Summary - HPA

Combine YAML, CPU threshold and limits to write an HPA.

Summary: HPA

In this group:

An HPA calculates the replica count with ceil(current × value / target) and clips it to the min–max range.
kubectl autoscale or autoscaling/v2 YAML: scaleTargetRef, minReplicas, maxReplicas, metrics.
The CPU percentage is relative to requests.cpu; without a request, <unknown>.
behavior adjusts the scaling speed and the scale-down stabilization window.
metrics-server is needed; for custom metrics the Prometheus Adapter or KEDA is used.

kubectl autoscale deploy api --cpu-percent=70 --min=2 --max=10
kubectl get hpa -w

Exercise CPU and memory metrics

Write an HPA's metrics section (starting with metrics:): for CPU type: Utilization, averageUtilization: 70; for memory type: AverageValue, averageValue: 400Mi.

Show example solution

metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 70
  - type: Resource
    resource:
      name: memory
      target:
        type: AverageValue
        averageValue: 400Mi

Test yourself

The HPA's TARGETS column shows cpu: <unknown>/70%. What are the two most likely causes?
- metrics-server is missing or requests.cpu isn't written in the containers
- maxReplicas is too low
- The image is wrong
- The namespace is wrong
Show answer

metrics-server is missing or requests.cpu isn't written in the containers — Check metrics-server with kubectl top pods, and the error with describe hpa.

Give an average-value target for memory.

        target:
          type: ____
          averageValue: 400Mi

Show answer

        target:
          type: AverageValue
          averageValue: 400Mi

Match.

HPA

Horizontal
VPA

Vertical
metrics-server

Metric source
behavior

Speed setting

Next lesson: What is RBAC?

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies