Lab 06: Autoscaling an OKS Cluster

Lab Objectives

In this lab, you will learn how to enable and test autoscaling on a Kubernetes cluster deployed with OKS.
The goal is to observe how the cluster automatically adds or removes nodes based on workload.

Lab Architecture

By the end of this lab, you will be able to:

• Configure a NodePool with autoscaling enabled.
• Deploy a variable workload on the cluster.
• Observe scale-up and scale-down behavior.

Environment Preparation

Before starting this lab, make sure you have:

• An OKS cluster.
  👉 Click here to access the OKS cluster creation lab if not already done.

Creating the NodePool with Autoscaling

⚠️ Important:

• Autoscaling is only supported for single-Subregion (single-AZ) NodePools.
• If your NodePool is deployed across multiple subregions (multi-AZ), the autoscaling: true option will not work.
• Nodes using local volumes (e.g., mounted via hostPath) are excluded from autoscaling and upgrade operations to prevent data loss.

---

Once your OKS cluster is up and running, we will create a NodePool configured to automatically adapt to workload.

Step 1: Create the NodePool YAML file

Create a file named nodepool.yaml with the following content:

apiVersion: oks.dev/v1beta2
kind: NodePool
metadata:
  name: nodepool-01
spec:
  desiredNodes: 2
  minNodes: 2
  maxNodes: 4
  autoscaling: true
  nodeType: tinav7.c2r4p1
  zones:
  - eu-west-2a
  rootVolume:
    storageType: gp2
    size: 20
  upgradeStrategy:
    maxUnavailable: 1
    maxSurge: 0
    autoUpgradeEnabled: true
    autoUpgradeMaintenance:
      durationHours: 1
      startHour: 12
      weekDay: Tue
  autoHealing: true

Apply the configuration to the cluster:

kubectl apply -f nodepool.yaml

Step 2: Verify the NodePool creation

List the created NodePools:

kubectl get nodepools.oks.dev

Example output:

NAME          TYPE             DESIRED NODES   ATTACHED   READY   RUNNING   PENDING   CURRENT PROCESSING PHASE
nodepool-01   tinav7.c2r4p1    2               2          2       2         0         idle

Also verify the cluster nodes:

kubectl get nodes

Example output:

NAME              STATUS   ROLES    AGE   VERSION
ip-10-50-34-159   Ready    <none>   12d   v1.32.5
ip-10-50-42-1     Ready    <none>   12d   v1.32.5

Testing Autoscaling with Real CPU Load

In this section, we will deploy an application that consumes CPU resources to observe the autoscaling behavior of the OKS cluster.

This method uses the vish/stress image to simulate CPU and memory load on each pod.

---

Step 1: Create the stress-test.yaml file

Create a file named stress-test.yaml with the following content:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: stress-test
spec:
  replicas: 7
  selector:
    matchLabels:
      app: stress
  template:
    metadata:
      labels:
        app: stress
    spec:
      containers:
      - name: stress
        image: vish/stress
        args: ["-cpus", "2"]
        resources:
          requests:
            cpu: "1500m"
            memory: "512Mi"
          limits:
            cpu: "2000m"
            memory: "1Gi"

Apply the deployment:

kubectl apply -f stress-test.yaml

Once the NodePool is ready, we will validate autoscaling by generating load.

Step 2: Observe cluster behavior

Monitor pod creation:

kubectl get pods

Example output:

NAME                           READY   STATUS    RESTARTS   AGE
stress-test-7dc77b76f5-22phs   1/1     Running   0          2m11s
stress-test-7dc77b76f5-d96bc   0/1     Pending   0          2m11s
stress-test-7dc77b76f5-gvqjc   1/1     Running   0          2m11s
stress-test-7dc77b76f5-nwbwp   0/1     Pending   0          2m11s
stress-test-7dc77b76f5-p9lsw   1/1     Running   0          2m11s
stress-test-7dc77b76f5-qg9lf   1/1     Running   0          2m11s
stress-test-7dc77b76f5-tls8c   1/1     Running   0          2m11s

Then monitor the number of nodes:

kubectl get nodepools.oks.dev

Example output:

NAME          TYPE             DESIRED NODES   ATTACHED   READY   RUNNING   PENDING   CURRENT PROCESSING PHASE
nodepool-01   tinav7.c2r4p1    4               3          3       3         0         idle
nodepool-01   tinav7.c2r4p1    4               3          3       3         0         reconciliation
nodepool-01   tinav7.c2r4p1    4               3          3       4         1         reconciliation

Testing Scale-Down

After verifying that scale-up works correctly, we will now test automatic scale-down.

---

Delete the deployment to remove the simulated CPU load:

kubectl delete -f stress-test.yaml

Monitor the cluster and node count:

kubectl get nodes

Example output:

NAME              STATUS   ROLES    AGE   VERSION
ip-10-50-33-211   Ready    <none>   14m   v1.32.5
ip-10-50-34-159   Ready    <none>   12d   v1.32.5
ip-10-50-42-1     Ready    <none>   12d   v1.32.5
ip-10-50-57-106   Ready    <none>   14m   v1.32.5

After a delay of around 10 minutes, the autoscaler removes unused nodes to reduce unused capacity and costs.

Once scale-down is complete, the number of nodes should return to the minimum value defined in the NodePool.

Example output:

NAME              STATUS   ROLES    AGE   VERSION
ip-10-50-34-159   Ready    <none>   12d   v1.32.5
ip-10-50-57-106   Ready    <none>   26m   v1.32.5

Expected Result

During this lab:

• The OKS cluster automatically added nodes when CPU load increased (scale-up).
• After about 10 minutes of inactivity, it removed excess nodes to return to the minimum configuration (minNodes: 2).