: ${{ env.YARN_VERSION }}
+
+      - name: Download every shard's raw coverage
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
+        with:
+          path: services/frontend/coverage/e2e/raw
+          pattern: frontend-e2e-coverage-raw-*
+
+      - name: Merge the shards into one report
+        run: |
+          yarn node ./scripts/convert-frontend-coverage.mjs \
+            --raw-dir coverage/e2e/raw \
+            --out-dir coverage/e2e
+
       - name: Upload frontend e2e coverage artifacts
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
@@ -762,6 +817,7 @@ jobs:
       - acceptance-features
       - fe-unit
       - fe-e2e
+      - fe-e2e-coverage
       - openapi-sync
     runs-on: ubuntu-latest
     timeout-minutes: 5

services/frontend/playwright.config.ts

@@ -9,21 +9,21 @@ export default defineConfig({
     timeout: 5_000,
   },
   fullyParallel: true,
-  // The e2e suite is I/O/wait-bound (browser navigation, rendering, mocked
-  // network) rather than CPU-bound, so oversubscribing the CI runner's vCPUs
-  // cuts wall-clock time: workers mostly await the browser, leaving CPU free.
-  // The bulk of the suite runs against a prebuilt app served by `vite preview`
-  // (see webServer below), which removes the dev-server compile bottleneck, so
-  // 8 workers on the 4-vCPU runner stays stable — concurrency-sensitive specs
-  // no longer race the compiler. See #424.
-  workers: process.env.CI ? 8 : undefined,
-  // Eight workers on the runner's four vCPUs stretch a local 0.8s test to ~6s, so a
-  // single starved assertion can blow the 5s cap on an otherwise green suite. The one
-  // retry also lets the trace above be captured, which retries: 0 made impossible.
-  retries: process.env.CI ? 1 : 0,
+  // CI runs the suite as six `--shard` slices, one job each, so a slice holds
+  // ~150 tests rather than 878. A short slice has no long tail to hide, which
+  // is what oversubscription bought: one worker per runner vCPU is enough, and
+  // it keeps a 0.8s test at 0.8s instead of stretching it toward the 5s cap.
+  workers: process.env.CI ? 4 : undefined,
+  // A retry under four workers would hide a real flake rather than absorb a
+  // starved assertion, so a failure is a failure. `trace` below is what makes
+  // that first failure readable.
+  retries: 0,
   reporter: "list",
   use: {
-    trace: "on-first-retry",
+    // retries: 0 means there is no second attempt to record, so the trace has
+    // to come off the first one. Passes are discarded, leaving a green run with
+    // no artifacts.
+    trace: "retain-on-failure",
     actionTimeout: 5_000,
     navigationTimeout: 5_000,
     // Every project but the motion one runs as a visitor who asked for reduced

services/frontend/scripts/run-e2e-with-coverage.mjs

@@ -6,7 +6,11 @@ import process from "node:process"
 
 const coverageDir = "coverage/e2e"
 const rawCoverageDir = `${coverageDir}/raw`
-const extraArgs = process.argv.slice(2)
+// CI shards the suite, so each slice only ever holds part of the picture: it
+// uploads its raw dir and a fan-in job merges the six into one report. Anything
+// but a shard still reports for itself.
+const report = !process.argv.includes("--no-report")
+const extraArgs = process.argv.slice(2).filter((arg) => arg !== "--no-report")
 
 rmSync(coverageDir, {recursive: true, force: true})
 
@@ -26,7 +30,7 @@ const e2eResult = spawnSync(
 
 const hasRawCoverage = existsSync(rawCoverageDir) && readdirSync(rawCoverageDir).some((file) => file.endsWith(".json"))
 
-if (hasRawCoverage) {
+if (hasRawCoverage && report) {
   const reportResult = spawnSync(
     "yarn",
     [
@@ -44,7 +48,7 @@ if (hasRawCoverage) {
   if (reportResult.status !== 0 && e2eResult.status === 0) {
     process.exit(reportResult.status ?? 1)
   }
-} else {
+} else if (!hasRawCoverage) {
   const message = `No raw coverage was captured in ${rawCoverageDir}`
   if (e2eResult.status === 0) {
     console.error(message)