some tests failed on macOS

[natsukium]

Hi, thanks for developing this awesome project.

Version :

4.0.0

Describe the bug :

I tried to use truvari on macOS, and when I ran the test, I got the following error.

error log

test_anno_grm ran in 2 sec with 34/0 lines to STDERR/OUT
 FAIL EXIT CODE (LINE 59)
-->     expected EX_OK, observed Unknown code: 1
    data = pd.concat(chunks, ignore_index=True)
  File "/nix/store/jhjlf5yy6ar5g7jl363pxzk6fqxwzj5s-python3.10-pandas-1.5.3/lib/python3.10/site-packages/pandas/util/_decorators.py", line 331, in wrapper
    return func(*args, **kwargs)
  File "/nix/store/jhjlf5yy6ar5g7jl363pxzk6fqxwzj5s-python3.10-pandas-1.5.3/lib/python3.10/site-packages/pandas/core/reshape/concat.py", line 368, in concat
    op = _Concatenator(
  File "/nix/store/jhjlf5yy6ar5g7jl363pxzk6fqxwzj5s-python3.10-pandas-1.5.3/lib/python3.10/site-packages/pandas/core/reshape/concat.py", line 422, in __init__
    objs = list(objs)
  File "/nix/store/v6f6y9mwmqj9v1jffiy4g96ili4pj5la-python3-3.10.11/lib/python3.10/multiprocessing/pool.py", line 873, in next
    raise value
AttributeError: 'types.SimpleNamespace' object has no attribute 'ref_filename'

test_anno_grm ran in 0 sec with 5/0 lines to STDERR/OUT
 FAIL EXIT CODE (LINE 48)
-->     expected EX_OK, observed Unknown code: 1
Traceback (most recent call last):
  File "<string>", line 4, in <module>
  File "/nix/store/kwv7hx63pw3djbd5ryk8365v25lh1ka6-python3.10-joblib-1.2.0/lib/python3.10/site-packages/joblib/numpy_pickle.py", line 650, in load
    with open(filename, 'rb') as f:
FileNotFoundError: [Errno 2] No such file or directory: 'test_results/grm.jl'

test_anno_trf ran in 2 sec with 26/0 lines to STDERR/OUT
 FAIL EXIT CODE (LINE 72)
-->     expected EX_OK, observed Unknown code: 1
    main()
  File "/private/tmp/nix-build-truvari-4.0.0.drv-0/source/truvari/__main__.py", line 102, in main
    TOOLS[args.cmd](args.options)
  File "/private/tmp/nix-build-truvari-4.0.0.drv-0/source/truvari/anno.py", line 59, in anno_main
    ANNOS[args.cmd][1](args.options)
  File "/private/tmp/nix-build-truvari-4.0.0.drv-0/source/truvari/annotations/trf.py", line 658, in trf_main
    for i in chunks:
  File "/nix/store/v6f6y9mwmqj9v1jffiy4g96ili4pj5la-python3-3.10.11/lib/python3.10/multiprocessing/pool.py", line 873, in next
    raise value
AttributeError: 'types.SimpleNamespace' object has no attribute 'args'

test_anno_trf_reg ran in 2 sec with 26/0 lines to STDERR/OUT
 FAIL EXIT CODE (LINE 83)
-->     expected EX_OK, observed Unknown code: 1
    main()
  File "/private/tmp/nix-build-truvari-4.0.0.drv-0/source/truvari/__main__.py", line 102, in main
    TOOLS[args.cmd](args.options)
  File "/private/tmp/nix-build-truvari-4.0.0.drv-0/source/truvari/anno.py", line 59, in anno_main
    ANNOS[args.cmd][1](args.options)
  File "/private/tmp/nix-build-truvari-4.0.0.drv-0/source/truvari/annotations/trf.py", line 658, in trf_main
    for i in chunks:
  File "/nix/store/v6f6y9mwmqj9v1jffiy4g96ili4pj5la-python3-3.10.11/lib/python3.10/multiprocessing/pool.py", line 873, in next
    raise value
AttributeError: 'types.SimpleNamespace' object has no attribute 'args'

After a bit of research, I found out that it is caused by how the process is started in the multiprocessing module.
(ref: https://docs.python.org/3/library/multiprocessing.html#contexts-and-start-methods)
I changed it from the macOS default of spawn to the Linux default of fork, and the tests passed.

To Reproduce :

Run bash repo_utils/truvari_sshtests.sh on macOS.

Expected behavior :

All tests pass.

Example Data :

Additional context :

The full log is here.
https://hydra.nixos.org/build/219117259/log

[ACEnglish]

Thank you for reporting this. I'm sure it wasn't easy to solve. I was able to recreate the issue and confirm that having multiprocessing.set_start_method('fork') before with multiprocessing.Pool.. calls can get tests to pass.

However, I don't think changing the set_start_method is a viable solution. The python docs state:

Changed in version 3.8: On macOS, the spawn start method is now the default. The fork start method should be considered unsafe as it can lead to crashes of the subprocess. See bpo-33725.

Possible solutions:

• Make it clear that some truvari anno commands will fail on macOS
• Make it clear that truvari is designed on/for unix and recommend docker (which is how I run truvari tests on my mac).
• Change design pattern to not depending on 'fork' sharing semantics. Specifically it's the global namespace.

And as I'm digging deeper into bpo-33725 and the issues it references, I'm thinking changing the code might need to be the solution as this non-explicit sharing may become not possible in a future python version.. maybe.

[natsukium]

Thanks for your response.

However, I don't think changing the set_start_method is a viable solution.

Yes, I don't think using set_start_method is a good way to fix it.

and thank you for starting the fix as quickly.
Let me know if I can help in any way.

[ACEnglish]

Functional tests off of develop now pass on mac. If you're still seeing a problem, please re-open this ticket.

These changes will be present in v4.1. However It'll be some time before that release is cut as there's at least two more features to add and a lot of user testing happening for a publication tied to the next version.