Set up a vertical pod autoscaler in a GKE cluster

To figure out which kubernetes objects are how much resource, a vertical pod autoscaler might be useful. It has a feature to either automatically update values, or suggest values. But it cannot be used with a horizontal pod autoscaler to update values if the horizontal pod autoscaler doesn’t use a custom or external metrics. In that case, use a multidimensional pod autoscaling.

In this post, it’s to test a vertical pod autoscaler by following this document and this article.

Besides, there are Pre-GA feature in GCP as of Feb 2023, MultidimPodAutoscaler, written in this document. And this behaves

  • HorizontalPodAutoscaler for CPU
  • VerticalPodAutoscaler for memory

Getting Started

Before using a VPA nor MPA, enable a vertical pod autoscaler on your GKE cluster. Then create a VPA and MPA for your workload.

The example of a VPA for a recommendation

For a recommendation configuration, set .spec.updatePolicy Off on a VPA.

At first, define a resource like Deployment for a VPA:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: php-apache
  namespace: test-vpa
spec:
  selector:
    matchLabels:
      run: php-apache
  template:
    metadata:
      labels:
        run: php-apache
    spec:
      containers:
      - name: php-apache
        image: k8s.gcr.io/hpa-example
        ports:
        - containerPort: 80
        resources:
          limits:
            cpu: 500m
            memory: 100M
          requests:
            cpu: 100m
            memory: 10M

For the above VPA, define VPA.

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: php-apache-vpa
  namespace: test-vpa
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind:       Deployment
    name:       php-apache
  updatePolicy:
    updateMode: "Off" # Only recommendation
  resourcePolicy:
    containerPolicies:
      - containerName: php-apache
        minAllowed:
          cpu: "100m"
          memory: "50Mi"

The example output looks like next

[personal|test-vpa] > kubectl get vpa php-apache-vpa -o yaml | yq .status.recommendation
containerRecommendations:
  - containerName: php-apache
    lowerBound:
      cpu: 100m
      memory: 50Mi
    target:
      cpu: 100m
      memory: 50Mi
    uncappedTarget:
      cpu: 1m
      memory: "11534336"
    upperBound:
      cpu: 100m
      memory: "52428800"

Note that we shouldn’t use a HPA controlling by CPU.

The example of a MPA

The target of the deployment is the same as VPA. Note that there is no recommendation mode currently.

# https://cloud.google.com/kubernetes-engine/docs/how-to/multidimensional-pod-autoscaling#updatemode
apiVersion: autoscaling.gke.io/v1beta1
kind: MultidimPodAutoscaler
metadata:
  name: php-apache-autoscaler
  namespace: test-vpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: php-apache
  goals:
    metrics:
    - type: Resource
      resource:
      # Define the target CPU utilization request here
        name: cpu
        target:
          type: Utilization
          averageUtilization: 60
  constraints:
    global:
      minReplicas: 1
      maxReplicas: 5
    containerControlledResources: [ memory ]
    container:
    - name: '*'
    # Define boundaries for the memory request here
      requests:
        minAllowed:
          memory: 10Mi
        maxAllowed:
          memory: 100Mi
  policy:
    updateMode: "Auto"

Then you can see a HPA and VPA created by this MPA

[personal|test-vpa] > kubectl get hpa,vpa | grep php-apache-autoscaler
horizontalpodautoscaler.autoscaling/php-apache-autoscaler   Deployment/php-apache   1%/60%    1         5         1          21h
verticalpodautoscaler.autoscaling.k8s.io/php-apache-autoscaler   Auto          11534336   True       21h

2024

Back to Top ↑

2023

Follow Kubernetes the Hard way

4 minute read

This article was written by just following Kelsey Hightower’s Kubernetes Hardway document to understand Kubernetes internal architecture.

Gcp Billing Analyze

less than 1 minute read

There are a few documents to manage billing data in BigQuery Attribution of committed use discount fees and credits How to export to BigQuery Structur...

Prometheus Metrics Overview on Grafana

1 minute read

In this post, some variables defined in Grafana are used for Prometheus metrics, including $__rate_interval: This article describes the benefit of this va...

Use Google Secret Manager in a GKE cluster

3 minute read

There are an awesome article about the options to use the Google Secret Manager and their pros and cons. In this article, use Secrets Store CSI Driver by fol...

Working around MySQL lock metadata

2 minute read

There are multiple documents about innodb locks on MySQL 5.7: InnoDB locking Locks Set by Different SQL Statements in InnoDB Using InnoDB Transaction ...

Upgrade Windows 10 to Windows 11

3 minute read

I used to use Windows 11, but for some reasons, the OS stopped working and I needed to clean-install it from Windows 10 from windows recovery environment.

Back to Top ↑

2022

MySQL backup and restore

1 minute read

In this article, explain how to backup MySQL database using Percona Xtrabackup. There are two binaries, innobackupex and xtrabackup. innobackupex is a wrappe...

tmux

1 minute read

Basic configuration

Back to Top ↑

2021

MySQL Replication

1 minute read

This configuration is for the version 5.7 and it’s minimum configuration in the official document.

jq cheetsheet

less than 1 minute read

jq is used to parse JSON result, format and output on the cli.

Compare static site generator

less than 1 minute read

There are many web sites to compare static site generator, but they miss some explanations that require to me. For some people, these features are important ...

Back to Top ↑

2020

Getting Started with Kubernetes Deployment

less than 1 minute read

The deployment is many use cases and in this page, they’re not described. For the details for those use cases or the concept of deployment, see official page.

Overview about MySQL Lock

2 minute read

This document is written for MySQL 5.7, so these contents may be not correct for other versions.

MySQL Performance

2 minute read

This document is written for MySQL 5.7, so these contents may be not correct for other versions. In this page, performance_schema is mainly discussed.

Git hooks

less than 1 minute read

Configurations

gitHub pages

3 minute read

Getting Started See Official tutorial for detail steps.

Gitconfig

1 minute read

Configuration The detail for gitconfig is written in official page.

git cli

less than 1 minute read

Written in March 2020.

MySQL Tuner

less than 1 minute read

MySQL Tuner tool This is a tool to review a configuration for MySQL server.

kubectl cheetsheet

less than 1 minute read

Collect recent error logs If the logs are outputted by zap, error messages are aggregated by checking level = error. This log does not work very well if the ...

Introduction to GCP Cloud endpoints

less than 1 minute read

The Cloud endpoint is actually the NGINX proxy which offers the following features on GCP. Authentication and validation Logging and monitoring in GCP

HTTP/2 for Go

1 minute read

http package in golang supports HTTP/2 protocols. It’s automatically configured.

Back to Top ↑

2019

Terraform overview

1 minute read

Basic concepts There are some basic components for terraform.

Protocol Buffers for Go with Gadgets

less than 1 minute read

gogo/protobuf is the library to store some extensions from golang/protobuf in this repository. There are some useful packages that golang/protobuf does not p...

Introduction to GCP Cloud CDN

less than 1 minute read

Target upstream services Cloud CDN can have only GCP load balancer as the upstream services. And GCP load balancer can configure one of followings for backen...

Getting Started with Google closure library

less than 1 minute read

Some JavaScript library depends on Google Closure. If you need to understand the behavior of such a library, you have to know closure. The official document ...

Back to Top ↑