Scaling the Deployment using kubectl

In this tutorial, you’ll learn how to scale an Azure Kubernetes Service (AKS) deployment and understand the per-node pod limit enforced by Azure CNI. We’ll cover:

1. Inspecting your current Deployment and Service
2. Scaling the Deployment to 5 replicas
3. Hitting the 30-pod per-node limit
4. Examining Namespaces and system pods

1. Inspect the current Deployment and Service

Before scaling, verify your application’s replica count and external endpoint.

# Check the Deployment replicas
kubectl get deployments
# Output example:
# NAME            READY   UP-TO-DATE   AVAILABLE   AGE
# Find the LoadBalancer’s public IP
kubectl get service
# NAME            TYPE           CLUSTER-IP      EXTERNAL-IP     PORT(S)         AGE
# kodekloudapp    LoadBalancer   10.0.199.121    20.247.251.108  80:30895/TCP    73m
# kubernetes      ClusterIP      10.0.0.1        <none>          443/TCP         124m

2. Scale to 5 replicas

Increase your Deployment to five pods:

kubectl scale deployment kodekloudapp --replicas=5
kubectl get deployment kodekloudapp
# NAME            READY   UP-TO-DATE   AVAILABLE   AGE
# kodekloudapp    5/5     5            5           76m

Validate load balancing

Open three incognito browser windows and navigate to your Service’s public IP. You should see traffic routed to different pods:

3. Scale to the pod limit (30 replicas)

When you created the AKS cluster, you configured Azure CNI with a maximum of 30 pods per node. Let’s push the Deployment to that limit:

kubectl scale deployment kodekloudapp --replicas=30
kubectl get deployment kodekloudapp
# NAME            READY   UP-TO-DATE   AVAILABLE   AGE
# kodekloudapp    16/30   30           16          78m

Only 16 pods are running; the rest remain Pending:

kubectl get pods
# NAME                           READY   STATUS    RESTARTS   AGE
# kodekloudapp-677fc758c5-5k92g  0/1     Pending   0          57s
# kodekloudapp-677fc758c5-bbp84  0/1     Pending   0          57s
# ...

Filter non-running pods

List pods that aren’t in the Running phase:

kubectl get pods --field-selector=status.phase!=Running

Inspect pod scheduling events

Describe one pending pod to see why it isn’t scheduled:

kubectl describe pod kodekloudapp-677fc758c5-5k92g

In the Events section you’ll find:

Warning  FailedScheduling   2m     default-scheduler   0/1 nodes are available: 1 Too many pods.
Normal   NotTriggerScaleUp  110s   cluster-autoscaler  max node group size reached

4. Namespaces and system pods

AKS uses several namespaces to isolate workloads. System pods in kube-system count toward your per-node limit.

kubectl get namespaces
# NAME              STATUS   AGE
# default           Active   129m
# kube-node-lease   Active   129m
# kube-public       Active   129m
# kube-system       Active   129m

Namespace overview

kubectl get pods --namespace kube-system
# NAME                                               READY   STATUS    RESTARTS   AGE
# ama-logs-rs-7f8bcb7c6f-5dlqq                       1/1     Running   0          131m
# coredns-59b6bf8b4f-lwt4c                           1/1     Running   0          131m
# metrics-server-7d74d8758-wfsdd                     2/2     Running   0          130m
# ...

Next steps

To work around the per-node pod limit, consider:

• Deploying multiple node pools with different --max-pods settings
• Switching to Kubenet or Azure CNI Overlay networks
• Splitting workloads across separate namespaces and node pools

Links and References

• Kubernetes Scaling Documentation
• AKS Networking Concepts
• Azure CNI IP Management