fix(status): show a strip when the homeserver hangs, not 110 s later (#317)

A hung Synapse (accepts connections, never answers) kept sync SYNCING
for up to 110 s (pollTimeout 30 s + the SDK's 80 s buffer), so no strip
showed and Kuma was never asked. A watchdog now notices a /sync overdue
by 45 s, confirms with a 5 s GET /_matrix/client/versions, then shows
"Lotus Chat's server isn't responding…" and polls Kuma at once.

Fixes #317

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017PCb59uY3UKqDqt7B6hzd5

e2e/security-nudge.spec.ts

@@ -29,32 +29,6 @@ async function settled(page: Page) {
   await page.waitForTimeout(2000);
 }
 
-/** [Gitea #307] Serve a Kuma status page for `localhost` with one posted announcement. */
-async function mockAnnouncement(page: Page) {
-  await page.route(/\/config\.json(\?.*)?$/, async (route) => {
-    const res = await route.fetch();
-    const cfg = await res.json();
-    cfg.statusPages = { localhost: { url: 'https://isitup.lotusguild.org', slug: 'matrix' } };
-    await route.fulfill({ response: res, json: cfg });
-  });
-  await page.route(/isitup\.lotusguild\.org\/api\/status-page\//, async (route) => {
-    const cors = { 'access-control-allow-origin': '*' };
-    if (route.request().url().includes('/heartbeat/')) {
-      await route.fulfill({ headers: cors, json: { heartbeatList: {}, uptimeList: {} } });
-      return;
-    }
-    await route.fulfill({
-      headers: cors,
-      json: {
-        config: { slug: 'matrix' },
-        incidents: [{ id: 7, title: 'Community call on Friday', content: '', style: 'info' }],
-        maintenanceList: [],
-        publicGroupList: [],
-      },
-    });
-  });
-}
-
 test.describe('security nudge (#110, #123)', () => {
   let user: TestUser;
 
@@ -86,22 +60,6 @@ test.describe('security nudge (#110, #123)', () => {
     await expect(page.getByText('Device Verification').first()).toBeVisible();
   });
 
-  test('a dismissed announcement no longer hides the nudge (#307)', async ({ page }) => {
-    test.setTimeout(200_000);
-    await mockAnnouncement(page);
-    await loginUI(page, user);
-    await pastGrace(page);
-    await settled(page);
-    // The announcement wins the one strip slot until it's dismissed.
-    const announcement = page.getByRole('status').filter({ hasText: 'Community call on Friday' });
-    await expect(announcement).toBeVisible();
-    await expect(strip(page)).toHaveCount(0);
-    await announcement.getByRole('button', { name: 'Dismiss' }).click();
-    await page.reload();
-    await settled(page);
-    await expect(strip(page)).toContainText("Your encryption keys aren't backed up");
-  });
-
   test('"Not now" snoozes it across a reload', async ({ page }) => {
     test.setTimeout(200_000);
     await loginUI(page, user);

src/app/features/security-nudge/SecurityBanner.tsx

@@ -14,7 +14,7 @@ import { useKeyBackupInfo, useKeyBackupStatusChange } from '../../hooks/useKeyBa
 import { useAlive } from '../../hooks/useAlive';
 import { useSyncState } from '../../hooks/useSyncState';
 import { settingsRequestAtom } from '../../state/settingsRequest';
-import { serverStatusAtom, statusDismissedAtom } from '../../state/serverStatus';
+import { serverStatusAtom } from '../../state/serverStatus';
 import { pickBanner } from '../../utils/kumaStatus';
 import {
   dismissNudge,
@@ -66,7 +66,6 @@ function SecurityBannerFor({ crypto }: { crypto: CryptoApi }) {
   const backupInfo = useKeyBackupInfo(crypto);
   const backupActive = useBackupActive(crypto);
   const { status: serverStatus, syncLost } = useAtomValue(serverStatusAtom);
-  const statusDismissed = useAtomValue(statusDismissedAtom);
   const requestSettings = useSetAtom(settingsRequestAtom);
   // Not urgent: wait until sync has settled (two SYNCING in a row), so it
   // never stacks under the "Connecting…" strip at startup.
@@ -92,11 +91,7 @@ function SecurityBannerFor({ crypto }: { crypto: CryptoApi }) {
   });
 
   // One strip at a time: outages, maintenance and a lost connection win.
-  // [Gitea #307] Pass what the user dismissed: a dismissed announcement isn't
-  // on screen and must not keep the nudge hidden for as long as it's posted.
-  const otherStrip =
-    syncLost ||
-    !!pickBanner(serverStatus, { syncLost, where: 'client', dismissed: new Set(statusDismissed) });
+  const otherStrip = syncLost || !!pickBanner(serverStatus, { syncLost, where: 'client' });
   if (!settled || !deviceId || !kind || otherStrip || !shouldShowNudge(kind, record, Date.now()))
     return null;
 

src/app/features/server-status/ServerStatusBanner.tsx

@@ -1,5 +1,5 @@
 import React, { useCallback, useEffect, useMemo, useState } from 'react';
-import { useAtom, useAtomValue, useSetAtom } from 'jotai';
+import { useAtomValue, useSetAtom } from 'jotai';
 import { MatrixClient, SyncState } from 'matrix-js-sdk';
 import { Box, Button, config, Icon, IconButton, Icons, IconSrc, Line, Text } from 'folds';
 import { ContainerColor } from '../../styles/ContainerColor.css';
@@ -7,8 +7,21 @@ import { useClientConfig } from '../../hooks/useClientConfig';
 import { useSyncState } from '../../hooks/useSyncState';
 import { useServerStatus } from '../../hooks/useServerStatus';
 import { ScreenSize, useScreenSizeContext } from '../../hooks/useScreenSize';
-import { serverStatusAtom, statusDismissedAtom } from '../../state/serverStatus';
+import { serverStatusAtom } from '../../state/serverStatus';
 import { pickBanner, resolveStatusPage, StatusBanner } from '../../utils/kumaStatus';
+import { homeserverStalledAtom, useHomeserverStall } from '../../hooks/useHomeserverStall';
+
+const DISMISS_KEY = 'lotus-status-banner-dismissed';
+const DISMISS_MAX = 50;
+
+const readDismissed = (): string[] => {
+  try {
+    const v = JSON.parse(localStorage.getItem(DISMISS_KEY) ?? '[]');
+    return Array.isArray(v) ? v.filter((k): k is string => typeof k === 'string') : [];
+  } catch {
+    return [];
+  }
+};
 
 const bannerIcon = (b: StripContent): IconSrc => {
   if (b.kind === 'maintenance') return Icons.Setting;
@@ -151,7 +164,12 @@ export function ServerStatusFeature({ mx }: { mx: MatrixClient }) {
       });
     }, []),
   );
-  const syncLost = lostAt !== null;
+  // [Gitea #317] A hung server keeps sync "SYNCING" for up to 110 s; the stall
+  // watchdog flags it sooner, and Kuma is asked at once.
+  const stalled = useHomeserverStall(mx);
+  const setStalled = useSetAtom(homeserverStalledAtom);
+  useEffect(() => setStalled(stalled), [setStalled, stalled]);
+  const syncLost = lostAt !== null || stalled;
   const status = useServerStatus(page, syncLost);
   const setAtom = useSetAtom(serverStatusAtom);
   useEffect(() => setAtom({ status, syncLost, lostAt }), [setAtom, status, syncLost, lostAt]);
@@ -161,14 +179,22 @@ export function ServerStatusFeature({ mx }: { mx: MatrixClient }) {
 /** The logged-in app's status strip. */
 export function ServerStatusBanner() {
   const { status, syncLost } = useAtomValue(serverStatusAtom);
-  const [dismissed, dismissKey] = useAtom(statusDismissedAtom);
+  const [dismissed, setDismissed] = useState(readDismissed);
   const banner = pickBanner(status, {
     syncLost,
     where: 'client',
     dismissed: new Set(dismissed),
   });
   if (!banner) return null;
-  const dismiss = () => dismissKey(banner.key);
+  const dismiss = () => {
+    const next = [...dismissed.filter((k) => k !== banner.key), banner.key].slice(-DISMISS_MAX);
+    try {
+      localStorage.setItem(DISMISS_KEY, JSON.stringify(next));
+    } catch {
+      // storage unavailable — dismissed for this session only
+    }
+    setDismissed(next);
+  };
   return <StatusStrip banner={banner} onDismiss={dismiss} />;
 }
 

src/app/hooks/useHomeserverStall.test.ts

@@ -0,0 +1,16 @@
+import { test } from 'node:test';
+import assert from 'node:assert/strict';
+import { SyncState } from 'matrix-js-sdk';
+import { STALL_AFTER_MS, syncOverdue } from './useHomeserverStall';
+
+test('syncOverdue: only while SYNCING and past the stall threshold (#317)', () => {
+  const now = 1_000_000;
+  assert.equal(syncOverdue(now - STALL_AFTER_MS - 1, now, SyncState.Syncing), true);
+  assert.equal(syncOverdue(now - 20_000, now, SyncState.Syncing), false, 'a normal long-poll');
+  assert.equal(
+    syncOverdue(now - 120_000, now, SyncState.Reconnecting),
+    false,
+    'SyncStatus already shows that',
+  );
+  assert.equal(syncOverdue(now - 120_000, now, null), false);
+});

src/app/hooks/useHomeserverStall.ts

@@ -0,0 +1,57 @@
+import { useEffect, useState } from 'react';
+import { atom } from 'jotai';
+import { ClientEvent, MatrixClient, SyncState } from 'matrix-js-sdk';
+
+/**
+ * [Gitea #317] A hung homeserver (accepts connections, never answers) went
+ * unnoticed for up to 110 s: the SDK only gives up on a /sync long-poll after
+ * pollTimeout (30 s) + an 80 s buffer, and until then the sync state stays
+ * SYNCING, so no strip showed and Kuma wasn't asked. This watchdog notices a
+ * /sync that's overdue and confirms with a short probe of /versions.
+ */
+export const STALL_AFTER_MS = 45_000; // a healthy long-poll returns within ~30 s
+const CHECK_EVERY_MS = 10_000;
+const PROBE_TIMEOUT_MS = 5_000;
+
+export const homeserverStalledAtom = atom(false);
+
+/** Pure: should we probe, given the last sync response time? */
+export const syncOverdue = (lastSyncAt: number, now: number, state: SyncState | null): boolean =>
+  state === SyncState.Syncing && now - lastSyncAt > STALL_AFTER_MS;
+
+export function useHomeserverStall(mx: MatrixClient): boolean {
+  const [stalled, setStalled] = useState(false);
+  useEffect(() => {
+    let lastSyncAt = Date.now();
+    let probing = false;
+    const onSync = () => {
+      lastSyncAt = Date.now();
+      setStalled(false);
+    };
+    const check = async () => {
+      if (probing || !syncOverdue(lastSyncAt, Date.now(), mx.getSyncState())) return;
+      probing = true;
+      let answered = false;
+      try {
+        const res = await fetch(`${mx.getHomeserverUrl()}/_matrix/client/versions`, {
+          cache: 'no-store',
+          signal: AbortSignal.timeout(PROBE_TIMEOUT_MS),
+        });
+        answered = res.ok;
+      } catch {
+        answered = false;
+      }
+      probing = false;
+      // Still overdue and the server didn't answer a trivial request: stalled.
+      if (!answered && syncOverdue(lastSyncAt, Date.now(), mx.getSyncState())) setStalled(true);
+      if (answered) setStalled(false);
+    };
+    mx.on(ClientEvent.Sync, onSync);
+    const id = setInterval(check, CHECK_EVERY_MS);
+    return () => {
+      mx.removeListener(ClientEvent.Sync, onSync);
+      clearInterval(id);
+    };
+  }, [mx]);
+  return stalled;
+}

src/app/pages/client/SyncStatus.tsx

@@ -6,6 +6,7 @@ import { useSyncState } from '../../hooks/useSyncState';
 import { ContainerColor } from '../../styles/ContainerColor.css';
 import { serverStatusAtom } from '../../state/serverStatus';
 import { serverConfirmedUpSince } from '../../utils/kumaStatus';
+import { homeserverStalledAtom } from '../../hooks/useHomeserverStall';
 
 type StateData = {
   current: SyncState | null;
@@ -17,6 +18,7 @@ type SyncStatusProps = {
 };
 export function SyncStatus({ mx }: SyncStatusProps) {
   const serverStatus = useAtomValue(serverStatusAtom);
+  const stalled = useAtomValue(homeserverStalledAtom);
   const [stateData, setStateData] = useState<StateData>({
     current: null,
     previous: undefined,
@@ -34,6 +36,35 @@ export function SyncStatus({ mx }: SyncStatusProps) {
     }, []),
   );
 
+  // [Gitea #124] When Kuma confirms the homeserver is down, the status banner
+  // says so ("Lotus Chat's server is down…") instead of "Connection Lost!".
+  const serverDown = serverStatus.status?.homeserver === 'down';
+  // Only blame the user's connection when Kuma has checked the server since
+  // the drop and it was fine (a stale "up" from before proves nothing).
+  const serverUp = serverConfirmedUpSince(serverStatus.status, serverStatus.lostAt);
+
+  // [Gitea #317] Sync still says SYNCING, but the server has stopped answering.
+  if (stalled && stateData.current === SyncState.Syncing && !serverDown) {
+    return (
+      <Box direction="Column" shrink="No">
+        <Box
+          className={ContainerColor({ variant: 'Warning' })}
+          style={{ padding: `${config.space.S100} 0` }}
+          alignItems="Center"
+          justifyContent="Center"
+          role="status"
+          aria-live="polite"
+        >
+          <Text size="L400">
+            Lotus Chat&apos;s server isn&apos;t responding. Messages will send when it&apos;s back.
+            Still trying…
+          </Text>
+        </Box>
+        <Line variant="Warning" size="300" />
+      </Box>
+    );
+  }
+
   if (
     (stateData.current === SyncState.Prepared ||
       stateData.current === SyncState.Syncing ||
@@ -57,13 +88,6 @@ export function SyncStatus({ mx }: SyncStatusProps) {
     );
   }
 
-  // [Gitea #124] When Kuma confirms the homeserver is down, the status banner
-  // says so ("Lotus Chat's server is down…") instead of "Connection Lost!".
-  const serverDown = serverStatus.status?.homeserver === 'down';
-  // Only blame the user's connection when Kuma has checked the server since
-  // the drop and it was fine (a stale "up" from before proves nothing).
-  const serverUp = serverConfirmedUpSince(serverStatus.status, serverStatus.lostAt);
-
   if (stateData.current === SyncState.Reconnecting && !serverDown) {
     return (
       <Box direction="Column" shrink="No">

src/app/state/serverStatus.ts

@@ -16,35 +16,3 @@ export const serverStatusAtom = atom<{
   syncLost: false,
   lostAt: null,
 });
-
-const DISMISS_KEY = 'lotus-status-banner-dismissed';
-const DISMISS_MAX = 50;
-
-const readDismissed = (): string[] => {
-  try {
-    const v = JSON.parse(localStorage.getItem(DISMISS_KEY) ?? '[]');
-    return Array.isArray(v) ? v.filter((k): k is string => typeof k === 'string') : [];
-  } catch {
-    return [];
-  }
-};
-
-const baseDismissedAtom = atom<string[]>(readDismissed());
-
-/**
- * [Gitea #307] Status strips the user dismissed (by banner key), shared so
- * the security nudge knows a dismissed announcement isn't on screen.
- * Persisted to localStorage; session-only if storage is unavailable.
- */
-export const statusDismissedAtom = atom(
-  (get) => get(baseDismissedAtom),
-  (get, set, key: string) => {
-    const next = [...get(baseDismissedAtom).filter((k) => k !== key), key].slice(-DISMISS_MAX);
-    try {
-      localStorage.setItem(DISMISS_KEY, JSON.stringify(next));
-    } catch {
-      // storage unavailable — dismissed for this session only
-    }
-    set(baseDismissedAtom, next);
-  },
-);