Custom Websocket url differ in JupyterLab and Jupyter Notebook Server

[sntgluca]

Dear all,

While searching for a workaround on another issue, I encountered a mismatch in the behaviour of JupyterLab and Jupyter Notebook server.
I post it here, as I think this is the most sensible place to discuss this, although it has partially to do with JupyterHub as well.

The mismatch is in the setup of the custom websocket_url parameter. It can be replicated with this minimal JupyterHub configuration:

c.JupyterHub.base_url = <BASE_URL>
c.Spawner.websocket_url = "ws://<ADDRESS>/<BASE_URL>"

where BASE_URL and<ADDRESS> are the custom url and location of my choice.
The final websocket address should have the form ws://<ADDRESS>/<BASE_URL>/user/{username}

The final websocket url built by the backend Jupyter session has the form

• ws://<ADDRESS>/<BASE_URL> in JupyterLab
• ws://<ADDRESS>/<BASE_URL>/user/{username} in Jupyter Notebook Server

This inconsistency is problematic, because I cannot customise the websocket_url parameter in a way that works for both JupyterLab and Jupyter Notebook server.
How would you address this problem?

Thanks
Gianluca

[sntgluca]

A quick answer about this, in case someone else came across the same problem.

Using notebook version 6.0.0, the web url is defined, for the traditional notebook server, in the file notebook/static/notebook/js/main.min.js

Adding the following lines in row 50727

+        this.ws_url = this.ws_url.replace(/^((?:\w+:)?\/\/(.+:\d+))(.*)/, '$1');
+        that.ws_url = that.ws_url.replace(/^((?:\w+:)?\/\/(.+:\d+))(.*)/, '$1');

removes the duplication from the notebook server websocket url, hence making both JupyterLab and Jupyter Notebook server behaving in the same way.

[jasongrout]

Which behavior do you think is more correct?

[sntgluca]

Hi @jasongrout !
I don't think there's an obvious answer to your question.

I found that modifying the notebook package's code was easier, and I considering JupyterLab is the default. Do you consider the patch above good enough?

The only argument for keeping the notebook package behaviour is that it seems simpler (as in, it doesn't strip arguments from the url). But I don't consider this a strong point.