jupyter in Binder

[ecervera]

@psychemedia There is a problem for running in MyBinder, since the system uses WebSockets so an open port is necessary (default port number:1234). I don't know if that is possible to do with MyBinder, or there is any workaround.

[psychemedia]

But things should be able to run inside the Binderised container? Things like jupyter-server-proxy will also proxy internal ports onto a path along port 80?

Do things like https://github.com/maartenbreddels/ipywebrtc and/or https://github.com/yuvipanda/jupyter-desktop-server contain any useful corollaries?

[ecervera]

Sure, it's possible but I don't know how to do it :-(

Have you tried https://robotbenchmark.net/?
It's basically Webots online, but no Jupyter notebooks.

[psychemedia]

I saw that, yes, but for generality I'm interested in containerised environments that can be run locally or hosted. Having base containers built according to Jupyter docker stack or repo2docker conventions means they should also be runnable as-is directly from multiuser Jupyterhub setups, if they are available. (I'm ever in search of ways of lowering barriers to entry for educators who want to make environments available to learners at class scale at least, particularly where they have little or no technical support.)

[ecervera]

I have started from scratch a new Dockerfile which installs Webots R2018b in a fresh Ubuntu system. You may have a look at the R2018b branch of the repository. Webots runs and you can try any sample world. For the moment, docker with nvidia support (and a nvidia card) is necessary for running it. I plan to remove this requirement later.

[ecervera]

I have created a new branch for Webots R2020a in order to run the web streaming service (got an error with the previous version).

[psychemedia]

RE: the nvidia/GPU, and using a Dockerfile that requires a GPU, where the dependencies / GPU requirements actually come in?

Does Webots require a GPU in some way, or does it just perform better if there is one available?

Are there switches in the Webots startup command that configure compute requirements (GPU (in)availability, RAM, num. of cores etc)?

[ecervera]

"An NVIDIA or AMD OpenGL (minimum version 3.3) capable graphics adapter with at least 512 MB of RAM is required."
from https://cyberbotics.com/doc/guide/system-requirements

However, it works in VirtualBox (without 3D acceleration), albeit more slowly.

Don't know about such switches.

[psychemedia]

Ah, ok - thanks.

When things return to normal I'll try to get a testing team to run it on various old/low spec devices to see to what extent it works for simple simulations at least.

[ecervera]

Some progress:
I have set up a Docker image based on Ubuntu 18.04 without OpenGL:
https://github.com/RobInLabUJI/WebotsLab
(it can also be run without Docker, see instructions in the repository).
Still struggling with jupyter-server-proxy, though. It seems that the websockets protocol has some issues.