Environment not persistent between sessions?

[rcpeene]

I think ever since the update where you select your environment upon launch, my environment no longer persists between sessions of Dandihub. This is a bit inconvenient since I have to reinstall The OpenScope Databook environment every time. Is there a way to manually save my environment?

[satra]

@rcpeene - let's use this opportunity to create a specific image for openscope. @asmacdo - could you please point carter to instructions for adding an image? i will let you also address carter's issue. perhaps some initiation process is overwriting something.

[asmacdo]

@rcpeene each user has persistent storage under /home/jovyan, so you'll just need to use a virtual environment somewhere in there. (The default conda env is in /opt, which is not persistent.)

For example:

$ python -m venv ~/venvs/my_venv
$ source ~/venvs/my_venv/bin/activate
(my_venv) $ 

For creating a new image we've got a short blurb in the docs: https://www.dandiarchive.org/handbook/50_hub/#custom-server-image. However, that has a broken links, and there's a bit more to it... Lets just develop those docs as we go and update when you're finished, (also feel free to ping me in MIT slack).

First youll want to add a new Dockerfile, ie Dockerfile.openscope https://github.com/dandi/dandi-hub/tree/b2f344a23bc726a2ff8690d97fdff637be539e54/images

Once that's built and pushed to dockerhub, it can be added as an image choice to the jupyterhub configuration.

dandi-hub/envs/shared/jupyterhub.yaml

Lines 116 to 125 in b2f344a

	choices:
	standard:
	display_name: "Standard"
	default: true
	kubespawner_override:
	image: "${singleuser_image_repo}:${singleuser_image_tag}"
	matlab:
	display_name: "MATLAB (must provide your own license)"
	kubespawner_override:
	image: "${singleuser_image_repo}:${singleuser_image_tag}-matlab"

We already have 2 choices, regular and Matlab. Youll need to add the new image as an option to each of the profiles.

[rcpeene]

Sorry for not getting back here;

For some environment reasons that we don't fully I understand, it is nontrivial for us to get a working docker container. Probably in the next several months we'll working on getting a docker container for the Databook and we'll check back.