Wasn't that a ride?
Kubernetes relies heavily on available resources spread throughout the cluster. Still, it cannot do magic. We need to help it out by defining resources we expect our containers will consume.
Even though Heapster is not the best solution for collecting metrics, it is already available in our Minikube cluster, and we used it to learn how much resources our applications use and, through that information, we refined our resource definitions. Without metrics, our definitions are pure guesses. When we guess, Kubernetes needs to guess as well. A stable system is a predictable system based on facts, not someone's imagination. Heapster helped us transform our assumptions into measurable facts which we fed into Kubernetes which, in turn, used them in its scheduling algorithms.
Exploration of resource definitions led us to Quality Of Service (QoS)...
