Jupyter kernels as a service

Only a proof of concept, but this is interesting

The idea is to run the kernel on one machine, and the jupyterlab instance on another.

A specific use case: run the kernel on a gadi PBS job and the jupyterlab instance on your laptop.

How is this different to OOD/ARE? Well it means the notebook session is persistent on your local machine, the kernel dies and you can restart it and reconnect, but notebook remains. And you can mix notebooks with remote kernels and local ones.

It also means you can use other client to connect to the kernel, like VS Code. Maybe this is useful to you @atteggiani?

1 Like

If this can play nicely with something like the nb_conda_kernels extension, that would make for a very convenient workflow…