Manage the culling of Jupyter Server with scheduled jobs to support JupyterHub

[thomasLeclaire]

Problem

In the context of JupyterHub, we should have some way to bypass the culling if some notebook are currently scheduled on the jupyterserver of an user.

Proposed solution

no real ideas :-(

[andrii-i]

Hi @thomasLeclaire. Thank you for creating this issue. Could you please elaborate on the problem providing a more detailed description of the problem and reproduction path?

[sunayansaikia]

@andrii-i -- If I'm not wrong, the requirement is that JupyterHub has a feature to cull Jupyterlab servers upon no user activity (ref: service-to-cull-idle-servers. However any jobs scheduled using 'jupyter-scheduler' probably is not considered as an actvity by Jupyterlab server and consequently JupyterHub will cull the jupyterlab server thinking that the jupyterlab server is idle (doing nothing).

Can there be enhancements done to 'jupyter-scheduler' to let the jupyterlab server somehow detect the jobs being scheduled as an 'activity' and ultimately not getting itself culled by JupyterHub? However, if no jobs are scheduled, I think this can be considered as a 'no activity'.

[andrii-i]

Thank you for the details.

[andrii-i]

Potentially related changes on the server (not released yet) jupyter-server/jupyter_server#1429

[mimikwang]

Ran into the same need - I have an idea of how this can be done - happy to submit a PR!

[andrii-i]

@mimikwang please feel free to open a PR, it would be much appreciated

[mimikwang]

@andrii-i FYI I've submitted a PR for this