IT 3300 : Virtualization

Kubernetes — Introduction

What is Kubernetes?

Open-source system to deploy, scale, and self-heal containerized
applications across a cluster of machines.

You declare the desired state; Kubernetes works to make reality match.

Why not just Docker/Compose?

• Compose runs an app on one host
• What about many hosts, failover, scaling, rolling updates?
• Kubernetes (k8s) orchestrates containers across a whole cluster

What it gives you

• Self-healing — restarts failed containers, replaces dead ones
• Scaling — run N replicas; scale up/down on demand
• Load balancing — spread traffic across replicas
• Rollouts/rollbacks — update with no downtime, undo cleanly
• Bin packing — fit workloads onto nodes by CPU/memory

Declarative desired state

• You describe what you want (3 replicas of this image)
• k8s figures out how to get and keep it there
• A control loop constantly reconciles desired vs. actual
• Same mindset as Compose, scaled to a cluster

Cluster anatomy

• Control plane — the brain (schedules, tracks state)
• Worker nodes — where your containers actually run
• You talk to the cluster with kubectl

Control plane pieces

• API server — the front door; everything goes through it
• etcd — the cluster's database (desired + actual state)
• scheduler — decides which node runs a new pod
• controller manager — runs the reconcile loops

Node pieces

• kubelet — the agent that runs pods on the node
• container runtime — actually runs containers
• kube-proxy — wires up service networking

The runtime changed — important

• Kubernetes no longer uses Docker on nodes
• dockershim was removed in k8s 1.24
• Nodes now run containerd (or CRI-O) via the CRI interface
• Your Docker-built OCI images still run perfectly

Ways to run a cluster

• k3s — lightweight, great for homelab/edge
• MicroK8s — simple snap install, easy add-node
• kind / minikube — local dev on a single machine
• GKE / EKS / AKS — managed cloud (Unit 4)

Key objects (preview)

• Pod — one or more containers, the smallest unit
• Deployment — manages replicas, scaling, rollouts
• Service — stable network endpoint for pods
• ConfigMap / Secret — configuration and credentials

Lab goals

• Read one Kubernetes case study; note the problem it solved
• Be able to name the control-plane and node components
• Explain, in a sentence, "declarative desired state"