serialize e2e tests due to flakiness (#3183)

.github/workflows/ci.yml

@@ -271,8 +271,6 @@ jobs:
       - name: E2E tests (Shard ${{ matrix.shard }}/${{ matrix.shardTotal }})
         env:
           FLAKINESS_ACCESS_TOKEN: ${{ secrets.FLAKINESS_ACCESS_TOKEN }}
-          # Self-hosted macOS runners can handle more parallelism
-          PLAYWRIGHT_PARALLELISM: ${{ contains(matrix.os.image, 'self-hosted') && '3' || '1' }}
           # Self-hosted macOS runners use fewer retries (1) vs GitHub-hosted CI (2)
           PLAYWRIGHT_RETRIES: ${{ contains(matrix.os.image, 'self-hosted') && '1' || '2' }}
         # You can add debug logging to make it easier to see what's failing

package.json

@@ -43,7 +43,7 @@
     "build": "npm run pre:e2e",
     "pre:e2e": "cross-env E2E_TEST_BUILD=true npm run package",
     "e2e": "playwright test",
-    "e2e:p3": "PLAYWRIGHT_RETRIES=1 PLAYWRIGHT_PARALLELISM=3 playwright test",
+    "e2e:fast": "PLAYWRIGHT_RETRIES=1 playwright test",
     "e2e:shard": "playwright test --shard",
     "storybook": "storybook dev -p 6006",
     "build-storybook": "storybook build"

playwright.config.ts

@@ -5,33 +5,21 @@ import { FAKE_LLM_BASE_PORT } from "./e2e-tests/helpers/test-ports";
 export { FAKE_LLM_BASE_PORT };
 
 const timestamp = new Date().toISOString().replace(/[:.]/g, "-");
-const parallelism = parseInt(process.env.PLAYWRIGHT_PARALLELISM || "1", 10);
 
-// Generate webServer configurations for each parallel worker
-// Each worker needs its own fake LLM server to avoid test interference
+// Single worker + one fake LLM server: parallel workers were flaky (shared state / timing).
 function generateWebServerConfigs(): PlaywrightTestConfig["webServer"] {
-  const configs: NonNullable<PlaywrightTestConfig["webServer"]> = [];
-
-  for (let i = 0; i < parallelism; i++) {
-    const port = FAKE_LLM_BASE_PORT + i;
-    configs.push({
-      // All servers run build to avoid race conditions since Playwright
-      // starts all webServer entries concurrently
-      command: `cd testing/fake-llm-server && npm run build && npm start -- --port=${port}`,
-      url: `http://localhost:${port}/health`,
-      // In CI, always start a fresh server; locally, reuse if one is already running
-      reuseExistingServer: !process.env.CI,
-    });
-  }
-
-  return configs;
+  return {
+    // Server runs build to avoid race conditions with concurrent webServer starts
+    command: `cd testing/fake-llm-server && npm run build && npm start -- --port=${FAKE_LLM_BASE_PORT}`,
+    url: `http://localhost:${FAKE_LLM_BASE_PORT}/health`,
+    // In CI, always start a fresh server; locally, reuse if one is already running
+    reuseExistingServer: !process.env.CI,
+  };
 }
 
 const config: PlaywrightTestConfig = {
   testDir: "./e2e-tests",
-  // Enable parallel test execution - E2E test builds skip the singleton lock
-  // Read parallelism from env var, default to 1 if not set
-  workers: parallelism,
+  workers: 1,
   retries: parseInt(
     process.env.PLAYWRIGHT_RETRIES ?? (process.env.CI ? "2" : "0"),
     10,