We have jupyterhub configured on kubernetes with Amazon AWS. In it, we have multiple profiles associated with different autoscaling groups that spawn different hardware configurations, and pods get allocated to those using taints. We autoscale when a new user requests a server and all nodes in that group are being used. We have no problem autoscaling for autoscaling groups with a min of 1 and above, but haven’t been able to trigger the autoscale up from 0 or down to 0.
Following the instructions in this link we have added the following tags to our autoscaling group:
Given our taints for pods to be on this group:
- key: “hub.jupyter.org/dedicated”
- key: “GPUJobExclusivity”
- key: “GPUtype”
- key: “numberOfGPUs”
Kubernetes was set up following this link
Kubernetes version: v1.14.1
Cloud being used: AWS
Host OS: Cent OS
Any idea what we could be missing?