Established tools and data visualization: Running desktop applications on JupyterHub

[lheagy]

"Under this project we will extend the types of software that can be delivered to users to include desktop-native applications that run on Linux distributions"

Examples of relevant desktop applications:

• ncview
• paraview

[lheagy]

I believe @yuvipanda has already (mostly?) addressed this (with a demo hub here: http://desktop-test-staging.datahub.berkeley.edu/).

So the main remaining item might just be finding a relevant demo or two? I can provide an example that uses Paraview. I know there were a couple subtleties with paraview specifically, so if this is too involved, we can definitely revisit.

[yuvipanda]

Hello!

github.com/yuvipanda/jupyter-desktop-server/ is the project that helps with this. Paraview needed OpenGL, which was a bit difficult. ncview should be doable though....

[lheagy]

@jhamman: is there someone / a group you have in mind who would benefit from having ncview on a JupyterHub? It would be nice to have a use-case to demonstrate this with.

[jhamman]

NCAR has to be one of the largest single user groups of ncview. I'm sure there would be broad interest in support for in the Casper/Cheyenne JupyterHub deployment.

[lheagy]

Thanks @jhamman!

@kmpaul, @andersy005: is this something you would be interested in helping push towards with the JupyterHub deployment at NCAR?

[tristanlt]

Dear @lheagy, here, we have a smart integration of JupyterHub with desktop functionality (for ncview, Matlab...).

JupyterHub service is based on OpenStack instances spawner (libcloudspawner). Each singleuser instance starts a VNC server (on vGPU if available) and a websocket proxy (VNC<>WS). This WS service is served by singleinstance proxy. Finally, users found one link on jupyter launcher, created with @yuvipanda jupyter-launcher-shortcuts, which redirect to noVNC with correct parameters (VNC-WS path and credentials).

I understand if you are wondering. But it really works well. :wink:

This is not yet publicly documented, but if someone is interested this could be a great motivation to write something.

Regards, Tristan

[fperez]

Thank you @tristanlt! We've also recently been playing with the Desktop integration in JupyterHub through Magic Castle, and I think it's a useful addition to the system for certain use cases.

It would be wonderful to have a few more details of your deployment, so we can make that integration as smooth as possible by default in JupyterHub (at least as a feature that can be activated on deployment with minimal effort). If we make it more "built-in", it will be easier to think of what the optimal UI for end users should be...

Pinging @consideRatio as well so we think about this from the perspective of the "landing" experience...