Problems with jupyterlab 0.34 when using a kernel that takes a while to start up

[Christian Schafmeister]

Hi there,

Could any kind jupyterlab developer give me an overview of what time-outs are important in the jupyter communication protocol?

We have developed a Common Lisp kernel that uses jupyter widgets including a few custom jupyterlab extensions (nglview and bqplot).

Our kernel normally takes about 30 seconds to start up because it invokes its slow, llvm-based compiler a few times at startup (we are working on starting up from images - so this will improve).

This slow startup appears to cause timing problems with the jupyter protocol that are very hard to debug.

It seems the jupyter protocol makes assumptions about how fast kernels can respond at startup and that might cause problems for other kernel developers - it cost me several weeks to figure this out.

The problems go away now that I wrote a fork-server that warms up our kernel and then forks children to service notebooks.

I've submitted this as an issue:   https://github.com/jupyterlab/jupyterlab/issues/5208

Cheers,

Christian Schafmeister

Professor, Chemistry Department

Temple University