Release 2.3.0 broke custom iterable datasets

[aapot]

Describe the bug

Trying to iterate examples from custom iterable dataset fails to bug introduced in torch_iterable_dataset.py since the release of 2.3.0.

Steps to reproduce the bug

next(iter(custom_iterable_dataset))

Expected results

next(iter(custom_iterable_dataset)) should return examples from the dataset

Actual results

/usr/local/lib/python3.7/dist-packages/datasets/formatting/dataset_wrappers/torch_iterable_dataset.py in _set_fsspec_for_multiprocess()
     16     See https://github.com/fsspec/gcsfs/issues/379
     17     """
---> 18     fsspec.asyn.iothread[0] = None
     19     fsspec.asyn.loop[0] = None
     20 

AttributeError: module 'fsspec' has no attribute 'asyn'

Environment info

• datasets version: 2.3.0
• Platform: Linux-5.4.188+-x86_64-with-Ubuntu-18.04-bionic
• Python version: 3.7.13
• PyArrow version: 8.0.0
• Pandas version: 1.3.5

[gugarosa]

Apparently, fsspec does not allow access to attribute-based modules anymore, such as fsspec.async.

However, this is a fairly simple fix:

• Change the import to: from fsspec import asyn;
• Change line 18 to: asyn.iothread[0] = None;
• Change line 19 to asyn.loop[0] = None.

[mariosasko]

Hi! I think it's easier to replace import fsspec with import fsspec.asyn and leave the rest unchanged. @gugarosa Are you interested in submitting a PR?

[gugarosa]

Perfect, it is even better!

Just submitted the PR: #4630.

Thank you!