Future issue with jupyter notebook 5.5 regarding hostname in docker ?

[tomjorquera]

By toying with upgrading jupyter to version 5.5 inside a container, I stumbled upon the following issue:
jupyter/notebook#3605

Apparently, jupyter-notebook behavior regarding the url has been changed in 5.5, and it will now try to display the host name instead of localhost. Inside docker, it means that the url that now appears will be something like https://8bc26195e104:8888/?... (which does not work when copy-pasted into the browser, host must be changed manually to localhost)

Feels like it's a somewhat significant issue for the project.

Maybe a workaround would be to propose an option to jupyter-nb to disable name resolution ?

[parente]

I agree this can be a confusion. In fact, the localhost link printed by Jupyter only ever directly worked if a user was running the docker container on a local machine. When running a container on a remote server, like a VM in the cloud, Jupyter Notebook still prints the URL as localhost because it is blissfully unaware that it is running in a container on another machine, as it should be.

Docker does allow a user to specify the fully qualified domain name of the container with docker run --hostname my.host.name command. I'm pretty sure this will impact what the notebook server will print when it starts. I'm not sure how useful it is in practice.

[parente]

I took a crack at updating the quick start steps to describe what hostname should be used to access the docker image when a container launches. Additional PRs against the docs are welcome.

I'm inclined to close this issue, as it's already being tracked upstream in jupyter/notebook and I don't have a good solution at the moment other than to continue to refine the instructions as people find specific fault with them.

[minrk]

The ip and port output from the notebook should generally be ignored when run in containers, since it's ~never right and never can be by default. Only in the very specific case of local docker where the forwarded and exposed ports match can it ever be correct. I'm not sure what the best solution is other than allowing override (jupyter/notebook#3668) which doesn't change the fact that the default will be wrong.

[parente]

Closing this issue as the docs have been updated and we don't have any ideas at the moment on how to make the docker behavior clearer. If anyone has a suggested improvement, we'll be happy to take a PR.