Control PlaneAmazon EKS fieldbook
Amazon EKS · Platform ArchitectureView Markdown source

Kubernetes in simple terms

Kubernetes is a continuous correction machine. You declare desired state. Controllers compare it with actual state and keep trying to close the gap.

The smallest useful vocabulary

Object Plain meaning
Pod Smallest scheduled unit; normally one application container
Deployment Keeps stateless Pods running and rolls out versions
StatefulSet Stable names and ordered behavior for stateful Pods
DaemonSet Runs one Pod on selected nodes
Job / CronJob Finite or scheduled work
Service Stable address for changing Pods
Gateway / Ingress Routes application traffic
Namespace Scope for names, quota, RBAC, and policy
ConfigMap / Secret Configuration and sensitive values

Labels identify objects. Selectors choose objects. A wrong selector can send traffic to the wrong Pods or leave a controller managing nothing.

Requests, limits, and health

A resource request is what the scheduler reserves. It drives placement, capacity, and cost. A limit is an upper boundary. Measure both; guesses create waste, CPU throttling, or memory kills.

  • Startup probe: “I am still starting.”
  • Readiness probe: “Do not send me traffic yet.”
  • Liveness probe: “I am stuck and a restart can fix me.”

Do not make liveness depend on a slow database. Restarting every Pod during a database incident makes the incident larger.

Feynman check

If three of five replicas disappear, a Deployment sees that actual state is smaller than desired state and creates three replacements. Kubernetes restores objects; it does not restore lost business data.

Control PlaneIndependent study material · verify production details in AWS documentation