Multiple environments on PAVICS

[aulemahal]

Would it be possible to have multiple python kernels accessible in the jupyter lab ?

A few pro arguments:

• Each individual env could be less complex
• Keep legacy versions of some packages when breaking changes occur
• Split envs by usual work : hydro vs climate (for example)
• Split the Jupyter lab env from the work envs
• Add other languages kernels ? (Julia, R)

I see that it might be difficult to implement this, but I guess it could save time on the other end ? If envs are smaller, maybe they could be easier to maintain than the current single large one ?

I don't know if it would replace the current multiple image logic ? I guess not, but would be cool if it did.

[tlvu]

I already thought about that. This means instead of multiples images, we will have multitples conda envs inside the same image.

From the user's perspective, this would be simpler since switching env no longer means shutting down and restarting a new image.

However from a logistic perspective, this means a gigantic image containing multiples conda envs. Currently, with only one single conda env, the image size is already 9G ! Pulling the current image is already slow !

Building multi conda env image while keeping old conda env from previous builds will be quite tricky as well.