Max CPU utilization on heterogeneous Kubernetes cluster with single-core process
Am I then limited by the smallest node in a heterogeneous cluster of nodes? If not, can I enforce choosing the node largest CPU (which would then limit me to the single node) or smallest CPU (which would then allow seamless and consistent parallel execution) or will Kubernetes do it in the background?
The scheduler does not have insight about what instances you use. So it will schedule the app to any node.
You can use Taints and Tolerations to classify the nodes and then you can declare to what kind of nodes you want your app to run on by setting tolerations. Or alternatively use nodeSelector and labels on nodes.