Scaling Applications in AKS

Azure Compute Solutions

Florin Angelescu

Azure Cloud Architect

Why scaling matters

Low Traffic

High Traffic

Applications rarely experience constant traffic:
- Some hours may be quiet
- Others bring spikes in usage
Without scaling:
- Workloads risk becoming overloaded
- Lead to poor performance or downtime

Why scaling matters

Kubernetes

Kubernetes provides mechanisms to adjust resources dynamically.
Applications can handle demand efficiently.
Scaling ensures:
- Users get consistent experiences
- Organizations optimize costs

Manual scaling

Manual Scaling

Increase or decrease the number of replicas in a deployment using kubectl scale.

Manual scaling

Manual Scaling

Scaling a web app from two replicas to four:

Ensures more pods are available to serve requests

Manual scaling

Manual Scaling

Useful for:

Planned product launches
Promotions
Stock sales

Horizontal Pod Autoscaler (HPA)

HPA

Horizontal Pod Autoscaler (HPA)

HPA

Monitors metrics (CPU and memory usage) then automatically adjusts the number of replicas.

Horizontal Pod Autoscaler (HPA)

HPA

Usage rises above a threshold:

More pods are created

Horizontal Pod Autoscaler (HPA)

HPA

Demand falls:

Pods are removed

Horizontal Pod Autoscaler (HPA)

HPA

Integrates with Azure Monitor.
Allows custom metrics (request latency, queue length).

Cluster Autoscaler

Autoscaler

Adds or removes nodes based on demand.
Pods cannot be scheduled due to insufficient resources:
- New nodes are provisioned automatically

Cluster Autoscaler

Autoscaler

Demand decreases:
- Unused nodes are removed to save costs
Infrastructure matches workload needs:
- Without over-provisioning

Best practices for scaling

Metrics

Define realistic resource requests and limits

Combine

Combine Horizontal Pod Autoscaler with Cluster Autoscaler

Test

Test scaling behavior under load

Cost

Monitor costs

Recap

Kubernetes

Scaling in AKS ensures applications adapt to demand automatically:
- Manual scaling
- Horizontal Pod Autoscaler
- Cluster Autoscaler
Balance performance and efficiency.

Let's practice!

Azure Compute Solutions