Comparison
Control planevsData plane
Control plane
the API that decides what should run went down, and everything already running carried on serving traffic perfectly.
The set of components that hold the desired state and decide what should happen — the API, the store behind it, the schedulers and controllers. It is where you make changes and where policy is enforced, and it is deliberately not in the request path of your users' traffic. Its failure mode is distinctive: nothing breaks immediately, but nothing can heal, scale or deploy either, so the outage becomes visible the moment something else fails.
Full entry →Data plane
the nodes running your containers and the proxies moving the packets keep working whether or not anyone is watching them.
The components that actually carry the work: the nodes running containers, the proxies forwarding requests, the storage serving reads. It is the part users experience, and it is designed to keep running from its last known instructions when the control plane is unavailable. Separating the two planes is the reason a managed cluster's upgrade can be boring and the reason a control-plane incident is measured in inability to change rather than in downtime.
Full entry →