Android: `Unable to Enter Isolate : Multiple mutators entering an isolate / Dart VM is shutting down`

[derolf]

I am decoding two MJPEG-streams in parallel and make heavy usage of MemoryImage (Flutter), decoding about 10-20 images per seconds.

After a few seconds, the app crashes with

/Dart    (14933): ../../third_party/dart/runtime/vm/dart_api_impl.cc: 1662: error: Unable to Enter Isolate : Multiple mutators entering an isolate / Dart VM is shutting down
E/DartVM  (14933): version=2.12.0-29.7.beta (beta) (Fri Nov 13 11:35:21 2020 +0100) on "android_arm"
E/DartVM  (14933): pid=14933, thread=14961, isolate_group=(nil)(0x0), isolate=(nil)(0x0)
E/DartVM  (14933): isolate_instructions=0, vm_instructions=a41c7240
E/DartVM  (14933):   pc 0xa427b025 fp 0xa2b7dab0 libflutter.so+0x13fd025
E/DartVM  (14933): -- End of DumpStackTrace
F/libc    (14933): Fatal signal 6 (SIGABRT), code -6 in tid 14961 (1.io)

Flutter doctor:

Doctor summary (to see all details, run flutter doctor -v):
[✓] Flutter (Channel dev, 1.24.0-10.1.pre, on Mac OS X 10.15.7 19H2 darwin-x64, locale en-DE)
[✓] Android toolchain - develop for Android devices (Android SDK version 30.0.2)
[✓] Xcode - develop for iOS and macOS (Xcode 12.2)
[✓] Android Studio (version 3.6)
[✓] VS Code (version 1.51.1)
[✓] Connected device (2 available)
• No issues found!

Same crash at the same location also with

[✓] Flutter (Channel beta, 1.24.0-10.2.pre, on Mac OS X 10.15.7 19H2 darwin-x64, locale en-DE)

[derolf]

Related to flutter/flutter#52576

[ravenblackx]

We have also observed this kind of crash, not even doing anything particularly demanding.

I'm only observing it from logs so I don't have a lot of details, but the crash occurred moments after downloading a 67kb png image (download completed at 01:39:26.507, and "Multiple mutators" error was the next line in the log at 01:39:27.133) so it seems very likely to be the same kind of image decoding error.

[derolf]

Dart team (@franklinyow). Anything we can do to help finding the root cause? Any small example to try on the target tablet?

[derolf]

I tried the following simple test on the target, but it didn't crash:

// @dart=2.11
import 'package:flutter/material.dart';
import 'package:flutter_test/flutter_test.dart';
import 'package:http/http.dart';

Future<void> main() async {
  WidgetsFlutterBinding.ensureInitialized();
  final buf = (await get('https://upload.wikimedia.org/wikipedia/en/a/a9/Example.jpg')).bodyBytes;
  for (;;) {
    final img = MemoryImage(buf);
    final status = await img.obtainCacheStatus(configuration: const ImageConfiguration(), handleError: (_, s) => print('$_ $s'));
    print(status);
  }
}

[franklinyow]

@a-siva Can you make sure I have the right area label? Thx

[ravenblackx]

I now think the problem may be in the network and/or file handling rather than the image decoding. (It's network in our case.)

flutter/flutter#72144 is related.

(derolf may be able to confirm or deny this suspicion by trying their code with the downloading but without the decoding?)

[derolf]

If I remove decoding and substitute it with just waiting (to generate pressure), it doesn't crash.

[derolf]

@franklinyow @a-siva How can I get a full backtrace?

*** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
Build fingerprint: 'google/cm_ford/ford:5.1.1/LMY48Y/525d8e95be:userdebug/test-keys'
Revision: '0'
ABI: 'arm'
pid: 13254, tid: 13304, name: 1.io  >>> com.example.home <<<
signal 6 (SIGABRT), code -6 (SI_TKILL), fault addr --------
    r0 00000000  r1 000033f8  r2 00000006  r3 00000000
    r4 a2b88db8  r5 00000006  r6 00000002  r7 0000010c
    r8 00000001  r9 a2b88b90  sl b6dee47d  fp b6df0fe4
    ip 000033f8  sp a2b88a68  lr b6e3d8f5  pc b6e610c4  cpsr 600d0010
backtrace:
    #00 pc 000380c4  /system/lib/libc.so (tgkill+12)
    #01 pc 000148f1  /system/lib/libc.so (pthread_kill+52)
    #02 pc 0001564b  /system/lib/libc.so (raise+10)
    #03 pc 00011cc5  /system/lib/libc.so (__libc_android_abort+36)
    #04 pc 0001005c  /system/lib/libc.so (abort+4)
    #05 pc 002c92c5  /data/app/com.example.home-2/lib/a

[a-siva]

@derolf is it possible to share your app code (just the decoding in parallel part) so we can have a reproduction of the crash and investigate.

[mkustermann]

This is very likely the same bug as flutter/flutter#72144 (the bug was introduced in flutter/engine#19843). It happens due to refcounted engine objects where destructor will try to delete persistent handle. This can only happen on mutator thread but due to refcounted object in C++ the last drop in refcount can happen on non-mutator thread, which leads to this bug.

Closing this issue, since it's almost surely the same.`