Skip to content
Alternate pod autoscaler implementation
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
controller
deploy
hack
pkg
.gitignore
.travis.yml
Dockerfile
LICENSE
Makefile
README.md
go.mod
go.sum
main.go

README.md

Basic Kubernetes Pod Autoscaler

The default Horizontal Pod Autoscaler has several shortcomings which can be listed as follows:

  1. The scaling is not restricted in terms of how pods are started and stopped at the same time.
  2. The scaling action is not based on historical data but on current usage.
  3. The thresholds for scaling up and scaling down are the same.

Usage

Create a scaler object in the following format:

apiVersion: arjunnaik.in/v1alpha1
kind: Scaler
metadata:
  name: example-scaler
  namespace: default
spec:
  evaluations: 2    // Number of evaluations in before scaling happens
  minReplicas: 1    // Minimum number of replicas
  maxReplicas: 10   // Maximum number of replicas
  scaleUp: 50       // Scale Up threshold in utilization percentage
  scaleDown: 20     // Scale Down threshold in utilization percentage
  scaleUpSize: 2    // Number of pods to scale up
  scaleDownSize: 1  // Number of pods to scale down
  target:
    kind: Deployment
    name: nginx
    apiVersion: apps/v1

In the above example the target field contains the scaling target. In this case the target is a Deployment with the name nginx. Evaluations indicates the number of minutes (cycles) before scaling happens. In this example, if the CPU utilization of a pod is more than 50% for more than 2 minutes then the deployment is scaled up. The scaleUpSize and scaleDownSize indicates the number of pods to be increased on successful scale up or scale down evaluations.

Dependencies

This setup expects Prometheus to be running in the cluster and configured to scrape pod resource metrics. The address for Prometheus can be passed through -prometheus-url flag.

You can’t perform that action at this time.