Controller Archetypes
This chapter explains the different types of controller archetypes.
The Kubernetes architecture is API centric and not API driven. The business logic runs outside the API server, in loosely coupled controllers which bi-directionally map the virtual representation of digital twins
We identify three archetypes for controllers and will explain them using Kubernetes native controllers Kubelet, Scheduler, ReplicaSet, Deployment, and their respective resource types or kinds.
Actuators
Actuators are controllers which map real-world capabilities into the virtual-world (with digital twins). They bring capabilities into the realm of the Kubernetes API and explain the vast expansion of the community and ecosystem beyond containerized workloads only. Crossplane, ExternalDNS, and Kubelet are examples of the actuator archetype.
The Kubelet controller is a fundamental component of Kubernetes. It acts as the node agent on each worker node. Each Kubelet first registers its node in the Nodes space and then watches for Pods that are assigned to it, essentially sharding the Pods space.
Aspect or Attribute Controllers
Aspect or attribute controllers operate in the virtual-world, on existing digital twins. Their core business logic function is to map a desired aspect by enriching (calculating) attributes of digital twins which are covered/managed by other controllers.
Besides Scheduler, for example, policy controllers such as Gatekeeper or Kyverno operate in this archetype category.
The Scheduler controller is a pivotal default component in the Kubernetes control plane, responsible for the assigning of pods to nodes. It reads available Node digital twin objects and keeps a watch on all unscheduled Pod objects. The binding process to compute a suitable node can be configured.
Ultimately, the Scheduler adjusts the Pod:spec:nodeName attribute of prior unscheduled Pods with its node assignment decision. This controller archetype obtains all information from the digital twins and only operates on digital twins.
Logical Object Controllers
Logical object controllers operate in the virtual-world only, but they introduce new logical types. Their business logic function is to create and manage useful abstractions. All custom controllers or operators (for example on OperatorHub) fall into this archetype.
They introduce new digital twin abstractions and often productize their offering using other abstractions, which at the end of the chain, typically spawn Kubernetes Pods with their workload.
The ReplicaSet and Deployment controller are prime example of useful abstractions for the orchestration of containers. ReplicaSet introduces a new digital twin type which creates and manages Pod objects with the intent to ensure that a stable set of replicas are running at any given time. And Deployment introduces a new type by instrumenting ReplicaSet to offer declarative lifecycle management (via rolling updates, rollback, and scaling).
