Horizontal Scaling

Kubernetes Cluster: Horizontal Scaling

Horizontal scaling for a Kubernetes Cluster can happen at two different levels: the platform can add or remove Kubernetes nodes, while Kubernetes itself can increase or reduce the number of application pods according to workload demand.
Platform-Managed Scaling Adds or removes Kubernetes nodes. Scaling triggers can automate node changes based on resource utilization.
Kubernetes-Managed HPA Horizontal Pod Autoscaler scales deployments up or down based on observed CPU utilization.
On this page

Horizontal Scaling Overview

Horizontal scaling increases or decreases capacity by changing the number of running instances instead of only increasing CPU or memory on an existing instance.

In a Kubernetes environment, node scaling and pod scaling solve different problems. Node scaling controls the infrastructure capacity available to the cluster, while pod scaling controls how many application replicas run inside that capacity.

Platform-Managed Horizontal Scaling

The platform can horizontally scale Kubernetes infrastructure by adding or removing cluster nodes according to workload requirements.

  • Increase the number of Kubernetes nodes when the cluster requires more infrastructure capacity.
  • Reduce the number of nodes when the additional capacity is no longer needed.
  • Configure scaling triggers to automate this process based on resource utilization.
  • Use platform horizontal scaling together with Kubernetes scheduling so workloads can be placed on newly available nodes.
Node-level scaling: Platform horizontal scaling changes the amount of infrastructure available to Kubernetes. It does not directly increase the number of replicas in an application deployment.

Kubernetes Horizontal Pod Autoscaler

Kubernetes provides the Horizontal Pod Autoscaler (HPA) to scale deployments automatically according to observed workload metrics.

  • HPA monitors application resource utilization.
  • The source documentation specifically highlights observed CPU utilization as a scaling signal.
  • When demand rises, HPA can increase the number of pod replicas.
  • When demand falls, HPA can reduce the replica count.
HPA operates at the application workload level. It requires enough cluster capacity to schedule the additional pods it creates.

Using Both Scaling Methods Together

Platform-managed node scaling and Kubernetes HPA complement each other.

Application Demand HPA increases pod replicas when the application requires more processing capacity.
Cluster Capacity Platform horizontal scaling adds Kubernetes nodes when existing infrastructure becomes insufficient.
Scale Down As utilization drops, unnecessary pod replicas and cluster nodes can be reduced.
Availability Together, both methods help keep infrastructure available when needed and maintain healthy application pods.
Horizontal scaling summary: Use Kubernetes HPA to scale application replicas and platform horizontal scaling to provide or remove the worker-node capacity those replicas require.

The source documentation also references an extended WordPress use case that demonstrates Kubernetes Cluster scaling in a practical application environment.

What’s next?