Can notebooks connect to a kernel together?

[yangsenyi0116]

I have 2 notebooks. I want to share variables between two notebooks. How can I connect the kernels of the two notebooks to the same one?

[jasongrout]

This is very easy to do in JupyterLab (open both notebooks, switch the kernel in one to point to the kernel of the other). It is very hard in classic notebook, since each notebook launches its own kernel session, and it does not provide an option in the GUI to select a kernel that is already running.

[yangsenyi0116]

I currently use the enterprise gateway, and I found that when the notebook is connected to the kernel, a request to /api/kernel/${kernel_id} is sent. Can the notebook fix this kernel_id through configuration?

[kevin-bates]

As Jason said, this can be done via JupyterLab (and irrespective of EG). Here are two notebooks screen shots with kernels running in a YARN cluster via EG, the second of which I started, then changed the kernel to be associated with the kernel used for Untitled5.ipynb:

Print variable i to demonstrate shared state. You can also see the execution counters appear to be shared as well...

Jupyter Notebook doesn't expose this functionality to the best of my knowledge.

[akashd11]

Hi team,
Just an extension to this problem,
The solution to share the kernel kills the isolation of kernels.
In my use case we have a variable of type spark session that is set in the builtins of python, I just want the kernel to share this variables i.e. is there any way I can set the my spark variable in builtins module while the kernel starts.

[krassowski]

@akashd11 I think you may be interested in the subshells, see:

• Support kernel subshells jupyterlab/jupyterlab#16963
• Add JEP for sub-shells enhancement-proposals#91

Once this this is merged you will be able to have two subshells for the same kernel, sharing the variables but performing execution independently. I believe that only the latest pre-release of ipykernel supports subshells as of today.