Loading dask.distribute.Client fails with 2022.12.0

[habi]

I have a project that uses an environment.yaml file, in which I load dask.
I am (currently) not pinning version numbers.

I've generated the conda environment on a new machine today and do this below in a Jupyter notebook.

import dask
from dask.distributed import Client
print(dask.__version__)
client = Client()

prints "2022.12.0" and fails with "AttributeError: 'Nanny' object has no attribute 'add_done_callback'"

The same code on a machine on which I've installed the environment a while ago prints "2022.05.0" and produces a Client/LocalCluster where I can connect to with dask-labextension.

[jrbourbeau]

Thanks for reporting @habi. What version of Python are you using? FWIW I've seen that error before when using Pyhon 3.11, which Dask doesn't currently support (xref #9691)

[habi]

Indeed, the fresh install on that machine installed Python 3.11.0, too.

Adding python=3.10 to my environment.yaml file and recreating the conda environment works as expected, e.g. I can create a Client/LocalCluster with dask version "2022.12.0".

As I'm not very proficient with working with environment files, it it customary to pin the 'current' versions of the imports?

[phobson]

@habi the specificity with which you're pinning something depends on your use case and personal preference. At any rate, I personally feel that one should always pin python version.

[jrbourbeau]

Great, thanks for the feedback @habi -- for now I'd stick with python<3.11. I'm going to close this as a duplicate of #9691. We're currently working on Python 3.11 support and hope to have it released soon.