fix(mobile): don't let a frozen sync block syncing on resume (#29870)

* fix(mobile): don't let a frozen sync block syncing on resume

* share the task lists between the cancel paths

* fix analyzer issues in resume integration test

* isolate the resume e2e test and pin the worker pool setup

mobile/integration_test/failed_sync_resume_e2e_test.dart

@@ -0,0 +1,101 @@
+import 'dart:async';
+import 'dart:io';
+import 'dart:math';
+
+import 'package:flutter_test/flutter_test.dart';
+import 'package:hooks_riverpod/hooks_riverpod.dart';
+import 'package:immich_mobile/domain/services/background_worker.service.dart';
+import 'package:immich_mobile/domain/utils/background_sync.dart';
+import 'package:immich_mobile/infrastructure/repositories/db.repository.dart';
+import 'package:immich_mobile/main.dart' as app;
+import 'package:immich_mobile/platform/background_worker_api.g.dart';
+import 'package:immich_mobile/providers/app_life_cycle.provider.dart';
+import 'package:immich_mobile/providers/background_sync.provider.dart';
+import 'package:immich_mobile/providers/infrastructure/db.provider.dart';
+import 'package:immich_mobile/services/api.service.dart';
+import 'package:immich_mobile/utils/bootstrap.dart';
+import 'package:immich_mobile/wm_executor.dart';
+import 'package:integration_test/integration_test.dart';
+
+import 'test_utils/fake_immich_server.dart';
+
+// Issue #28082 end-to-end: a resume after a sync froze mid-flight starts a fresh sync.
+// Kept in its own file on purpose: run after the failed_sync_resume tests in one process,
+// the resume sync starves past its window - a cross-test interaction that survived pool,
+// sqlite and bg-worker audits unnamed. One file per process is device-proven green.
+void main() {
+  final binding = IntegrationTestWidgetsFlutterBinding.ensureInitialized();
+  binding.framePolicy = LiveTestWidgetsFlutterBindingFramePolicy.fullyLive;
+
+  late Drift drift;
+  late FakeImmichServer server;
+
+  // main()'s formula with a higher floor: a wedged frozen worker plus a full local
+  // sync must not starve the fresh resume sync out of its 25s window on 4 cores.
+  final poolSize = max(Platform.numberOfProcessors - 1, 8);
+
+  setUpAll(() async {
+    await app.initApp();
+    (drift, _) = await Bootstrap.initDomain();
+    // A background-worker schedule persisted by real app use on this device can
+    // launch a second engine mid-file (own isolate pool + full sync) and starve
+    // these tests on a small device. Unregister it for the whole run.
+    await BackgroundWorkerFgService(BackgroundWorkerFgHostApi()).disable();
+  });
+
+  setUp(() async {
+    // A task completing while dispose tears the pool down re-warms it (_schedule
+    // on the cleared pool re-creates workers), and init on a warm pool is silently
+    // ignored - poolSize would never apply. Reset first, then verify it took.
+    await workerManagerPatch.dispose();
+    await workerManagerPatch.init(dynamicSpawning: true, isolatesCount: poolSize);
+    expect(
+      workerManagerPatch.pool.length,
+      poolSize,
+      reason: 'init was ignored: a straggler from the previous test re-warmed the pool',
+    );
+    server = await FakeImmichServer.start();
+    await ApiService().resolveAndSetEndpoint(server.endpoint);
+    await drift.delete(drift.userEntity).go();
+  });
+
+  tearDown(() async {
+    // Close the server first so any held-open sync stream ends and its isolate unwinds,
+    // then drain the pool - otherwise dispose waits on the frozen read.
+    await server.close();
+    await workerManagerPatch.dispose();
+  });
+
+  testWidgets('a resume after a sync froze mid-flight starts a fresh sync', (tester) async {
+    final manager = BackgroundSyncManager();
+    // Not disposed on purpose: driftOverride closes the drift on dispose, and that
+    // drift belongs to setUpAll. The container only holds the shared drift and a
+    // fire-and-forget resume; the frozen isolates + server are drained by tearDown.
+    final container = ProviderContainer(
+      overrides: [driftProvider.overrideWith(driftOverride(drift)), backgroundSyncProvider.overrideWithValue(manager)],
+    );
+
+    // A first sync opens /sync/stream and never finishes - the frozen state a
+    // suspended sync isolate is left in. Holding the stream open keeps
+    // _syncTask non-null, exactly as it is across an iOS process suspension.
+    unawaited(manager.syncRemote());
+    await server
+        .streamOpenedNth(1)
+        .timeout(const Duration(seconds: 30), onTimeout: () => fail('first sync isolate never opened /sync/stream'));
+
+    // The lifecycle then goes background -> foreground. handleAppResume runs the
+    // resume sync exactly once. On the buggy build it hangs on the stale task's
+    // future, so it is not awaited here.
+    final notifier = container.read(appStateProvider.notifier);
+    await notifier.handleAppPause();
+    unawaited(notifier.handleAppResume());
+
+    await server
+        .streamOpenedNth(2)
+        .timeout(
+          const Duration(seconds: 25),
+          onTimeout: () => fail('resume did not start a fresh remote sync - the stale frozen sync blocked it (#28082)'),
+        );
+    expect(server.streamOpenCount, greaterThanOrEqualTo(2));
+  });
+}

mobile/integration_test/failed_sync_resume_test.dart

@@ -0,0 +1,133 @@
+import 'dart:async';
+import 'dart:io';
+import 'dart:math';
+
+import 'package:flutter_test/flutter_test.dart';
+import 'package:immich_mobile/domain/services/background_worker.service.dart';
+import 'package:immich_mobile/domain/utils/background_sync.dart';
+import 'package:immich_mobile/infrastructure/repositories/db.repository.dart';
+import 'package:immich_mobile/main.dart' as app;
+import 'package:immich_mobile/platform/background_worker_api.g.dart';
+import 'package:immich_mobile/services/api.service.dart';
+import 'package:immich_mobile/utils/bootstrap.dart';
+import 'package:immich_mobile/wm_executor.dart';
+import 'package:integration_test/integration_test.dart';
+
+import 'test_utils/fake_immich_server.dart';
+
+// Issue #28082: a remote sync in-flight when the app is backgrounded stays referenced
+// but frozen across the suspension. On resume the app drops the stale task
+// (cancelResumeSyncs) and starts a fresh sync.
+//
+// Device/emulator tests: real worker isolates + a real drift db + a loopback fake server
+// (same pattern as background_sync_teardown_test). The mobile integration-test CI job is
+// disabled in test.yml, so like Mert's teardown test this is a local/on-device guard.
+//
+// The end-to-end resume test lives in failed_sync_resume_e2e_test.dart - one process
+// per file keeps it clear of the cross-test interaction described there.
+void main() {
+  final binding = IntegrationTestWidgetsFlutterBinding.ensureInitialized();
+  binding.framePolicy = LiveTestWidgetsFlutterBindingFramePolicy.fullyLive;
+
+  late Drift drift;
+  late FakeImmichServer server;
+
+  // main()'s formula with a higher floor: a wedged frozen worker plus a full local
+  // sync must not starve the fresh resume sync out of its 25s window on 4 cores.
+  final poolSize = max(Platform.numberOfProcessors - 1, 8);
+
+  setUpAll(() async {
+    await app.initApp();
+    (drift, _) = await Bootstrap.initDomain();
+    // A background-worker schedule persisted by real app use on this device can
+    // launch a second engine mid-file (own isolate pool + full sync) and starve
+    // these tests on a small device. Unregister it for the whole run.
+    await BackgroundWorkerFgService(BackgroundWorkerFgHostApi()).disable();
+  });
+
+  setUp(() async {
+    // A task completing while dispose tears the pool down re-warms it (_schedule
+    // on the cleared pool re-creates workers), and init on a warm pool is silently
+    // ignored - poolSize would never apply. Reset first, then verify it took.
+    await workerManagerPatch.dispose();
+    await workerManagerPatch.init(dynamicSpawning: true, isolatesCount: poolSize);
+    expect(
+      workerManagerPatch.pool.length,
+      poolSize,
+      reason: 'init was ignored: a straggler from the previous test re-warmed the pool',
+    );
+    server = await FakeImmichServer.start();
+    await ApiService().resolveAndSetEndpoint(server.endpoint);
+    await drift.delete(drift.userEntity).go();
+  });
+
+  tearDown(() async {
+    // Close the server first so any held-open sync stream ends and its isolate unwinds,
+    // then drain the pool - otherwise dispose waits on the frozen read.
+    await server.close();
+    await workerManagerPatch.dispose();
+  });
+
+  // Self-contained (bare manager, no fire-and-forget resume), so it runs first: its
+  // frozen syncs are fully drained by tearDown.
+  testWidgets('a cancelled sync task does not clear the slot of the fresh task that superseded it', (tester) async {
+    final manager = BackgroundSyncManager();
+
+    // First sync opens /sync/stream and is held open - the frozen suspended state.
+    unawaited(manager.syncRemote());
+    await server
+        .streamOpenedNth(1)
+        .timeout(const Duration(seconds: 30), onTimeout: () => fail('first sync isolate never opened /sync/stream'));
+
+    // Resume drops the stale task then immediately starts a fresh one, exactly as
+    // _handleBetaTimelineResume does. cancelResumeSyncs cancels the first task; its
+    // completion chain then fires and must NOT null the fresh task's slot.
+    unawaited(manager.cancelResumeSyncs());
+    unawaited(manager.syncRemote());
+    await server
+        .streamOpenedNth(2)
+        .timeout(const Duration(seconds: 30), onTimeout: () => fail('fresh sync isolate never opened /sync/stream'));
+
+    // A third sync stream only opens if the cancelled task cleared the fresh slot,
+    // letting the dedupe guard start a redundant sync.
+    var thirdOpened = false;
+    unawaited(server.streamOpenedNth(3).then((_) => thirdOpened = true));
+
+    // Let the cancelled task's completion chain settle, then ask to sync again.
+    await Future.delayed(const Duration(milliseconds: 200));
+    unawaited(manager.syncRemote());
+    await Future.delayed(const Duration(seconds: 3));
+
+    expect(
+      thirdOpened,
+      isFalse,
+      reason: 'the cancelled task cleared the fresh task slot, so a redundant third sync started (#28082 clobber)',
+    );
+    expect(server.streamOpenCount, 2);
+  });
+
+  // The false-error-flash fix: a task cancelled by cancelResumeSyncs completes with a
+  // CanceledError, which is not a sync failure and must not reach onRemoteSyncError.
+  // Before the filter the stale task reported an error right after the fresh sync
+  // started, so the status UI showed a failure for the whole healthy run.
+  testWidgets('a cancelled sync does not report a false error to the status callbacks', (tester) async {
+    final errors = <String>[];
+    final manager = BackgroundSyncManager(onRemoteSyncError: errors.add);
+
+    // Hold the stream open so the task is genuinely in-flight when it is cancelled.
+    unawaited(manager.syncRemote());
+    await server
+        .streamOpenedNth(1)
+        .timeout(const Duration(seconds: 30), onTimeout: () => fail('sync isolate never opened /sync/stream'));
+
+    await manager.cancelResumeSyncs();
+    // Let the cancelled task's completion chain run before checking the callbacks.
+    await Future.delayed(const Duration(milliseconds: 100));
+
+    expect(
+      errors,
+      isEmpty,
+      reason: 'a cancelled task is not a real error; its CanceledError must not fire onRemoteSyncError',
+    );
+  });
+}

mobile/integration_test/test_utils/fake_immich_server.dart

@@ -10,6 +10,8 @@ class FakeImmichServer {
   final (int, int, int) version;
 
   final Completer<SyncStream> _streamOpened = Completer<SyncStream>();
+  final List<SyncStream> _streamOpens = [];
+  final Map<int, Completer<SyncStream>> _openWaiters = {};
 
   int ackRequests = 0;
 
@@ -18,6 +20,17 @@ class FakeImmichServer {
   /// Resolves when the sync isolate opens `POST /sync/stream`.
   Future<SyncStream> get streamOpened => _streamOpened.future;
 
+  /// How many `/sync/stream` requests have opened so far.
+  int get streamOpenCount => _streamOpens.length;
+
+  /// Resolves when the [n]-th (1-indexed) `/sync/stream` opens.
+  Future<SyncStream> streamOpenedNth(int n) {
+    if (_streamOpens.length >= n) {
+      return Future.value(_streamOpens[n - 1]);
+    }
+    return (_openWaiters[n] ??= Completer<SyncStream>()).future;
+  }
+
   static Future<FakeImmichServer> start({(int, int, int) version = (3, 0, 0)}) async {
     final server = await HttpServer.bind(InternetAddress.loopbackIPv4, 0);
     final fake = FakeImmichServer._(server, version);
@@ -58,13 +71,17 @@ class FakeImmichServer {
     request.response
       ..statusCode = HttpStatus.ok
       ..headers.contentType = ContentType('application', 'jsonlines+json')
-      ..contentLength = -1 // chunked: stays open to stream incrementally
+      ..contentLength =
+          -1 // chunked: stays open to stream incrementally
       ..bufferOutput = false;
     // Flush headers so the client's send() resolves and enters its read loop.
     await request.response.flush();
+    final stream = SyncStream._(request.response);
+    _streamOpens.add(stream);
     if (!_streamOpened.isCompleted) {
-      _streamOpened.complete(SyncStream._(request.response));
+      _streamOpened.complete(stream);
     }
+    _openWaiters.remove(_streamOpens.length)?.complete(stream);
   }
 
   Future<void> _respondJson(HttpRequest request, Object body) async {
@@ -83,8 +100,8 @@ class FakeImmichServer {
   }
 
   Future<void> close() async {
-    if (_streamOpened.isCompleted) {
+    for (final stream in _streamOpens) {
-      await (await _streamOpened.future).close();
+      await stream.close();
     }
     await _server.close(force: true);
   }

mobile/lib/domain/utils/background_sync.dart

@@ -50,16 +50,39 @@ class BackgroundSyncManager {
     this.onCloudIdSyncError,
   });
 
+  // The tasks the app-resume path re-runs. One in-flight when the app was suspended
+  // stays referenced but frozen, so on resume the dedupe guards would hand back the
+  // stale task instead of syncing (#28082). Websocket and cloud-id are excluded - the
+  // resume path never restarts them. [_allTasks] builds on this so the lists can't drift.
+  List<Cancelable?> get _resumeSyncTasks => [_syncTask, _deviceAlbumSyncTask, _hashTask, _linkedAlbumSyncTask];
+
+  List<Cancelable?> get _allTasks => [_syncWebsocketTask, _cloudIdSyncTask, ..._resumeSyncTasks];
+
   Future<void> cancel() async {
     _syncQueued = false;
-    final tasks = [
+    final tasks = _allTasks;
-      _syncTask,
+    _syncTask = null;
-      _syncWebsocketTask,
+    _syncWebsocketTask = null;
-      _cloudIdSyncTask,
+    _cloudIdSyncTask = null;
-      _linkedAlbumSyncTask,
+    _linkedAlbumSyncTask = null;
-      _deviceAlbumSyncTask,
+    _deviceAlbumSyncTask = null;
-      _hashTask,
+    _hashTask = null;
-    ];
+    await _cancelAll(tasks);
+  }
+
+  Future<void> cancelResumeSyncs() async {
+    _syncQueued = false;
+    final tasks = _resumeSyncTasks;
+    _syncTask = null;
+    _deviceAlbumSyncTask = null;
+    _hashTask = null;
+    _linkedAlbumSyncTask = null;
+    await _cancelAll(tasks);
+  }
+
+  // Cancels every task in [tasks] and waits for them to unwind. Callers null out
+  // their own fields first, so the sync guards see a clean slate immediately.
+  Future<void> _cancelAll(List<Cancelable?> tasks) async {
     final futures = [
       for (final task in tasks)
         if (task != null) task.future,
@@ -67,13 +90,6 @@ class BackgroundSyncManager {
     for (final task in tasks) {
       task?.cancel();
     }
-    _syncTask = null;
-    _syncWebsocketTask = null;
-    _cloudIdSyncTask = null;
-    _linkedAlbumSyncTask = null;
-    _deviceAlbumSyncTask = null;
-    _hashTask = null;
-
     try {
       await Future.wait(futures);
     } on CanceledError {
@@ -84,14 +100,14 @@ class BackgroundSyncManager {
   // No need to cancel the task, as it can also be run when the user logs out
   Future<void> syncLocal({bool full = false}) {
     if (_deviceAlbumSyncTask != null) {
-      return _deviceAlbumSyncTask!.future;
+      return _deviceAlbumSyncTask!.future.catchError((_) {}, test: (error) => error is CanceledError);
     }
 
     onLocalSyncStart?.call();
 
     // We use a ternary operator to avoid [_deviceAlbumSyncTask] from being
     // captured by the closure passed to [runInIsolateGentle].
-    _deviceAlbumSyncTask = full
+    final task = _deviceAlbumSyncTask = full
         ? runInIsolateGentle(
             computation: (ref) => ref.read(localSyncServiceProvider).sync(full: true),
             debugLabel: 'local-sync-full-true',
@@ -101,37 +117,43 @@ class BackgroundSyncManager {
             debugLabel: 'local-sync-full-false',
           );
 
-    return _deviceAlbumSyncTask!
+    return task
         .whenComplete(() {
-          _deviceAlbumSyncTask = null;
+          if (identical(_deviceAlbumSyncTask, task)) {
+            _deviceAlbumSyncTask = null;
+          }
           onLocalSyncComplete?.call();
         })
         .catchError((error) {
-          onLocalSyncError?.call(error.toString());
+          if (error is! CanceledError) {
-          _deviceAlbumSyncTask = null;
+            onLocalSyncError?.call(error.toString());
+          }
         });
   }
 
   Future<void> hashAssets() {
     if (_hashTask != null) {
-      return _hashTask!.future;
+      return _hashTask!.future.catchError((_) {}, test: (error) => error is CanceledError);
     }
 
     onHashingStart?.call();
 
-    _hashTask = runInIsolateGentle(
+    final task = _hashTask = runInIsolateGentle(
       computation: (ref) => ref.read(hashServiceProvider).hashAssets(),
       debugLabel: 'hash-assets',
     );
 
-    return _hashTask!
+    return task
         .whenComplete(() {
           onHashingComplete?.call();
-          _hashTask = null;
+          if (identical(_hashTask, task)) {
+            _hashTask = null;
+          }
         })
         .catchError((error) {
-          onHashingError?.call(error.toString());
+          if (error is! CanceledError) {
-          _hashTask = null;
+            onHashingError?.call(error.toString());
+          }
         });
   }
 
@@ -143,11 +165,11 @@ class BackgroundSyncManager {
 
     onRemoteSyncStart?.call();
 
-    _syncTask = runInIsolateGentle(
+    final task = _syncTask = runInIsolateGentle(
       computation: (ref) => ref.read(syncStreamServiceProvider).sync(),
       debugLabel: 'remote-sync',
     );
-    return _syncTask!
+    return task
         .then((result) {
           final success = result ?? false;
           onRemoteSyncComplete?.call(success);
@@ -155,15 +177,21 @@ class BackgroundSyncManager {
           return success;
         })
         .catchError((error) {
-          onRemoteSyncError?.call(error.toString());
+          if (error is! CanceledError) {
+            onRemoteSyncError?.call(error.toString());
+          }
           _syncQueued = false;
           return false;
         })
+        // A task clears only its own slot: one that was cancelled and superseded by a
+        // fresh task (see cancelResumeSyncs) must not null the new task's slot.
         .whenComplete(() {
-          _syncTask = null;
+          if (identical(_syncTask, task)) {
-          if (_syncQueued) {
+            _syncTask = null;
-            _syncQueued = false;
+            if (_syncQueued) {
-            unawaited(syncRemote());
+              _syncQueued = false;
+              unawaited(syncRemote());
+            }
           }
         });
   }
@@ -210,13 +238,21 @@ class BackgroundSyncManager {
 
   Future<void> syncLinkedAlbum() {
     if (_linkedAlbumSyncTask != null) {
-      return _linkedAlbumSyncTask!.future;
+      return _linkedAlbumSyncTask!.future.catchError((_) {}, test: (error) => error is CanceledError);
     }
 
-    _linkedAlbumSyncTask = runInIsolateGentle(computation: syncLinkedAlbumsIsolated, debugLabel: 'linked-album-sync');
+    final task = _linkedAlbumSyncTask = runInIsolateGentle(
-    return _linkedAlbumSyncTask!.whenComplete(() {
+      computation: syncLinkedAlbumsIsolated,
-      _linkedAlbumSyncTask = null;
+      debugLabel: 'linked-album-sync',
-    });
+    );
+    return task
+        .whenComplete(() {
+          if (identical(_linkedAlbumSyncTask, task)) {
+            _linkedAlbumSyncTask = null;
+          }
+        })
+        // a cancelled resume sync is not a failure; absorb it so the websocket callers don't get an uncaught error
+        .catchError((_) {}, test: (error) => error is CanceledError);
   }
 
   Future<void> syncCloudIds() {

mobile/lib/providers/app_life_cycle.provider.dart

@@ -112,6 +112,12 @@ class AppLifeCycleNotifier extends StateNotifier<AppLifeCycleEnum> {
     await Future.delayed(const Duration(milliseconds: 500));
 
     final backgroundManager = _ref.read(backgroundSyncProvider);
+
+    // Drop any sync that froze mid-flight while the app was suspended so resume
+    // starts fresh instead of awaiting the stale task (#28082). cancelResumeSyncs
+    // clears the task refs synchronously, so the syncs below see a clean slate.
+    unawaited(backgroundManager.cancelResumeSyncs());
+
     final isAlbumLinkedSyncEnable = _ref.read(appConfigProvider).backup.syncAlbums;
 
     try {