How to Use Kubernetes 1.35 In‑Place Pod Resize with VPA – A Step‑by‑Step Guide

Kubernetes 1.35 introduces the GA In‑Place Pod Resize capability, allowing CPU and memory limits of running containers to be changed without pod eviction. Combined with the Vertical Pod Autoscaler (VPA) in the new InPlaceOrRecreate mode, stateful workloads such as MySQL, Redis, and Kafka can scale dynamically without downtime.

Why the update matters

Prior to 1.35, adjusting a pod’s resources required VPA to evict the pod, wait for a new pod to start, and then resume the application—an approach that can break database connections, roll back transactions, and invalidate caches. The new feature keeps the pod UID and container ID unchanged, applies changes instantly via the /resize sub‑resource, and falls back to eviction only when the node lacks sufficient capacity.

Prerequisites

Minikube (latest version) – local Kubernetes cluster

kubectl v1.35+

Git – any version to clone the VPA repository

Allocate at least 4 CPU cores and 4 GiB memory to Minikube so the VPA components can run.

Step‑by‑step implementation

Step 1: Start a Minikube cluster

minikube start --cpus=4 --memory=4096

kubectl cluster-info

Step 2: Enable Metrics Server

minikube addons enable metrics-server

kubectl top nodes

Successful output shows CPU and memory usage.

Step 3: Install VPA components

git clone https://github.com/kubernetes/autoscaler.git

cd autoscaler/vertical-pod-autoscaler

./hack/vpa-up.sh

kubectl get pods -n kube-system | grep vpa

You should see three running pods: vpa-recommender, vpa-updater, and vpa-admission-controller.

Step 4: Deploy a deliberately under‑provisioned NGINX demo

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-vpa-demo
  namespace: vpa-demo
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nginx-vpa-demo
  template:
    metadata:
      labels:
        app: nginx-vpa-demo
    spec:
      containers:
      - name: nginx
        image: nginx:latest
        resources:
          requests:
            cpu: "50m"   # intentionally low
            memory: "64Mi" # intentionally low

The low requests give VPA room to recommend higher values.

Step 5: Create a VPA in recommendation‑only mode

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: nginx-vpa
  namespace: vpa-demo
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: nginx-vpa-demo
  updatePolicy:
    updateMode: "Off"   # only recommend, no adjustment
  resourcePolicy:
    containerPolicies:
    - containerName: nginx
      minAllowed:
        cpu: "25m"
        memory: "32Mi"
      maxAllowed:
        cpu: "1"
        memory: "512Mi"
      controlledResources: ["cpu", "memory"]

After 2‑3 minutes, retrieve the recommendation: kubectl get vpa nginx-vpa -n vpa-demo -o yaml The output shows a target of 25 mCPU and 256 Mi memory, indicating the original 50 m request was higher than needed.

Step 6: Switch to InPlaceOrRecreate mode

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: nginx-vpa
  namespace: vpa-demo
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: nginx-vpa-demo
  updatePolicy:
    updateMode: "InPlaceOrRecreate"   # key mode
  resourcePolicy:
    containerPolicies:
    - containerName: nginx
      controlledValues: "RequestsAndLimits"

controlledValues: RequestsAndLimits

makes VPA adjust both request and limit while preserving their original ratio.

InPlaceOrRecreate works as follows:

Attempts in‑place resize via the /resize sub‑resource.

Pod UID stays the same; containers are recreated but the pod is not restarted.

If the node lacks resources, it falls back to eviction and recreation.

Both requests and limits are updated proportionally.

Verify the change without pod restart:

kubectl describe pod -n vpa-demo -l app=nginx-vpa-demo | grep -A 3 "Requests:"

Key observations: pod Age unchanged, Restart Count is 0, and resource requests reflect VPA’s recommendation.

Step 7: Load‑test to trigger dynamic resizing

apiVersion: v1
kind: Pod
metadata:
  name: load-generator
  namespace: vpa-demo
spec:
  containers:
  - name: load-generator
    image: busybox
    command:
    - /bin/sh
    - -c
    - |
      while true; do
        wget -q -O- http://nginx-service
      done

Monitor VPA suggestions every 30 seconds:

kubectl get vpa nginx-vpa -n vpa-demo -o jsonpath='{.status.recommendation.containerRecommendations[0].target}'

As load rises, CPU targets increase; when the load stops, resources gradually fall back. The pod never restarts.

Cleanup

# Delete the demo namespace
kubectl delete namespace vpa-demo
# Uninstall VPA
cd autoscaler/vertical-pod-autoscaler
./hack/vpa-down.sh

Key takeaways and best practices

Kubernetes 1.35 enables true in‑place resource adjustments, ideal for stateful services.

InPlaceOrRecreate first tries live resize, falling back to eviction only when necessary, preserving pod identity.

Use RequestsAndLimits to keep request/limit ratios consistent.

Suitable scenarios: databases, caches, message queues, long‑running batch jobs, and any workload sensitive to restarts.

Unsuitable for short‑lived, stateless services where traditional scaling suffices.

Ensure nodes have enough free capacity and the container runtime supports in‑place resize; validate in a test environment before production rollout.