data icon indicating copy to clipboard operation
data copied to clipboard
verified publisher icon pytorch

MultiProcessingReadingService is stuck for distributed training on windows

Open ejguan opened this issue 3 years ago • 1 comments

🐛 Describe the bug

After https://github.com/pytorch/data/pull/827 is landed, our nightly release becomes broken for distributed test on windows. It basically will time out on windows distributed testing.

Needs to investigate.

Versions

main

ejguan avatar Oct 24 '22 18:10 ejguan

Two blocking tests:

  • test_proto_rs_dl2
  • test_proto_rs_determinism

ejguan avatar Oct 24 '22 21:10 ejguan