Kubernetes Clusters

Kubernetes is the Operating System of the cloud. But running the OS involves two parts: the Control Plane (The Master) and the Data Plane (The Workers).

Managed vs. Self-Hosted

• Hard Mode (Kops/Kubeadm): You manage the Control Plane (etcd, api-server) on EC2 instances. Don't do this unless you are a bank or a massive tech corp.
• Easy Mode (EKS/GKE/AKS): AWS/Google manages the Control Plane. You just manage the Worker Nodes.

---

Node Pools (Data Plane)

Organize your worker nodes into groups based on hardware.

1. General Purpose: M5/T3 instances. For web apps.
2. Compute Optimized: C5. For batch processing.
3. Spot Instances: Cheap, unreliable instances. Great for stateless workers, but you must handle interruptions (Graceful Shutdown).

---

Upgrade Strategy

Upgrading Kubernetes is terrifying. It forces a restart of every container in your cluster.

The Blue/Green Node Strategy:

1. You are on Version 1.26 (Node Group A).
2. Spin up a new Node Group B on Version 1.27.
3. Taint Group A so no new pods land there.
4. Drain Group A. This evicts pods, forcing them to reschedule onto Group B.
5. Once empty, delete Group A.