How can we implement multiple Kubernetes worker pools with
Mulitple Kubernetes Node pools can be defined in the
kops manifests (YAML).
How can we implement multiple worker pools within our
kops Kubernetes clusters?
We define multiple node pools in the
kops manifest (YAML) to define imperative Kubernetes architectures.
Inside the manifest, which looks like a standard Kubernetes resource, is a section for
kind: InstanceGroup that allows for the definition of any number of node pools, each with its own characteristics like instance type.
An example manifest is provided here.
An example of a project with the Van Valen Research Lab at Caltech demonstrates where a GPU node pool was added that is scaled down to zero by default.
Using the cluster autoscaler, when a pod is scheduled with the proper labels, the node pool is automatically scaled up.