Automated JupyterHub deployment testing

[kabilar]

Requirements

• [ ] Test configuration of GPU nodes (e.g. install GPU drivers, etc.)
• [ ] Test scaling of deployment for multiple instance types

Possible implementation tool

• Selenium

[asmacdo]

To verify GPU envs

pip install torch torchvision torchaudio

python  test-cuda.py

import torch
if torch.cuda.is_available():
    print ("SUCCESS!")

[asmacdo]

Some kind of CI testing around autoscaling would be nice, maybe we can test how it reacts by watching Karpenter CRDs while disabling the Karpenter reconcile loop? https://github.com/dandi/dandi-hub/issues/129#issuecomment-2020746330 Then we wouldn't need to actually pay for the Nodes to be created.

[asmacdo]

Test Case: 3 users creating a "Base" profile should all fit on 1 node.