Kubernetes Core Concepts

核心 · Key Idea

In one line: Kubernetes is cluster-level container orchestration. You write desired state (YAML) and its control loops continuously reconcile reality toward that state. If concepts feel overwhelming, start by memorizing 5 objects: Pod / Deployment / Service / Ingress / Namespace.

What it is#

# Deployment: 3 replicas of a web app
apiVersion: apps/v1
kind: Deployment
metadata: { name: web }
spec:
  replicas: 3
  selector: { matchLabels: { app: web } }
  template:
    metadata: { labels: { app: web } }
    spec:
      containers:
        - name: app
          image: ghcr.io/me/web:1.2.0
          ports: [{ containerPort: 8080 }]
          resources:
            limits: { cpu: 500m, memory: 512Mi }
            requests: { cpu: 100m, memory: 128Mi }
---
# Service: stable DNS name for these Pods
apiVersion: v1
kind: Service
metadata: { name: web }
spec:
  selector: { app: web }
  ports: [{ port: 80, targetPort: 8080 }]

Analogy#

打个比方 · Analogy

You don't tell a worker "lay this brick" (imperative). You hand a blueprint (YAML) to a project manager (controller), who has a crew constantly inspecting: missing brick → add it; misaligned → straighten; collapsed → rebuild — that's declarative + control loops.

Five most-used objects#

PodPod

Smallest scheduling unit, one or more tightly-coupled containers sharing network namespace and volumes.

DeploymentDeployment

Manages stateless Pod lifecycles — rolling upgrades / rollbacks.

ServiceService

Stable DNS name + virtual IP fronting a set of Pods.

IngressIngress

L7 routing — maps external hostnames / paths to Services (implemented by nginx / Traefik / etc.).

NamespaceNamespace

Logical partition inside the cluster (dev / prod / kube-system).

How it works#

K8s is fundamentally a pile of controllers watching the API server and reconciling.

Key commands#

kubectl get pods -A                            # all namespaces
kubectl describe pod web-xxxxx
kubectl logs -f web-xxxxx
kubectl exec -it web-xxxxx -- sh
kubectl apply -f manifest.yaml
kubectl rollout status deploy/web
kubectl rollout undo deploy/web
kubectl top pod
kubectl explain deployment.spec.template.spec.containers   # field docs

Practical notes#

Always set resource requests/limits — scheduler uses requests; limits prevent one Pod from saturating the node.
Don't kubectl exec to fix things in place — edit YAML → apply, otherwise the next rollout wipes your fix.
Three probes: livenessProbe (alive?), readinessProbe (accepting traffic?), startupProbe (slow boot tolerance).
Pods are ephemeral — can be evicted or rescheduled at any time. Apps must be stateless or store state in PVC / external DB.
Rolling updates: default RollingUpdate, controlled via maxUnavailable / maxSurge.
HPA: kubectl autoscale deploy/web --cpu-percent=70 --min=2 --max=10.
Learning ladder: minikube / k3s → deploy a real app → learn Helm → then service mesh / operators.

Easy confusions#

Pod

K8s scheduling unit, **may contain 1–N tightly-coupled containers**.
Shared network + volumes.

Container

A running instance inside a Pod.
One element in the `containers:` array.