ci(hub-client-e2e): fix Playwright install hang (Node pin) + smoke-all seed/sync flakiness

.github/workflows/hub-client-e2e.yml

@@ -82,11 +82,24 @@ jobs:
           chmod +x tree-sitter-linux-x86
           sudo mv tree-sitter-linux-x86 /usr/local/bin/tree-sitter
 
-      # Node.js and dependencies
+      # Node.js and dependencies.
+      #
+      # PINNED to 24.15.0 — do NOT bump to a bare '24' (which floats to
+      # >=24.16). Node 24.16.0 introduced a yauzl stream-destruction
+      # regression that hangs `for await` over `openReadStream`, which
+      # makes `playwright install chromium` hang FOREVER right after the
+      # browser download hits 100% (extraction deadlocks; the job then
+      # burns the full 6h cap). It is fixed in Playwright >= 1.60.0, but
+      # we pin @playwright/test at ^1.50.0 -> 1.58.0. 24.15.0 is the last
+      # Node release before the regression, so it keeps Playwright 1.58 /
+      # Chromium 145 and leaves the visual-regression baselines unchanged.
+      # Refs: microsoft/playwright#40724, fixed by #40747 (PW 1.60.0).
+      # REMOVE this pin once @playwright/test is bumped to >= 1.60.0 (and
+      # regenerate the visual baselines for the newer Chromium then).
       - name: Set up Node.js
         uses: actions/setup-node@v6
         with:
-          node-version: '24'
+          node-version: '24.15.0'
           cache: 'npm'
 
       - name: Install npm dependencies
@@ -127,41 +140,8 @@ jobs:
       - name: Pre-build hub binary
         run: cargo build --bin hub
 
-      # Cache the downloaded browser binaries across runs. Keyed on the
-      # lockfile so a Playwright version bump invalidates the cache and
-      # re-downloads; otherwise we skip the 175 MB chromium download
-      # entirely.
-      - name: Cache Playwright browsers
-        uses: actions/cache@v4
-        with:
-          path: ~/.cache/ms-playwright
-          key: playwright-${{ runner.os }}-${{ hashFiles('package-lock.json') }}
-
-      # Install Playwright browsers.
-      #
-      # The full Chromium (Chrome for Testing) build is served from a
-      # SINGLE mirror with no fallback: cdn.playwright.dev/chrome-for-
-      # testing-public (Playwright's cftUrl() helper). cdn.playwright.dev
-      # is fronted by Azure Front Door, and from the (also-Azure) GitHub
-      # runners that front door sends all ~167 MiB and then holds the
-      # connection open without EOF — so Playwright's download promise
-      # never resolves and the step hangs until the 6h job cap. With no
-      # fallback mirror for Chrome for Testing, Playwright can't recover.
-      # (Carlos report, 2026-05; confirmed from the 100%-then-hang in the
-      # run-26760896905 log.)
-      #
-      # Fix: point the chromium download host at the GCS origin that
-      # cdn.playwright.dev merely redirects to. That origin serves the zip
-      # with a correct content-length and a clean EOF. On linux-x64 BOTH
-      # the full Chromium and the headless-shell resolve via cftUrl(), so
-      # this one var fixes both; ffmpeg's name isn't "chromium*" so the
-      # override doesn't touch it and it keeps its 3-mirror fallback.
-      # PLAYWRIGHT_DOWNLOAD_CONNECTION_TIMEOUT is a belt-and-suspenders so
-      # any future post-100% stall aborts in 2 min instead of hanging.
+      # Install Playwright browsers
       - name: Install Playwright
-        env:
-          PLAYWRIGHT_CHROMIUM_DOWNLOAD_HOST: https://storage.googleapis.com/chrome-for-testing-public
-          PLAYWRIGHT_DOWNLOAD_CONNECTION_TIMEOUT: '120000'
         run: |
           cd hub-client
           npx playwright install --with-deps chromium

hub-client/e2e/helpers/projectFactory.ts

@@ -161,6 +161,17 @@ export async function bootstrapProjectSet(
  * synced project set is initialized; otherwise the App lands on the
  * needs-migration screen.
  *
+ * Before returning, this waits until the seeded project is actually present
+ * in the *connected, synced* project set — not just written to IDB. That
+ * closes the race behind the smoke-all flakiness (bd-3nzyd): `addProject`
+ * only writes IDB, and the app reconciles IDB→set on the status→connected
+ * transition, which does NOT re-fire for a project seeded *after* the set is
+ * already connected. So we wait for the real peer connection, run the
+ * idempotent reconciler ourselves, and wait for the project to appear. The
+ * full Automerge sync path stays exercised end-to-end — the test just stops
+ * racing it. Waits are bounded so a genuinely unreachable sync server fails
+ * loudly here instead of surfacing 75s later as a preview-render timeout.
+ *
  * Returns the local project ID (UUID) used in URL navigation.
  */
 export async function seedProjectInBrowser(
@@ -175,6 +186,34 @@ export async function seedProjectInBrowser(
       const hooks = window.__quartoTest;
       if (!hooks) throw new Error('__quartoTest missing — rebuild with VITE_E2E=1');
       const entry = await hooks.projectStorage.addProject(indexDocId, syncServer, name);
+
+      const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));
+      const deadline = Date.now() + 30000;
+
+      // 1) Wait for the real project-set peer connection (the app's implicit
+      //    5s waitForPeer is too tight in CI; give it a generous window).
+      while (!hooks.projectSet.isConnected() && Date.now() < deadline) {
+        await sleep(100);
+      }
+      if (!hooks.projectSet.isConnected()) {
+        throw new Error(
+          'Project set did not reach connected state within 30s — sync server unreachable?',
+        );
+      }
+
+      // 2) Land the seeded IDB entry into the synced set (idempotent), then
+      //    wait until it is observably present before the caller navigates.
+      while (!hooks.projectSet.getProject(indexDocId) && Date.now() < deadline) {
+        await hooks.reconcileProjectSet();
+        if (hooks.projectSet.getProject(indexDocId)) break;
+        await sleep(100);
+      }
+      if (!hooks.projectSet.getProject(indexDocId)) {
+        throw new Error(
+          `Seeded project ${indexDocId} never appeared in the connected project set within 30s`,
+        );
+      }
+
       return entry.id;
     },
     { indexDocId, syncServer, name },

hub-client/e2e/helpers/testHooks.ts

@@ -24,10 +24,14 @@
  * ```
  */
 import type * as projectStorage from '../../src/services/projectStorage';
+import type * as projectSet from '../../src/services/projectSetService';
+import type { reconcileIntoConnectedProjectSet } from '../../src/services/projectSetReconciler';
 import type * as wasmRenderer from '../../src/services/wasmRenderer';
 
 export interface QuartoTestHooks {
   projectStorage: typeof projectStorage;
+  projectSet: typeof projectSet;
+  reconcileProjectSet: typeof reconcileIntoConnectedProjectSet;
   wasmRenderer: typeof wasmRenderer;
 }
 

hub-client/src/test-hooks.ts

@@ -13,15 +13,32 @@
  * out of the production bundle entirely.
  */
 import * as projectStorage from './services/projectStorage';
+import * as projectSet from './services/projectSetService';
+import { reconcileIntoConnectedProjectSet } from './services/projectSetReconciler';
 import * as wasmRenderer from '@quarto/preview-runtime';
 
 declare global {
   interface Window {
     __quartoTest?: {
       projectStorage: typeof projectStorage;
+      // The live project-set service singleton (same instance the app uses),
+      // so the E2E suite can observe real connection/sync state — e.g. wait
+      // for `isConnected()` and `getProject(indexDocId)` after seeding before
+      // navigating, instead of racing the implicit reconcile-on-connect.
+      projectSet: typeof projectSet;
+      // Idempotent IDB→synced-set reconciler. The app runs this only on the
+      // status→connected transition, which does not re-fire for a project
+      // seeded after the set is already connected; the suite invokes it
+      // explicitly so a seeded project deterministically lands in the set.
+      reconcileProjectSet: typeof reconcileIntoConnectedProjectSet;
       wasmRenderer: typeof wasmRenderer;
     };
   }
 }
 
-window.__quartoTest = { projectStorage, wasmRenderer };
+window.__quartoTest = {
+  projectStorage,
+  projectSet,
+  reconcileProjectSet: reconcileIntoConnectedProjectSet,
+  wasmRenderer,
+};