Asset registration doesn't work with a remote Jupyter server

[joshuamanns]

When using a library like Interact that relies on AssetRegistry on a remote Jupyter server, 404s are thrown for assets.
See this thread: JuliaGizmos/Interact.jl#270

[shashi]

Thanks for the issue! @travigd you needed to solve this problem too right?

One idea which can potentially work: I believe there's only a ZMQ link between the kernel and the Jupyter server. So we can make our plugin serve files by fetching them via ZMQ instead of directly reading from disk. IMO this is still much better / proper than trying to send files over WebSocket by writing some Julia code.

[twavv]

@shashi This is a little bit different than what I was talking about (which is running a Jupyter kernel on a remote server rather than the entire notebook/lab server).

At any rate, are you using JupyterHub? What is the url of your notebook and what URL is the AssetRegistry using (i.e. what URL do you see 404 errors for in the browser console)?

[shashi]

Oh I assumed "server" means kernel in the issue title...

[twavv]

Yeah, remote kernel (insofar as remote to the web server) is a pretty rare case as far as I know. The only common case is "enterprise" kernels (i.e. things that run on clusters like Spark).

@joshuamanns Are you still having issues?

[ilanpillemer]

I am having this issue

URL of notebook : https:///jupyter/user/ilan.pillemer@@dev/lab/workspaces/auto-0
URL of 404 : https:///jupyter/user/ilan.pillemer@@dev//assetserver/ced635b163795d11ca32b58b98ce168337af879a-style.css

[twavv]

Is it the double-slash that's causing the issue? I've not had these kinds of issues on JupyterHub myself.

[ilanpillemer]

when I tried removing the double slash and pasting into the url bar, it still didnt give me the css file.

$ echo $JUPYTERHUB_SERVICE_PREFIX

/jupyter-dev/user/ilan.pillemer<snipped>/

$ echo $JULIA_ASSETREGISTRY_BASEURL

$ echo $ENV

$

So the double slash is caused because JUPYTERHUB_SERVICE_PREFIX has a trailing slash.
According to this jupyterhub/jupyterhub#997 its always meant to have a trailing slash if it exists.

But when I simply tried pasting the URL without the double slash it still seemed to not work.
So I think its definitely part of the problem, but maybe not all of the problem.

Do you know how I can diagnose to see what the URL it is expecting should be? ie where the asset server is running on?
If you tell me how to diagnose further so we can get the bottom of it and get this working, I will use those steps to diagnose and analyse. But I am unsure how to determine what the expected URL should be. I am new to Julia so may need some guidance on how to override values etc, to see if different changes will solve the problem.

by the way this is from a regular notebook, the above comment a day or so ago was trying to get it working in JupyterLab. But the same thing happens in both.

[twavv]

Is WebIO pre-installed on the server?

The assetregistry extension is a Jupyter server extension, which needs to be installed before the Jupyter notebook instance starts (there's no way to activate it after the Jupyter notebook has started).

[ilanpillemer]

Currently I run from the notebook. (in the Jupyter Hub)

using Plots
using Interact
using WebIO
WebIO.install_jupyter_nbextension

or alternatively the lab extension.

and then restart the kernel.

currently I believe we use this to start-up the Julia kernel in the k8s cluster as part of Jupyter Hub running the K8s cluster.

RUN julia -e 'import Pkg; Pkg.update()' && \
    (test $TEST_ONLY_BUILD || julia -e 'import Pkg; Pkg.add("HDF5")') && \
    julia -e "using Pkg; pkg\"add IJulia\"; pkg\"precompile\""

I would assume we would need to change this to include WebIO and Interact. Any suggestion on the best way to do that? FYI @dbricare

[twavv]

Yep, just add WebIO and run WebIO.install_jupyter_nbextension() as part of that same run.

RUN julia -e 'import Pkg; Pkg.update()' && \
    (test $TEST_ONLY_BUILD || julia -e 'import Pkg; Pkg.add("HDF5")') && \
    julia -e '\
        using Pkg; pkg"add IJulia WebIO"; pkg"precompile"; \
        using WebIO; WebIO.install_jupyter_nbextension(); \
    '

[ilanpillemer]

that made all the difference.

However I still needed to run the below in the notebook and then type CMD-R.. but after that it worked 🎉

Anyway to avoid having to do that?

using WebIO
WebIO.install_jupyter_nbextension