Prevent process/variable space sharing on multi-user setups

[albertmartga]

( Related to #26 )

We're experimenting with a JupyterHub instance (tljh) to serve multiple customers on the same instance, and found that Matlab seems to have a single master "session" for all users running any notebook from any server within JupyterHub.

This means that, for example, if two users are interacting with their own notebooks and both have a data variable it'll always hold the value from the latest assignment, and users will be constantly stepping onto each other.

This is different from the python kernel, where each notebook seems to use its own independent interpreter session.

Any plans to support this use case? Ideally it'd also support multiple license, so each user in jupyter server (or every notebook) will also have to provide their own license.

Thanks!

[prabhakk-mw]

@albertmartga

Thank you for your post!

As far as I understand the Jupyter Eco system, the expectation is that every user that logs into JupyterHub gets his own Jupyter server and his own HOME folder.

Users are thereby isolated from each other at the Jupyter Server Level, and by consequence the file system as well.

Based on the information you've provided, it sounds like you have multiple different users logging into a Jupyter Server instance as the same system user. i.e: The same entity from a system point-of-view.

It seems like this specific use case is akin to a single user looking for isolation between his MATLAB notebooks. We are currently working towards enabling such a feature. Where by each notebook, will have to specify that is requires a separate MATLAB for itself.

That said, we would still have issues meeting your desired licensing model.
Given that all your users are sharing the HOME folder, it would be hard to enable different forms of licensing for them. We could perhaps disable the caching of licensing information all together, thereby requiring your users to enter licensing information for every MATLAB notebook that they spin up.

Even then, there would still be ways for them to "step" on each others toes, as there is no isolation at the system level.

Unfortunately, I don' think any solution we provide will fully serve this style of using Jupyter.

Please do correct any of my assumptions listed above.

For now, I'll mark this request as waiting for the "One MATLAB per notebook" feature.