Fixing a Jupyter kernel connection issue

N.B. This post was migrated from oli-hall.github.io to oli-hall.com on 18/04/2019

Another quick post, this time around Jupyter notebooks. I quite often use these for prototyping and playing around with ideas in Python, and I tend to have a scratchpad notebook in most of my projects. However, in a recent project, after pip installing jupyter, I found myself with a kernel that refused to connect. Checking the console output (the output in the terminal window used to launch the Jupyter notebook server), I found the following:

[I 14:36:02.130 NotebookApp] Kernel started: 3e7f0bc3-6b24-4a95-b0cf-b729ac4dccd8
 [I 14:36:02.788 NotebookApp] Adapting to protocol v5.1 for kernel 3e7f0bc3-6b24-4a95-b0cf-b729ac4dccd8
 /Users/oli-hall/Code/pymidi/bin/env/lib/python3.5/site-packages/notebook/base/zmqhandlers.py:284: RuntimeWarning: coroutine 'get' was never awaited
   super(AuthenticatedZMQStreamHandler, self).get(*args, **kwargs)

This didn't look particularly great. Something low down was clearly not working as it should. Fortunately, in this age of Google and Stack Overflow, the answer was not far away. It seems that jupyter depends on the tornado web server, and doesn't pin the dependency. This means that right now, installing jupyter installs tornado 6.0, which apparently is throwing these ZeroMQ issues that are causing the kernel to disconnect from the notebook somehow (I've been meaning to dig deeper, but have yet to find the time). Anyhow, forcing a downgrade of tornado with pip to 5.1.1 solves the issue:

pip uninstall tornado
 pip install tornado==5.1.1

If you want to dig in deeper, there's an issue open on the Jupyter GitHub project to track it. Hopefully this'll get fixed soon (in fact, it likely will be fixed by the time I publish this), but nevertheless, this may come in useful for someone if they have the same symptoms.