Parametrized kernels for passing arguments in to starting up kernels

[balamuruganvg]

Please add support for parametrized kernels so that one can pass custom arguments in to the kernels from the client (API handlers).

@SylvainCorlay @jasongrout

[jasongrout]

@SylvainCorlay - I couldn't find an issue from the previous discussion (last year?) on kernel arguments. Is there one?

CC also @takluyver @minrk

[kevin-bates]

Thanks for opening this issue. FWIW, we've had some discussion of parameterized kernels over in Enterprise Gateway. This is more along the lines of configuring the kernel runtime env but would also apply to explicit parameters as well.

[balamuruganvg]

Thanks for the EG pointer @kevin-bates, its quite helpful.

[SylvainCorlay]

I could find the notes from the last dev meeting about parameterized kernels:

https://paper.dropbox.com/doc/Day-1-Kernels-jupyter_client-IPython-Notebook-server--Ab6UGuDRv_0wv95Z4NcGb1j8Ag-MyS7d8X4wkkhRQy7wClXY#:uid=671520827638401135554679&h2=Parameterised-kernelspecs

[kevin-bates]

A proof of concept relative to kernel providers can be found here: https://github.com/takluyver/jupyter_kernel_mgmt/pull/22

Note that I believe the core functionality would be to have metadata that describes the parameters (ideally adhering to a defined schema) returned from /api/kernelspecs, processed and prompted by front-end machinery to produce a simple dict of name/value pairs, then submitted in the /api/kernels POST body.

The nice thing about the kernel provider approach is that each provider would define and understand the parameters it supports. Of course the current model, given rich enough metadata, could perform validation in a similar manner.