[ENH] - Better way to use local packages on nebari

[krassowski]

Feature description

I find myself limited by nebari with respect to working on larger analyses (multiple notebooks spread across directory tree). Locally, I would either:

• start JupyterLab with modified PYTHONPATH variable to include the root of the analysis repo and have relative imports and other fun things for free
• install package in editable mode in the appropriate environment

However in nebari I cannot do either (because of lack of exposed spawner customization and conda store respectively). Nebari now offers documentation on How to Develop Local Packages on Nebari, however I find the documented solution unfit for the scenario of using Nebari for data analysis in notebooks.

In particular, I find the requirement to include on top of each notebook the block:

import sys
sys.path.append('/home/myusername/venv_myenv/lib/python3.10/site-packages/')

cumbersome and making it difficult to collaborate.

It is really counter-productive having to add this in each notebook, also if another user wants to run it they need to change the user name.

Developing in ~/shared would help a bit, but would make that snippet even larger:

import sys
import pathlib
sys.path.append(pathlib.Path('~/shared/venv_myenv/lib/python3.10/site-packages/').expanduser())

Parametrized kernels (https://github.com/jupyter/enhancement-proposals/pull/87) could help here once they lands (it may be worth dedicating some time to help it land).

Value and/or benefit

More happy users

Anything else?

No response

[krassowski]

Another approach is taken by https://bluss.github.io/pyproject-local-kernel/ which detects python environment based on pyproject.toml file. This would be amazing to have. It also allows to swap out the kernel startup arguments so we could pass custom PYTHONPATH.

I am not sure if we would have a consensus to add it to base docker image (maybe?), but an option to install it (without having to ship own docker image) would be great.

[krassowski]

Of note the

import sys
import pathlib
sys.path.append(pathlib.Path('~/shared/venv_myenv/lib/python3.10/site-packages/').expanduser())

does not really work when the package is meant to be in repository which is in the user folder.

[Adam-D-Lewis]

Pinging @kcpevey due to prior interest in the topic

[kcpevey]

I completely agree that the current status is functional but not practical. I raised this as an issue on this repo a while back.

I'm not sure how pyproject-local-kernel would work with conda-store. I think the proper solution would be to sort out how to make this work within conda-store.