Kernel detected as dead on refresh of notebook page. Dead kernel icon never cleared after that even though the kernel is working.

Hi folks,

I'm using a notebook under JupyterHub and with a custom kernel. Things work fine when I first launch the notebook, but, if I refresh the page, about two seconds after it is reloaded, I will get the "The kernel appears to have died..." message, and the Dead kernel red button/icon will be added to the status bar. At the same time, the custom kernel status will change to a bomb.

After a few seconds, the custom kernel icon will change back to idle, and the kernel will correctly execute code from the notebook, but the Dead kernel icon will remain, and never gets cleared.

This is on notebook 6.4.12. I tried 6.5.1, but it doesn't play nicely with some custom widgets we are using... However, 6.5.1 also exhibited the same behavior about the Dead kernel notification.

The notebook is running in a container, launched by the JupyterHub Docker spawner and based on jupyter/minimal-notebook:python-3.9.12, but with a couple of changes to maintain compatibility with our kernel and widgets (ipykernel 5.5.5, ipywidgets 7.5.1, and, as mentioned, notebook 6.4.12 instead of the 6.4.11 that came with the image).

Please let me know any additional details you need. I would also appreciate any pointers to documentation on how the Dead kernel detection happens and how the message is supposed to get cleared.

Thanks,
Paul

Reproduce in the notebook page using the reported JupyterHub Docker environment by refreshing a running custom-kernel notebook and observing the kernel status and dead-kernel icon. Start by tracing how the dead-kernel notification is triggered and cleared; done means the icon and message clear after the kernel returns to idle and executes code successfully.

Written by the indexing model from the issue text.