You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Dec 3, 2021. It is now read-only.
One of the biggest existing blockers to being able to use Jupyter notebooks en masse for lessons is that the dependencies for all lessons are baked into the Jupyter image. This image is used for all lessons that use a jupyter notebook.
This is extremely non-ideal for a few reasons:
It's impossible to support multiple lessons with conflicting requirement versions
Anyone that wants to use a jupyter notebook must first get their notebook's dependencies added to the jupyter environment. Not only is this non-obvious, but it crosses the boundary between curriculum and platform, which means that a new platform release would need to occur before the author could make use of the new image.
We need to come up with a solution to this that allows notebook dependencies to be totally contained within the curriculum and serviceable by the author. The jupyter image that we distribute as part of the platform should be able to ingest this and provide the right environment.
Just spitballing, but maybe stages can state their dependencies in a requirements.txt file alongside the notebook. That much seems obvious enough and should be easy to enforce in Antidote on lesson ingest. The problem will be figuring out how to safely build these environments without introducing a bunch of runtime uncertainty. I am not a fan of just-in-time environment construction, where we do an install of a requirements.txt file at startup. The system is already pretty reliant on 3rd-party systems to be online, this would just introduce pypi to the mix. Unfortunately I can't think of any alternatives at the moment, so maybe this is the right solution for now.
The text was updated successfully, but these errors were encountered:
One of the biggest existing blockers to being able to use Jupyter notebooks en masse for lessons is that the dependencies for all lessons are baked into the Jupyter image. This image is used for all lessons that use a jupyter notebook.
This is extremely non-ideal for a few reasons:
We need to come up with a solution to this that allows notebook dependencies to be totally contained within the curriculum and serviceable by the author. The jupyter image that we distribute as part of the platform should be able to ingest this and provide the right environment.
Just spitballing, but maybe stages can state their dependencies in a
requirements.txt
file alongside the notebook. That much seems obvious enough and should be easy to enforce in Antidote on lesson ingest. The problem will be figuring out how to safely build these environments without introducing a bunch of runtime uncertainty. I am not a fan of just-in-time environment construction, where we do an install of arequirements.txt
file at startup. The system is already pretty reliant on 3rd-party systems to be online, this would just introduce pypi to the mix. Unfortunately I can't think of any alternatives at the moment, so maybe this is the right solution for now.The text was updated successfully, but these errors were encountered: