NormalizedCache.merge(records:) in conformance InMemoryNormalizedCache

[AlexPan1992]

Summary

InMemoryNormalizedCache is not thread-safe when multiple GraphQL queries/mutations execute concurrently. When multiple threads attempt to merge records into the cache simultaneously, it causes memory corruption crashes with the error: "___BUG_IN_CLIENT_OF_LIBMALLOC_POINTER_BEING_FREED_WAS_NOT_ALLOCATED".

The crash occurs in RecordSet.merge(record:) when Dictionary operations are performed concurrently without proper synchronization.

Version

2.0.4

Steps to reproduce the behavior

1. Create an ApolloStore with default InMemoryNormalizedCache
2. Execute multiple GraphQL queries/mutations concurrently from different threads (e.g., using async/await with different Task contexts)
3. These operations will trigger cache merge operations simultaneously
4. The app will crash with memory corruption error

Logs

Crashed: com.apple.root.user-initiated-qos.cooperative
0  libsystem_kernel.dylib         0xb0cc __pthread_kill + 8
1  libsystem_pthread.dylib        0x7810 pthread_kill + 268
2  libsystem_c.dylib              0x77f64 abort + 124
3  libsystem_malloc.dylib         0x15a64 malloc_vreport + 892
4  libsystem_malloc.dylib         0x156dc malloc_report + 64
5  libsystem_malloc.dylib         0x9714 ___BUG_IN_CLIENT_OF_LIBMALLOC_POINTER_BEING_FREED_WAS_NOT_ALLOCATED + 76
6  libswiftCore.dylib             0x27e60 <redacted> + 56
7  libswiftCore.dylib             0x28a7c <redacted> + 152
8  MyApp                  0x39f7fc specialized _NativeDictionary.copy() + 4305451004
9  MyApp                  0x39edf4 specialized _NativeDictionary.setValue(_:forKey:isUnique:) + 4305448436
10 MyApp                  0x3a49c4 RecordSet.merge(record:) + 4305471940
11 MyApp                  0x3a375c RecordSet.merge(records:) + 4305467228
12 MyApp                  0x39e6d0 protocol witness for NormalizedCache.merge(records:) in conformance InMemoryNormalizedCache + 4305446608
13 libswift_Concurrency.dylib     0x62d58 <redacted> + 288
14 libswift_Concurrency.dylib     0x641cc <redacted> + 156
15 libdispatch.dylib              0x13f48 _dispatch_root_queue_drain + 364
16 libdispatch.dylib              0x146fc _dispatch_worker_thread2 + 180
17 libsystem_pthread.dylib        0x137c _pthread_wqthread + 232
18 libsystem_pthread.dylib        0x8c0 start_wqthread + 8

Anything else?

No response

[calvincestari]

Thanks for creating the issue @AlexPan1992. It would be extremely helpful if you could you put together a standalone reproduction case that demonstrates the crash.

Do you know if this crash occurs with versions 2.0.3 or below? I ask because 2.0.4 made changes to the read/write locking mechanism in ApolloStore so it'll be helpful to know whether this is a newly introduced issue or something pre-existing?

[AlexPan1992]

Hi @calvincestari, appreciate your help.

2.0.4 is the first version I released to Production, so I haven’t used 2.0.3 before.

However, I created a small unit test (RocketReserverTests.swift) to verify this issue and was able to reproduce the behavior you described.

With the exact same test, 2.0.4 results in a crash, while 2.0.3 runs without any issues.

Here is the example I used:

2.0.4 (crash): AlexPan1992/iOSTutorial@619ebff

2.0.3 (successfully): AlexPan1992/iOSTutorial@b259c5c

[calvincestari]

That's awesome, thanks @AlexPan1992!

[AnthonyMDev]

A PR with a fix is up. We'll get this shipped out soon.

[AnthonyMDev]

Thanks so much @AlexPan1992 for the excellent reproduction case!

[github-actions]

Do you have any feedback for the maintainers? Please tell us by taking a one-minute survey. Your responses will help us understand Apollo iOS usage and allow us to serve you better.

[markusfassbender-lb]

Thanks for this fix! @AnthonyMDev Is there any chance to get this released soon as we also experience it in our app?

[calvincestari]

I'll queue up a release for today, I think there are a couple merges waiting to go out.