frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix flaky SIGSEGV crash in test_profile_memory (#136304)

Browse Source 
Fixes https://github.com/pytorch/pytorch/issues/132331

We need another barrier here to ensure that the main thread doesn't stop the profiler while other threads are still using it (and crash).  I can reliably reproduce the issue with `pytest -v test/profiler/test_cpp_thread.py -k test_profile_memory --flake-finder`.

### Testing

`pytest -v test/profiler/test_cpp_thread.py --flake-finder` all passes.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/136304
Approved by: https://github.com/briancoutinho
This commit is contained in:
Huy Do
2024-09-20 02:56:47 +00:00
committed by PyTorch MergeBot
parent d45b0151e5
commit fe0e9fb385
1 changed files with 10 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
test/profiler/test_cpp_thread.cpp
View File

@ -47,6 +47,8 @@ void start_threads(int thread_count, int iteration_count, bool attach) {
static std::atomic<int> barrier = 0;
barrier = 0;
static std::atomic<int> another_barrier = 0;
another_barrier = 0;
thread_local bool enabled_in_main_thread = false;
std::vector<std::thread> threads;
@ -78,6 +80,14 @@ void start_threads(int thread_count, int iteration_count, bool attach) {
}
ProfilerEventHandler::Handler->emulateTraining(iteration, id);
// We need another barrier here to ensure that the main thread doesn't
// stop the profiler while other threads are still using it. This fixes
// https://github.com/pytorch/pytorch/issues/132331
++another_barrier;
while (another_barrier % thread_count) {
std::this_thread::yield();
}
}
});
}