custom-metric-autoscaling.md

Autoscaling

Back to overview

Custom Metric Autoscaling

General Introduction

For scaling pods depending on custom metrics you can also use the Horizontal Pod Autoscaler (HPA).

To make your custom metric available to the HPA, you need to make them available in the API Server. This can be done by

1. Collecting your custom metrics with Prometheus
2. Attaching your prometheus metrics to the Metrics Server with [Prometheus Adapter](https://github.com/DirectXMan12/ k8s-prometheus-adapter)
3. Configure your HPA to use the custom metric for scaling


For more information see: https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale-walkthrough/#autoscaling-on-multiple-metrics-and-custom-metrics

Example

1. Infrastructure Setup

Prometheus Operator

# TODO: Try out with helm chart
kubectl apply -f custom-metric-scaling

Prometheus Adapter

helm install -f metric-config.yaml --name prometheus-adapter-release stable/prometheus-adapter

Pod Metrics

helm install -f metric-config.yaml --name prometheus-adapter-release stable/prometheus-adapter

Example:

type: Pods
pods:
  metric:
    name: packets-per-second
  target:
    type: AverageValue
    averageValue: 1k

External Metrics

See: https://github.com/Azure/azure-k8s-metrics-adapter implemented with https://github.com/kubernetes-incubator/custom-metrics-apiserver

https://github.com/DirectXMan12/k8s-prometheus-adapter/blob/master/docs/walkthrough.md