Open PandaGab opened 1 year ago
Hey @PandaGab, it seems I can't reproduce this issue. Can you try to run almost the same setup (Linux, Py3.10, ClearML 1.11.1) but on a different machine with attrs
installed from PyPI and not some private package index/registry. I have a supposition that the issue might be due to +computecanada
suffix, but I'm not sure.
Hello, I also think it is coming from there but it should not affect (and crash) the usage of ClearML right?
Sorry but I was not able to try what you mentioned. But it seems the behavior of attrs
have changed between two versions 22.2.0 and 23.1.0 (both installed from the private package registry). To be clear, when running pip list
, I get 22.2.0+computecanada
and 23.1.0+computecanada
respectively:
>>> import attr
>>> attr.__version__
'22.2.0'
to
>>> import attr
>>> attr.__version__
<stdin>:1: DeprecationWarning: Accessing attr.__version__ is deprecated and will be removed in a future release. Use importlib.metadata directly to query for attrs's packaging metadata.
'23.1.0+computecanada'
Which could explain why my ClearML setup was working fine before but it stopped working after an update.
EDIT: Downgrading attrs
did solve the problem.
python -m pip install -U attrs==22.2.0+computecanada
Describe the bug
Importing clearml fails when running (
import clearml
).The error is:
I added the following print at line 305 in
clearml/utilities/version.py
and got the following:It seems (clearml/utilities/version.py) to be looking at the version of the package
attrs
currently installed on the cluster, which is set to the following versionEnvironment
1.11.1
3.10.2
Linux
Thank you for your help