Comparison
ClustervsNode
Cluster
you stopped addressing individual machines and started submitting work to a pool that decides where it lands.
A set of nodes under one control plane, presented as a single place to run things. The unit matters more than it looks: a cluster is a blast radius, an upgrade unit, a security boundary of sorts and a billing boundary, so 'how many clusters' is one of the more consequential platform decisions a team makes. One big cluster is cheaper and packs better; many small ones fail independently and let teams upgrade on their own schedule.
Full entry →Node
a machine joins the pool, advertises how much CPU and memory it has, and starts being given other people's workloads.
One machine — usually a virtual one — that has joined a cluster and runs an agent taking work from the control plane. Almost everything about capacity, cost and failure domains resolves to nodes: how many, how big, in which zones, and how quickly a replacement appears. The reason node size is an argument is that big nodes bin-pack better and lose more when one dies.
Full entry →