@whyjz and @espg observed a bug that we can reproduce quite easily now, and I've diagnosed in some more detail. This is the bug.
Whenever we have loaded something big into memory, in this case, a 38 MB .pickle object, the CPU spikes and remain spiked even after the cell execution is completed.
We observe that the next cell execution is delayed, and then happens fast.
We observe that the CPU goes down to 0 again after a while
We observe that when we run another print("hello world") it goes back to 100% CPU again
Narrowing the problem
I've not been able to reproduce this with JupyterLab + Python 3.9 on my local computer.
I've not been able to reproduce this with a kernel started in the Classical notebook interface, or the Retrolab interface.
I've observed that the classical notebook interface and the retrolab interface, started to work with the jupyterlab created kernel, will be slow - but not if they initialized the kernels themselves.
We don't have JupyterLab --collaborative mode enabled, so it's not about that.
My current guess is that some JuptyerLab extension is messing things up for us.
How to debug this?
I'm not sure how to debug what goes wrong, we don't have an error. I would like to disable a JupyterLab extension one at the time and see if things resolve, but how to do that?
@whyjz and @espg observed a bug that we can reproduce quite easily now, and I've diagnosed in some more detail. This is the bug.
Narrowing the problem
My current guess is that some JuptyerLab extension is messing things up for us.
How to debug this?
I'm not sure how to debug what goes wrong, we don't have an error. I would like to disable a JupyterLab extension one at the time and see if things resolve, but how to do that?