KEDA CPU Scaling

Welcome to this guide on implementing CPU-based autoscaling in Kubernetes using KEDA and the built-in Horizontal Pod Autoscaler (HPA). You’ll learn how KEDA monitors CPU metrics, sets thresholds, and dynamically adjusts your pod replicas to match workload demands.

How KEDA Leverages CPU Metrics

KEDA integrates with the Kubernetes HPA to fetch CPU metrics and make scaling decisions. The following diagram illustrates the flow:

1. Metrics Server exposes CPU usage for each pod.
2. HPA retrieves metrics and compares against defined thresholds.
3. KEDA’s ScaledObject configures HPA targets and min/max replica counts.
4. Pods scale out/in based on real-time CPU usage.

Prerequisites

Before you begin, ensure your cluster meets the following requirements:

• Metrics Server is deployed and operational.
• Pods specify CPU requests (and optionally limits) to enable accurate metrics.
• Kubernetes v1.27+ or compatible version for HPA and Custom Metrics API.

Defining CPU Requests in Your Deployment

Specify resource requests and limits in your Deployment manifest so the Metrics Server can report CPU usage correctly:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: cpu-app
  namespace: default
spec:
  replicas: 1
  selector:
    matchLabels:
      app: cpu-app
  template:
    metadata:
      labels:
        app: cpu-app
    spec:
      containers:
        - name: app
          image: images.my-company.example/app:v4
          resources:
            requests:
              cpu: "500m"
              memory: "128Mi"
            limits:
              cpu: "1000m"
              memory: "256Mi"

Configuring the CPU Trigger in KEDA

KEDA supports two CPU trigger modes:

Example trigger configuration:

triggers:
  - type: cpu
    metadata:
      type: Utilization      # 'Utilization' or 'AverageValue'
      value: "60"            # percentage or absolute value
      containerName: ""      # optional: target specific container

Creating the ScaledObject

A ScaledObject ties your Deployment to KEDA’s scaling logic. Below is a sample manifest that scales cpu-app between 1 and 5 replicas when CPU utilization exceeds 50%:

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: cpu-app-scaledobject
  namespace: default
spec:
  scaleTargetRef:
    name: cpu-app         # must match the Deployment name
  minReplicaCount: 1
  maxReplicaCount: 5
  triggers:
    - type: cpu
      metadata:
        type: Utilization
        value: "50"        # CPU utilization threshold

Summary of Key Resources

References

• KEDA Documentation
• Horizontal Pod Autoscaler
• Metrics Server Installation