VPA CPU Memory in Action Demo

In this hands-on guide, you’ll learn how to configure Kubernetes’ Vertical Pod Autoscaler (VPA) for both CPU and memory. We’ll cover:

1. Deploying a sample Flask application
2. Applying a VPA manifest with updates disabled
3. Generating load and inspecting VPA Updater logs (no evictions yet)
4. Enabling automatic updates and observing pod evictions and resource adjustments

1. Deploy the Sample Application

Apply the Deployment and Service for our Flask app:

kubectl apply -f /root/vpa-cpu-app.yml
# deployment.apps/flask-app created
# service/flask-app-service created

Verify the pods and service are running:

kubectl get pods,svc | grep flask-app

2. Review and Apply the VPA Configuration

Inspect the VPA manifest below. The updateMode is set to Off, so no live adjustments will occur:

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: flask-app
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: flask-app
  updatePolicy:
    updateMode: "Off"
  evictionRequirements:
    - resources: ["cpu", "memory"]
      changeRequirement: TargetHigherThanRequests
  resourcePolicy:
    containerPolicies:
      - containerName: "*"
        minAllowed:
          cpu: 100m
          memory: 100Mi
        maxAllowed:
          cpu: 800m
          memory: 500Mi
        controlledResources: ["cpu", "memory"]

Apply the VPA object:

kubectl apply -f /root/vpa-cpu-vpa.yml
# verticalpodautoscaler.autoscaling.k8s.io/flask-app created

Describe the Initial Recommendation

With no traffic and updateMode: Off, the VPA will only show its configured bounds:

kubectl describe vpa flask-app
# Recommendation:
#   Container Name:  flask-app
#   Lower Bound:     Cpu: 100m
#   Target:          Cpu: 100m
#   Uncapped Target: Cpu: 25m
#   Upper Bound:     Cpu: 800m

3. Generate Load Without Evictions

Start a load test to drive up CPU usage:

sh load.sh
# Hello, World!Hello, World!Hello, World!...

Now tail the VPA Updater logs in the kube-system namespace. Since updates are disabled, you’ll see skip messages:

kubectl logs -f deployment/vpa-updater -n kube-system
# I0104 12:28:33.901166 updater.go:150] skipping VPA object default/flask-app because its mode is not "Recreate" or "Auto"
# W0104 12:28:33.901197 updater.go:166] no VPA objects to process

4. Enable Automatic Updates

Switch the VPA into Auto mode so it can evict and restart pods with new resource requests:

kubectl edit vpa flask-app
# Change updatePolicy.updateMode from "Off" to "Auto"

Confirm the change:

kubectl get vpa flask-app -o=jsonpath='{.spec.updatePolicy.updateMode}'
# Auto

5. Observe Pod Evictions & Resource Adjustments

With load still running, watch the Updater logs:

kubectl logs -f deployment/vpa-updater -n kube-system
# I0104 12:39:33.906873 update_priority_calculator.go:146] pod accepted for update default/flask-app-67b666c5fc-k9rtg with priority 586 - processed recommendations: flask-app: target: 587m; uncappedTarget: 587m;
# I0104 12:39:33.906962 updater.go:228] evicting pod default/flask-app-67b666c5fc-k9rtg
# I0104 12:39:33.919716 event.go:298] Event(...): type: 'Normal' reason: 'EvictedByVPA' Pod was evicted by VPA Updater to apply resource recommendation.

Check the updated pods and their ages:

kubectl get pods
# NAME                                READY STATUS    RESTARTS AGE
# flask-app-67b666c5fc-41b5s         1/1   Running   0        43s
# flask-app-67b666c5fc-96jfj         1/1   Running   0        103s

You should now see new pods running with the VPA-recommended CPU and memory requests. This completes the VPA demo.

Links and References

• Vertical Pod Autoscaler Overview
• VPA GitHub Repository

Further Reading

• Kubernetes Resource Management
• Autoscaling in Kubernetes