Debugging stuck kernels

[miohtama]

What would be the best way to debug a misbehaving kernel (ipykernel).

Sometimes Kernel dies and sometimes it does not start and is stuck in the starting phase.

I did not find any user interface-facing way to diagnose what could be wrong with the kernel. I have also not seen any diagnosing/debugging options with any other Jupyter implementations.

Running cocalc v2 Docker image.

• Can I view the logs / map them to the host filesystem
• Would it be possible to get logs to Docker stdout and use Docker's own logging system for tracking this (seems to be a common practice)

[haraldschilly]

I usually try in a terminal: jupyter console --kernel=[name], where the "name" comes from the list given by jupyter kernelspec list.

[williamstein]

Yes, that's exactly what I do first as well. It's not 100%, but it helps with any generic problems that would impact all Jupyter clients.

If it's something fairly cocalc specific, enabling this DEBUG flag can be helpful:

https://github.com/sagemathinc/cocalc/blob/master/src/packages/jupyter/kernel/kernel.ts#L27

Note that I just massively refactored the Jupyter code in cocalc during the last week, so if you don't build a new image you might not see packages/jupyter. This is in master now though.