[python-package][dask] handle failures parsing worker host names

[jameslamb]

Contributes to #3867.

This PR fixes mypy errors in dask.py

python-package/lightgbm/dask.py:99: error: Invalid index type "Optional[str]" for "Dict[str, _HostWorkers]"; expected type "str"
python-package/lightgbm/dask.py:383: error: Invalid index type "Optional[str]" for "defaultdict[str, Set[int]]"; expected type "str"

This is mypy correctly pointing out the urlparse(something).hostname can be None, but lightgbm.dask treats as if it is guaranteed to be non-None.

from urllib.parse import urlparse

print(urlparse("https://127.0.0.1:123").hostname)
# 127.0.0.1

print(urlparse("127.0.0.1:123").hostname)
# None

The changes in this PR add explicit handling for that case, so that if it happens lightgbm will raise a more specific and helpful error.

Confirmed that this suppresses the error by running the following.

mypy \
    --exclude='python-package/compile/|python-package/build' \
    --ignore-missing-imports \
    python-package/

[jmoralez]

LGTM! I think right now the _group_workers_by_host function can't get invalid addresses since it gets them from dask, however it's nice to have that function tested in case it ever needs to be used somewhere else.

[jameslamb]

can't get invalid addresses since it gets them from dask

You're right, I don't know any specific case right now where a worker address could make it to this method that doesn't have a hostname in the parsing result from urlparse().

But I think mypy is right here that it is POSSIBLE for urlparse(address).hostname to be None, and I think we'll be glad to have this specific error message there to make debugging faster.

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.