search

oceanhackweek / jupyter-image

jupyter-image
MIT License
2 stars 7 forks source link

An initial Python attempt at splitting images #43

Closed abkfenris closed 2 years ago

abkfenris commented 2 years ago

In order to hopefully allow for faster image loading, 2i2c is letting us set up separate images for Python and R (and possible Python + ML libraries or other massive stuff).

I've started by creating a new directory for a Python image with a few updates to match the current 2i2c environment, and slimmed down further using micromamba.

I've also created a re-usable workflow for building and pushing Docker images, that we can then reference from image/directory specific workflows.

I'm fully expecting needing to tinker on the workflows to get it building and pushing as expected.

abkfenris commented 2 years ago

I'm gonna merge without review so that I can debug and get 2i2c testing the images.

ocefpaf commented 2 years ago

I'm gonna merge without review so that I can debug and get 2i2c testing the images.

Damn, I was too slow!

abkfenris commented 2 years ago

Thanks. I was going to keep dependency changes minimal at this stage, just so that we can get to the point of testing with 2i2c sooner. I've also had enough weirdness around Python 3.10 to leave that for a later PR.

Of course I also moved to Micromamba as a base at the same time, so my judgement may be lacking.

ocefpaf commented 2 years ago

Good strategy! For the GHA actions bumps we should probably adopt dependabot, not sure if we can enable it only for GHA. The Python 3.10 is not important. I'm mostly curious about what is broken.

abkfenris commented 2 years ago

I nearly wrote 'thanks, dependabot' You can enable it just for actions, but I would probably set it up for the Docker FROM images too.

Largely some weirdness about types.