Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Get fastplotlib working in a docker container and binder or jupyterhub #398

Closed
kushalkolar opened this issue Feb 6, 2024 · 3 comments
Closed

Comments

@kushalkolar
Copy link
Member

kushalkolar commented Feb 6, 2024

It's important for us to fully figure out how to use fastplotlib in docker on machines with a GPU. @apasarkar has been battling this for some time and so far we suspect the issue is that the nvidia-docker container does not contain the necessary graphics libraries (libglx etc. and related libs) to run vulkan applications. We also think libvulkan1 is required. We've tried installing some of these into the container with no luck so far (will post more details soon). We also suspect that the vulkan drivers might need to be installed on the host in order for the container to be able to use them.

Also tagging @billbrod since he has figured out Flatiron's binder and this is important for the FENS workshop.

related #376

@billbrod
Copy link

billbrod commented Feb 6, 2024

I wrote my notes for getting the nemos workshop working on Flatiron's binder on the readme. Binder recommends against specifying the Dockerfile explicitly, and instead using some combination of the other config files they support.

I ended up getting it working with two conda environment files (environment.yml and environment-cuda.yml), plus a postBuild bash script. Binder sees the environment.yml file, uses conda to install it (which includes python, pip, cuda), then runs the postBuild script, in which the first line installs is CONDA_OVERRIDE_CUDA="11.8" conda env update -p $NB_PYTHON_PREFIX --file environment-cuda.yml (this second file includes jax and all our other requirements; the rest of postBuild does nemos-specific things: downloading the data, building the notebooks, and moving things around). Being able to set CONDA_OVERRIDE_CUDA was the key to getting it working, putting everything in a single environment file didn't install jax with cuda support.

If you want to just have a binder that works on the Flatiron cluster, this should suffice for you, but if you're looking for a general docker solution, it's probably not enough.

@kushalkolar
Copy link
Member Author

Caitlin has got this working in docker with this, leads me to think if there's something about host drivers that we're missing:

https://github.com/hantman-lab/animal-soup/blob/main/Dockerfile

@kushalkolar
Copy link
Member Author

Closing this since we've been using it on the Flatiron cluster for a while now, it's a matter of having all the graphics libs installed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants