K20s
View on GitHub

Smart Kubernetes Autoscaling.

K20s is a smart Kubernetes controller that provides sophisticated, policy-driven workload autoscaling and rightsizing.

Get Started View on GitHub

Beyond the HPA: Stable, Predictive Scaling

Unlike the native Horizontal Pod Autoscaler (HPA), K20s leverages long-term, historical data from Prometheus to make more intelligent and stable scaling decisions. We eliminate the jerky, oscillatory scaling based on momentary spikes.

It introduces the ResourceOptimizerProfile Custom Resource, allowing you to declaratively define powerful optimization policies for every application in your cluster.

Utilization Over Time
HPA (Jerky)
K20s (Stable)

Optimization, Elevated.

Policy-Driven Optimization

Choose between Scale, Resize, and Recommend policies to declaratively control exactly how your workloads are optimized.

Historical Data Analysis

Integrates directly with Prometheus to analyze resource utilization over extended periods, providing deep context for stable rightsizing.

Configurable "Goldilocks Zone"

Define your ideal CPU utilization ranges to keep applications performant while maximizing cost-effectiveness—not too hot, not too cold, just right.

Built-in Stability & Safety

Features like cooldown periods and explicit min/max boundaries prevent oscillatory scaling and ensure your cluster remains stable and predictable.

Clear Observability

A built-in status dashboard and custom Prometheus metrics give you a clear, transparent view of every decision the controller makes.

Native Kubernetes Experience

Configure scaling policies directly through Kubernetes Custom Resources—just use kubectl and YAML.

Installation

Prerequisites

  • A running Kubernetes cluster.
  • kubectl installed and configured.
  • Prometheus installed in the cluster and accessible via a service.

For detailed setup instructions, including using make install, see the Makefile Help Documentation.

1. Install the Controller

Run the following command to install the controller, CRD, and all necessary resources. Replace vX.Y.Z with the desired version from the GitHub Releases page. 

kubectl apply -f https://github.com/OpScaleHub/K20s/releases/download/vX.Y.Z/install.yaml

2. Verify the Installation

Check that the controller pod is running in the k20s-system namespace.

kubectl get pods -n k20s-system

How to Use K20s

K20s is configured using the ResourceOptimizerProfile Custom Resource. This allows you to define your scaling and optimization strategies declaratively, right alongside your other Kubernetes manifests.

Create a YAML file for your profile, specifying the target workloads via a label selector, your desired CPU utilization thresholds, and the optimization policy to apply.

Example Profile

This profile targets any deployment with the label app: sample-app. It will attempt to keep the average CPU utilization between 30% and 75% by scaling the number of replicas up or down. 

# sample-profile.yaml
apiVersion: optimizer.k20s.opscale.ir/v1
kind: ResourceOptimizerProfile
metadata:
  name: sample-app-optimizer
spec:
  selector:
    matchLabels:
      app: sample-app
  cpuThresholds:
    minPercent: 30
    maxPercent: 75
  optimizationPolicy: Scale

Apply the profile to your cluster:

kubectl apply -f sample-profile.yaml

For a complete reference of the ResourceOptimizerProfile fields, check the API Reference Documentation.

Ready to Optimize Your Fleet?

Get started by deploying the controller and creating your first optimization profile.

Deploy K20s Now