VPA CPU Memory in Action Demo

Welcome — this lesson runs a full Vertical Pod Autoscaler (VPA) lab so you can observe recommendations and automated updates in action. We’ll deploy a sample Flask app, apply a VPA, generate load, and watch how the updater accepts recommendations and evicts pods to apply new resource requests. High-level sequence:

Deploy the sample Flask application (Deployment + Service).
Apply a VPA with updateMode: Off to collect recommendations only.
Generate load against the app so VPA can observe usage.
Inspect the VPA updater logs while Off to confirm no evictions occur.
Switch the VPA to Auto (or Recreate) to enable automatic application.
Watch the updater evict pods and confirm new pods run with updated requests.

A slide titled "Lab Steps" showing a vertical timeline with four numbered tasks: 01 Deploy Application, 02 Deploy VPA Configuration, 03 Initiate Load Test, and 04 Monitor Logs. The left side has a teal gradient panel with the title and the right side lists the steps with colorful numbered markers.

Summary of the demo flow:

Step	Command / Artifact	Purpose
1	`kubectl apply -f /root/vpa-cpu-app.yml`	Deploy Flask Deployment + Service so the app is reachable.
2	`kubectl apply -f /root/vpa-flask-app-vpa.yml`	Apply VPA with `updateMode: Off` to gather recommendations.
3	`sh load.sh`	Generate sustained load so VPA can build resource observations.
4	`kubectl logs -f vpa-updater-<pod> -n kube-system`	Inspect updater behavior while updates are disabled.
5	`kubectl edit vpa flask-app` (change to `Auto`) or run `kubectl get vpa flask-app -o jsonpath='{.spec.updatePolicy.updateMode}'`	Enable automatic application of recommendations.
6	`kubectl get pods`	Confirm pods are recreated and new pods reflect VPA recommendations.

For more on VPA concepts, see the Kubernetes docs: Vertical Pod Autoscaler.

A presentation slide titled "Lab Steps" with a blue gradient panel on the left. On the right are two numbered steps: "05 Update VPA Configuration" and "06 Monitor Logs to see VPA in action."

1) Deploy the sample app

Apply the manifest for the sample Flask application. The manifest creates both a Deployment and a Service so the app can be load-tested.

kubectl apply -f /root/vpa-cpu-app.yml

Expected output:

deployment.apps/flask-app created
service/flask-app-service created

Confirm the Deployment and Service are created and pods are running before proceeding.

2) VPA configuration (initial)

Create a VPA that targets the flask-app Deployment. Start with updateMode: Off so the VPA only computes and exposes recommendations without evicting pods.

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

This manifest:

Targets the flask-app Deployment.
Uses updateMode: Off so the updater will not evict pods (only recommendations are generated).
Sets minAllowed and maxAllowed to constrain suggestions.
Controls cpu and memory resources.

Apply the VPA manifest:

kubectl apply -f /root/vpa-flask-app-vpa.yml

Inspect VPA recommendations and bounds:

kubectl describe vpa flask-app

Sample (truncated) kubectl describe vpa output:

Recommendation:

    Container Recommendations:

        Container Name:  flask-app

        Lower Bound:

            Cpu:  100m

        Target:

            Cpu:  100m

        Uncapped Target:

            Cpu:  25m

        Upper Bound:

            Cpu:  800m

At this stage the updater provides recommendations but does not evict or modify pods.

3) Start a load test

Generate traffic so the VPA collects realistic resource usage metrics. For this lab the provided script is used:

sh load.sh

Sample console output (repeated responses):

Hello, World!Hello, World!Hello, World!Hello, World!Hello, World!Hello, World!

Allow the load to run for several minutes so VPA accumulates observations and refines its recommendations.

4) Inspect the VPA updater logs (while updateMode is Off)

While updateMode remains Off, confirm the updater is collecting data but not evicting pods. Identify the updater pod name (example shown below uses vpa-updater-59469d986c-gw5jf — replace with your pod name):

kubectl logs -f vpa-updater-59469d986c-gw5jf -n kube-system

Example logs showing the updater skipping VPA objects that aren’t in an active update mode:

I0104 12:28:33.901166        1 updater.go:150] skipping VPA object default/flask-app because its mode is not "Recreate" or "Auto"
W0104 12:28:33.901197        1 updater.go:166] no VPA objects to process

VPA behavior is split into two concerns: (1) computing recommendations based on observed usage, and (2) applying those recommendations by evicting pods so new pods start with updated requests. updateMode: Off disables automatic application; use Auto or Recreate to enable eviction and automatic application.

5) Enable automatic updates

When you’re ready for the VPA to apply recommendations, change updateMode to Auto (or Recreate if you require a different eviction behavior). Editing in-place:

kubectl edit vpa flask-app

Change the updatePolicy block to:

updatePolicy:
  updateMode: "Auto"

Verify the change:

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

Expected output:

Auto

Enabling Auto (or Recreate) will allow the VPA updater to evict pods to apply recommended resource requests. Be prepared for transient pod restarts and possible brief disruptions to service. Use during maintenance windows if running in production.

6) Watch the updater apply recommendations

With Auto enabled and load in effect, the updater will accept recommendations, evict selected pods, and let the Deployment create new pods with the updated resource requests. Tail the updater logs again to observe this flow:

kubectl logs -f vpa-updater-59469d986c-gw5jf -n kube-system

Example log lines showing accepted recommendations and eviction events:

I0104 12:39:33.906873       1 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 12:39:33.906922       1 update_priority_calculator.go:146] pod accepted for update default/flask-app-67b666c5fc-fpw77 with priority 586 - processed recommendations:
flask-app: target: 587m; uncappedTarget: 587m;

I0104 12:39:33.906962       1 updater.go:228] evicting pod default/flask-app-67b666c5fc-k9rtg

I0104 12:39:33.919716       1 event.go:298] Event(v1.ObjectReference{Kind:"Pod", Namespace:"default", Name:"flask-app-67b666c5fc-k9rtg", UID:"ea3379ed-eeef-4b2e-b0b4-9ef30e3fe009", APIVersion:"v1", ResourceVersion:"4137", FieldPath:""}): type: 'Normal' reason: 'EvictedByVPA' Pod was evicted by VPA Updater to apply resource recommendation.

These lines show the updater computed a target CPU (~587m) and evicted pods so the Deployment could recreate them with the new resource requests.

7) Confirm new pods are running with updated resource requests

After evictions and recreations, list pods and verify ages and resource requests. New pods created after eviction will have a low AGE value:

kubectl get pods

Example output:

NAME                          READY  STATUS   RESTARTS  AGE
flask-app-67b666c5fc-41b5s    1/1    Running   0        43s
flask-app-67b666c5fc-96jf9    1/1    Running   0        103s

To inspect the resource requests of a pod’s containers, describe the pod or check the Deployment’s pod template after the VPA has updated the requests:

kubectl describe pod flask-app-67b666c5fc-41b5s
# or
kubectl get deployment flask-app -o yaml

Look for the resources.requests fields to verify they reflect the VPA recommendations.

That completes the end-to-end demo. You can now practice applying VPA manifests, generating load, and watching the updater behavior. For additional reading, consult:

Kubernetes VPA in the Autoscaler repo: https://github.com/kubernetes/autoscaler/tree/master/vertical-pod-autoscaler
Kubernetes documentation: https://kubernetes.io/docs/

Manual Scaling

Horizontal Pod Autoscaler (HPA)

Vertical Pod Autoscaler (VPA)

Cluster Proportional Autoscaler (CPA)

Kubernetes Event-Driven Autoscaling (KEDA)

Introduction

VPA CPU Memory in Action Demo

1) Deploy the sample app

2) VPA configuration (initial)

3) Start a load test

4) Inspect the VPA updater logs (while updateMode is Off)

5) Enable automatic updates

6) Watch the updater apply recommendations

7) Confirm new pods are running with updated resource requests

Watch Video

Practice Lab

​1) Deploy the sample app

​2) VPA configuration (initial)

​3) Start a load test

​4) Inspect the VPA updater logs (while updateMode is Off)

​5) Enable automatic updates

​6) Watch the updater apply recommendations

​7) Confirm new pods are running with updated resource requests

Watch Video

Practice Lab

1) Deploy the sample app

2) VPA configuration (initial)

3) Start a load test

4) Inspect the VPA updater logs (while updateMode is Off)

5) Enable automatic updates

6) Watch the updater apply recommendations

7) Confirm new pods are running with updated resource requests