Autoscaling

Scale FreeRADIUS pods automatically based on CPU utilization.

---

Overview

The operator manages a Kubernetes HorizontalPodAutoscaler (HPA) for each RadiusCluster that has autoscaling enabled. This lets your RADIUS infrastructure scale with demand — handling peak authentication loads without over-provisioning during quiet periods.

Prerequisites

Your cluster must have the Metrics Server installed. Verify with:

kubectl top nodes

If this returns metrics, you're ready. If not, install the Metrics Server first.

Enable Autoscaling

apiVersion: radius.operator.io/v1alpha1
kind: RadiusCluster
metadata:
  name: production
  namespace: radius
spec:
  image: freeradius/freeradius-server:3.2.3
  replicas: 2
  resources:
    requests:
      cpu: 250m
      memory: 256Mi
    limits:
      cpu: "1"
      memory: 512Mi
  autoscaling:
    enabled: true
    minReplicas: 2
    maxReplicas: 10
    targetCPUUtilizationPercentage: 70

Note

When autoscaling is enabled, the spec.replicas field is ignored. The HPA controls the replica count between minReplicas and maxReplicas.

Configuration

Field	Default	Description
enabled	false	Set to true to create the HPA
minReplicas	1	Floor for pod count — never scale below this
maxReplicas	10	Ceiling for pod count — never scale above this
targetCPUUtilizationPercentage	80	Target average CPU usage across all pods

How It Works

1. When autoscaling.enabled is true, the operator creates an HPA targeting the FreeRADIUS Deployment
2. The HPA monitors average CPU utilization across pods
3. When utilization exceeds the target, the HPA increases replicas (up to maxReplicas)
4. When utilization drops, the HPA decreases replicas (down to minReplicas)
5. If you set autoscaling.enabled back to false, the operator deletes the HPA and spec.replicas takes effect again

Resource Requests Are Required

The HPA calculates CPU utilization as a percentage of the CPU request. If you don't set resources.requests.cpu, the HPA has no baseline and cannot make scaling decisions.

resources:
  requests:
    cpu: 250m       # Required for HPA to work
    memory: 256Mi

Sizing Guidelines

Deployment Size	minReplicas	maxReplicas	Target CPU
Development	1	3	80%
Small office (< 500 users)	2	5	75%
Campus (500–5000 users)	3	10	70%
Enterprise (5000+ users)	5	20	65%

These are starting points — adjust based on your observed authentication patterns and response time requirements.

Disable Autoscaling

Set enabled: false and specify your desired replica count:

spec:
  replicas: 3
  autoscaling:
    enabled: false

The operator deletes the HPA and sets the Deployment to 3 replicas.