Kubernetes

• Key points are in the visual diagram above.

A single Docker container can't survive node crashes, roll out zero-downtime updates, or scale to 50 replicas during a product launch. K8s manages the lifecycle of agent pods, load balances traffic, and integrates with secrets, config maps, and autoscaling.

Kubernetes is an air-traffic control tower for containers — it decides which runway (node) each plane (pod) uses, reroutes when one goes down, and adds more gates when passenger volume spikes.

Scenario

Deploy an agent API as a Deployment with 3 replicas, expose via ClusterIP Service, route `api.example.com/agent` through Ingress with TLS, and HPA scales 3→20 pods when p95 latency exceeds 2s.

Solution

In Agent Runtime & Production, apply Kubernetes to this scenario: Deploy an agent API as a Deployment with 3 replicas, expose via ClusterIP Service, route `api. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Kubernetes (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Kubernetes happens in the code.

1# deployment.yaml (excerpt)2apiVersion: apps/v13kind: Deployment4metadata:5  name: agent-api6spec:7  replicas: 38  template:9    spec:10      containers:11        - name: agent12          image: myregistry/agent-api:v1.2.013          ports:14            - containerPort: 800015          readinessProbe:16            httpGet:17              path: /health18              port: 800019          envFrom:20            - secretRef:21                name: openai-key

• Treating Kubernetes as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for kubernetes

Quick recap — the most important points from this module.