cyberbotics/webots-server

Repository containing necessary files to run a server able to run Webots simulation

Webots Server

This repository contains the files to be deployed on a server machine to run Webots simulations online.

The documentation on how to set-up a Webots simulation server is provided in the Webots user guide.

webots-server's Issues

Unnecessary fields in docker compose files

I recently did some tests and it seems that some fields inside the docker compose files are unnecessary: privileged, everything concerning XAUTHORITY and QT_X11_NO_MITSHM

Does anybody know why they were set in the first place?

Additionally, it seems that Intel graphics cards (not our case) need an additional volume to be shared with the host to be able to use the GPU: /dev/dri

Error if "persistantDockerImages" is not present

If persistantDockerImages is not defined in the config file, an error occurs when the server tries to get config['persistantDockerImages'].

I think that the correct behavior should be that if persistantDockerImages is not present, then we should keep all images.

Notify the client on failure.

When the server does not load because it is not on the allowed list for example the client stays on stand-by and does not get notified.

AttributeError: 'str' object has no attribute 'decode' when trying to run simulation server with GPU

Hey guys, great job on the repo so far. Here's a simple error I encountered when trying to start the servers with a GPU:

deployments-webots-1  | Traceback (most recent call last):
deployments-webots-1  |   File "/usr/local/lib/python3.8/dist-packages/tornado/web.py", line 1711, in _execute
deployments-webots-1  |     result = method(*self.path_args, **self.path_kwargs)
deployments-webots-1  |   File "simulation_server.py", line 633, in get
deployments-webots-1  |     gpu = nvmlDeviceGetName(nvmlHandle).decode('utf-8')
deployments-webots-1  | AttributeError: 'str' object has no attribute 'decode'

The fix is simple -- I just removed the call to decode, since nvmlDeviceGetName returns a str already.

Not sure if this is reproducible on other platforms: I'm running this in a Docker container on Windows 10 (x86_64) with GPU passthrough enabled via Docker Desktop on the WSL 2 engine.

If you want to make line simulation_server.py:633 more resilient, I'd suggest adding a check for a byte or str before attempting to decode.

Thanks!