When designing a Kubernetes cluster, we don't just need to worry about how we are going to configure our deployment objects to host our applications, or how we are going to configure our service objects to provide communication across our pods—where all this is hosted is also important. Therefore, we also need to take into account the resources that are required to bring balance to our application workloads and our control plane.
Kube-sizing
etcd considerations
We will require at least a three-node etcd cluster in order for it to be able to support itself in case one node fails. Because etcd uses a distributed census algorithm called Raft, odd-numbered clusters are recommended. This is because, in order for an...