Some interactive jupyter widgets unusable slow with the Python plugin + Remote-SSH plugin

[tjahn]

Hello,

I am experiencing some really laggy interactive jupyter widgets when using the Remote-SSH extension.

The following python code creates an interactive jupyter widget in vscode showing a 3d Sphere rendered on the server and streamed to the jupyter widget:

# %%
import pyvista as pv
pv.start_xvfb() #if X11 DISPLAY is available on the server you may remove this line
sphere = pv.Sphere()
plotter = pv.Plotter(notebook=True)
plotter.add_mesh(sphere)
plotter.show(use_ipyvtk=True)

The interactive jupyter widget is visible but unusable laggy when running this code on a remote Server using the SSH-Extension.

I expected that the Python extension starts a jupyter server on the remote host, forwards the port to my local working machine and connects the interactive python tab to this port. However, this doesn't seem to be correct, since everything is working fine when I do this manually:

• If I use a local vscode project and let the python extension start the default jupyter server, everything is working fine.
• If I use a local vscode project and connect to an existing jupyter lab running on the remote Server (by forwarding a local port to the remote server "ssh -L 8888:remoteServer:8888 remoteServer"), everything is working fine.
• If I use the Remote-SSH plugin to connect to the server and then connect to the same existing jupyter lab, the widget is unusable laggy.
• If I use the Remote-SSH plugin and let the jupyter extension start its own default jupyter host, the widget is unusable laggy

Steps to Reproduce:

1. python3 -m pip install pyvista vtk
2. Connect to remote OpenSuse Server using the Remote-SSH plugin
3. Run the code interactively
4. Try using the mouse to interacting with the sphere

My system information:

vscode : 1.55.2
OS: MacOS 11.2.3
Remote OS: OpenSuse 15.2 Leap
pyvista : 0.29.1
IPython : 7.16.1
ipyvtk_simple : 0.1.4

[roblourens]

I guess the difference is just whether the extension host is running remotely, but I guess this is just due to the latency between your host and remote? Can you explain what aspect of it is laggy? Don't know about this scenario in particular, but there can be some code running locally, some in the EH, some in the kernel.

[tjahn]

When I use the mouse to interact with the window it takes 5-15 seconds to move. Also when it starts, it only renders with roughly 1 fps.

[tjahn]

I noticed the same problem with matplotlib. I installed ipympl with pip and then used "%matplotlib inline" to get an interactive window in vscode. Again its working fine on a local project.

# %%
import numpy as np
import matplotlib.pyplot as plt
%matplotlib widget

# %%
plt.plot( np.sin(np.linspace(0,100,1000)) )

[roblourens]

Could there just be high latency between your local machine and remote?

[tjahn]

Wouldnt it have been slow too when I connect to an existing jupyter kernel from a local vscode project in that case?

I didn't notice an especially high latency during my everyday work with ssh and vscode remote (without interactive plots) either.

[roblourens]

It's working fine for me on a remote. You might be interested in our new native notebook experience in vscode - it may work better there. https://devblogs.microsoft.com/python/introducing-the-jupyter-extension-for-vs-code/

You can also ask this on the python repo, they are the experts of python notebooks and matplotlib widgets and can give better advice.

https://github.com/microsoft/vscode-python/issues