Merge pull request 'fix(status): show a strip when the homeserver hangs, not 110 s later (#317)' (#380) from fix/stall-watchdog into lotus

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/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">