diff --git a/test/distributed/test_c10d_nccl.py b/test/distributed/test_c10d_nccl.py index 73c29c640da..15e38033310 100644 --- a/test/distributed/test_c10d_nccl.py +++ b/test/distributed/test_c10d_nccl.py @@ -2913,6 +2913,9 @@ class NcclErrorHandlingTest(MultiProcessTestCase): # nccl error happening before rank 0 timeouts time.sleep(4) + # Mimicing all ranks sensing the timeout, abort + process_group.abort() + if prev_nccl_async_error_handling is not None: os.environ[ "TORCH_NCCL_ASYNC_ERROR_HANDLING"