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

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

@@ -9,6 +9,7 @@ import { useServerStatus } from '../../hooks/useServerStatus';
 import { ScreenSize, useScreenSizeContext } from '../../hooks/useScreenSize';
 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;
@@ -163,7 +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]);

src/app/features/settings/account/Profile.tsx

@@ -67,7 +67,6 @@ import {
   makePresetId,
   normalizeLabel,
 } from '../../../utils/statusPresets';
-import { writeStatusExpiry } from '../../../utils/statusExpiry';
 
 type ProfileProps = {
   profile: UserProfile;
@@ -509,16 +508,12 @@ function ProfileStatus() {
       if (msg && delayMs > 0) {
         // Persist the expiry timestamp; the always-mounted StatusExpiryMonitor
         // (ClientNonUIFeatures) fires the auto-clear even when Settings is closed.
-        const expiresAt = Date.now() + delayMs;
-        localStorage.setItem(STATUS_EXPIRY_KEY(userId), String(expiresAt));
-        // [Gitea #300] …and on every other Lotus device, via account data.
-        writeStatusExpiry(mx, msg, expiresAt).catch(() => undefined);
+        localStorage.setItem(STATUS_EXPIRY_KEY(userId), String(Date.now() + delayMs));
       } else {
         localStorage.removeItem(STATUS_EXPIRY_KEY(userId));
-        writeStatusExpiry(mx, msg).catch(() => undefined);
       }
     },
-    [saveStatus, userId, mx],
+    [saveStatus, userId],
   );
 
   const handleSubmit: FormEventHandler<HTMLFormElement> = (evt) => {
@@ -558,7 +553,6 @@ function ProfileStatus() {
     setStatusMsg('');
     localStorage.removeItem(STATUS_MSG_KEY(userId));
     localStorage.removeItem(STATUS_EXPIRY_KEY(userId));
-    writeStatusExpiry(mx, '').catch(() => undefined);
     // Preserve the user's chosen presence when clearing the status message.
     mx.setPresence({
       presence: presenceStateFromSetting(presenceStatus, hidePresence),

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/ClientNonUIFeatures.tsx

@@ -91,7 +91,6 @@ import {
   shouldNotifyThreadReply,
   THREAD_NOTIFICATIONS_FALLBACK_BEHAVIOR,
 } from '../../utils/threadNotifications';
-import { readStatusExpiry, statusExpiryAction, writeStatusExpiry } from '../../utils/statusExpiry';
 
 // Grace period after the initial sync settles before invite notifications arm, so
 // the async invite-atom population lands first and isn't mistaken for new invites.
@@ -428,7 +427,12 @@ function StatusExpiryMonitor() {
     // permanently and the status stayed set forever. Retry the write (honouring
     // Retry-After), and only forget the status once the server has taken it.
     let clearing = false;
-    const clearStatus = () => {
+    const check = () => {
+      if (clearing) return;
+      const stored = localStorage.getItem(expiryKey);
+      if (!stored) return;
+      const ts = parseInt(stored, 10);
+      if (!ts || Date.now() < ts) return;
       clearing = true;
       setPresenceWithRetry(mx, {
         presence: presenceStateFromSetting(presenceStatusRef.current, hidePresenceRef.current),
@@ -438,31 +442,12 @@ function StatusExpiryMonitor() {
           noteLocalStatusWrite();
           localStorage.removeItem(msgKey);
           localStorage.removeItem(expiryKey);
-          writeStatusExpiry(mx, '').catch(() => undefined);
         })
         .catch(() => undefined)
         .finally(() => {
           clearing = false;
         });
     };
-    const check = () => {
-      if (clearing) return;
-      const stored = localStorage.getItem(expiryKey);
-      const ts = stored ? parseInt(stored, 10) : 0;
-      if (ts && Date.now() >= ts) {
-        clearStatus();
-        return;
-      }
-      // [Gitea #300] An expiry set on another device (account data): clear the
-      // status if it is still that one, otherwise just forget the expiry.
-      const action = statusExpiryAction(
-        readStatusExpiry(mx),
-        mx.getUser(userId)?.presenceStatusMsg,
-        Date.now(),
-      );
-      if (action === 'clear') clearStatus();
-      else if (action === 'forget') writeStatusExpiry(mx, '').catch(() => undefined);
-    };
 
     check();
     // Poll often enough that a rate-limited retry lands promptly, but not so

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/utils/statusExpiry.test.ts

@@ -1,22 +0,0 @@
-import { test } from 'node:test';
-import assert from 'node:assert/strict';
-import { parseStatusExpiry, statusExpiryAction } from './statusExpiry';
-
-test('parseStatusExpiry (#300)', () => {
-  assert.deepEqual(parseStatusExpiry({ status_msg: '🍔 Lunch', expires_at: 5 }), {
-    statusMsg: '🍔 Lunch',
-    expiresAt: 5,
-  });
-  assert.equal(parseStatusExpiry({}), undefined);
-  assert.equal(parseStatusExpiry({ status_msg: '', expires_at: 5 }), undefined);
-  assert.equal(parseStatusExpiry({ status_msg: 'x', expires_at: '5' }), undefined);
-});
-
-test('statusExpiryAction: clears only the status that was set to expire (#300)', () => {
-  const e = { statusMsg: '🍔 Lunch', expiresAt: 1000 };
-  assert.equal(statusExpiryAction(e, '🍔 Lunch', 999), 'none');
-  assert.equal(statusExpiryAction(e, '🍔 Lunch', 1000), 'clear');
-  assert.equal(statusExpiryAction(e, '🎮 Gaming', 2000), 'forget');
-  assert.equal(statusExpiryAction(e, '', 2000), 'forget');
-  assert.equal(statusExpiryAction(undefined, 'x', 2000), 'none');
-});

src/app/utils/statusExpiry.ts

@@ -1,46 +0,0 @@
-import { MatrixClient } from 'matrix-js-sdk';
-import { AccountDataEvent } from '../../types/matrix/accountData';
-import { getAccountData } from './room';
-
-/**
- * [Gitea #300] The status message's auto-clear time, in account data so every
- * Lotus device knows it. Before, only the device that set "clear after 1 hour"
- * (localStorage) ever cleared it, and only while it was open.
- */
-export type StatusExpiry = { statusMsg: string; expiresAt: number };
-
-/** Pure: parse `io.lotus.status_expiry` content. */
-export function parseStatusExpiry(content: unknown): StatusExpiry | undefined {
-  const c = content as { status_msg?: unknown; expires_at?: unknown } | undefined;
-  if (!c || typeof c.status_msg !== 'string' || typeof c.expires_at !== 'number') return undefined;
-  if (!c.status_msg || !Number.isFinite(c.expires_at)) return undefined;
-  return { statusMsg: c.status_msg, expiresAt: c.expires_at };
-}
-
-/**
- * Pure: what to do with a due expiry. Clear the server status only if it is
- * still the one that was set to expire; if it has since changed (e.g. on
- * another device), just forget the expiry.
- */
-export function statusExpiryAction(
-  entry: StatusExpiry | undefined,
-  currentStatusMsg: string | undefined,
-  now: number,
-): 'none' | 'clear' | 'forget' {
-  if (!entry || now < entry.expiresAt) return 'none';
-  return (currentStatusMsg ?? '') === entry.statusMsg ? 'clear' : 'forget';
-}
-
-export const readStatusExpiry = (mx: MatrixClient): StatusExpiry | undefined =>
-  parseStatusExpiry(getAccountData(mx, AccountDataEvent.LotusStatusExpiry)?.getContent());
-
-/** Record (or with no `expiresAt`, remove) the shared auto-clear time. */
-export async function writeStatusExpiry(
-  mx: MatrixClient,
-  statusMsg: string,
-  expiresAt?: number,
-): Promise<void> {
-  const content =
-    statusMsg && expiresAt !== undefined ? { status_msg: statusMsg, expires_at: expiresAt } : {};
-  await mx.setAccountData(AccountDataEvent.LotusStatusExpiry as never, content as never);
-}

src/types/matrix/accountData.ts

@@ -26,10 +26,6 @@ export enum AccountDataEvent {
   // [Gitea #104] Synced subset of the Lotus Settings object (utils/settingsSync).
   LotusSettings = 'io.lotus.settings',
 
-  // [Gitea #300] When the status message auto-clears: { status_msg, expires_at },
-  // so any Lotus device can clear it (not only the one that set it).
-  LotusStatusExpiry = 'io.lotus.status_expiry',
-
   SecretStorageDefaultKey = 'm.secret_storage.default_key',
 
   CrossSigningMaster = 'm.cross_signing.master',