harden e2e getState against transient /api/state 500s (ISS-127)

2026-09-22 17:29:45 +00:00
parent a8c4e93dc1
commit ae91bf6a5d
12 changed files with 89 additions and 33 deletions
@@ -47,9 +47,14 @@ interface StateBody {
 }

 async function getState(page: Page): Promise<StateBody> {
-  const res = await page.request.get('/api/state');
-  expect(res.status()).toBe(200);
-  return (await res.json()) as StateBody;
+  // ISS-127: /api/state intermittently 500s under scale-1000 dind load; a throw here would
+  // instantly fail any expect.poll callback, so retry briefly instead.
+  for (let attempt = 1; ; attempt++) {
+    const res = await page.request.get('/api/state');
+    if (res.status() === 200) return (await res.json()) as StateBody;
+    if (attempt >= 3) throw new Error(`/api/state returned ${res.status()} after 3 attempts`);
+    await page.waitForTimeout(500);
+  }
 }

 // Test hook (FLIGHT_TEST_HOOKS=1): set the world rate and VERIFY from the response that it stuck.