I am very new to JupyterHub. Please forgive me, if my questions are stupid. I have successfully started “The littlest JupyterHub”, and I am still struggling with the “JuypterHub on Kubernetes” docs. My goal is something in between:
I want to run JupyterHub on some tiny host (a small vserver, or a micro EC2 instance), because it is typically accessed only once per week, and I want to avoid costs. This tiny JupyterHub should spawn notebook servers using AWS Kubernetes (EKS), because every now and then I need hundreds of notebooks.
Is such a setup possible? Can I find a description somewhere, how to do the setup?
Thanks for asking. I was not aware of the costs of an EKS cluster without worker nodes. I thought I’d only pay for the worker nodes.
Yes, after finding that an unused cluster costs 73$ per month, I’ll probably try to spin up a cluster on the first request and delete it when it is idle for a few hours. I’ll try to find out what I have to change in JupyterHub to make this happen.
Or maybe you have a better suggestion?
EKS can also run with Fargate nodes, this might allow you to use Z2JH without paying for a full EKS cluster- since the JupyterHub hub and proxy pods don’t need much resources you can leave them running on EKS Fargate. This isn’t something I’ve looked into though so I might be wrong, but if Z2JH works with EKS Fargate that’s a very nice option for running a scalable cluster using the standard Helm chart.
If you research this please report back, it’d be very useful for others!