Kubernetes Autoscaling : Build Efficient, Cost-Optimized Clusters with KEDA and Karpenter.

Format:

Book

Author/Creator:

Melendez, Christian.

Language:

English

Subjects (All):

Kubernetes.

Cloud computing.

Virtual computer systems.

Physical Description:

1 online resource (421 pages)

Edition:

1st ed.

Place of Publication:

Birmingham : Packt Publishing, Limited, 2025.

Summary:

Learn to scale workloads efficiently, reduce cloud costs, and optimize performance with real-world strategies for event-driven and infrastructure-level scaling Key Features Autoscale Kubernetes workloads and infrastructure using KEDA and Karpenter Improve performance, reduce cloud costs, and eliminate resource waste with smarter scaling Work with.

Contents:

Intro

FM

Copyright

Forewords

Contributors

Table of Contents

Preface

Free Benefits with Your Book

Part 1: Getting started with Kubernetes Autoscaling

Chapter 1: Introduction to Kubernetes Autoscaling

Technical requirements

Scalability foundations

A bit of history

Horizontal and vertical scaling

Vertical scaling

Horizontal scaling

Kubernetes architecture

Efficient Kubernetes data planes

What do I mean by efficiency?

Challenges and considerations

Kubernetes autoscaling categories

Application workloads

Data plane nodes

Hands-on lab: Creating a Kubernetes cluster

Local Kubernetes cluster with Kind

Installing Kind

Creating a Kind cluster

Cloud Kubernetes cluster in AWS

Creating an Amazon EKS cluster

Removing the Amazon EKS cluster

Summary

Get This Book's PDF Version and Exclusive Extras

Chapter 2: Workload Autoscaling Overview

Challenges of autoscaling workloads

How does the Kubernetes scheduler work?

Configuring requests

Configuring limits

What if you don't specify resource requests or limits?

Pod configuration example

What if the pod exceeds the resource limits?

Recommendations for configuring resources and limits

Workload rightsizing

Monitoring

Prometheus and Grafana

Hands-on lab: Setting up Prometheus and Grafana

Hands-on lab: Determining the right size of an application

Establishing defaults

Establishing default requests and limits

Hands-on lab: Setting up default requests for CPU and memory

Workload autoscalers

HorizontalpodAutoscaler (HPA)

VerticalpodAutoscaler (VPA)

Kubernetes Event-driven Autoscaling (KEDA)

Chapter 3: Workload Autoscaling with HPA and VPA.

Technical requirements

The Kubernetes Metrics Server

The Metrics Server: The what and the why

Hands-on lab: Setting up the Metrics Server

Hands-on lab: Using the Metrics Server

HPA basics

How HPA scales resources

Defining HPA scaling policies

Hands-on lab: Scaling using basic metrics with HPA

Deploy the sample application

Create the HPA autoscaling policy

Run load tests

Watch autoscaling working

HPA and custom metrics

How does HPA work with custom metrics?

Hands-on lab: Scaling using custom metrics with HPA

Deploy the Prometheus Adapter

Deploy the Service monitor

Run load tests and see HPA in action

VPA basics

How VPA scales resources

Defining VPA scaling policies

Hands-on lab: Automatic vertical scaling with VPA

Deploy VPA components

Run load tests and see VPA in action

How to work with HPA and VPA together

Part 2: Workload Autoscaling and KEDA

Chapter 4: Kubernetes Event-Driven Autoscaling - Part 1

KEDA: What it is and why you need it

KEDA's architecture

KEDA Operator

KEDA Metrics Server

Admission Webhooks

What Kubernetes objects can KEDA scale?

KEDA scalers

KEDA CRDs

ScaledObject

ScaledJob

TriggerAuthentication and ClusterTriggerAuthentication

Hands-on lab: Installing KEDA

Scaling Deployments

Hands-on lab: Scaling using latency

Controlling autoscaling speed

Scaling to zero

Hands-on lab: Scaling from/to zero

Deploying a RabbitMQ queue

Deploying the sample application

Deploying a message producer

Watching KEDA in action

Cleanup

Scaling Jobs

Hands-on lab: Scaling jobs

Exploring the ScaledJob rule.

Deploying the sample application

Chapter 5: Kubernetes Event-Driven Autoscaling - Part 2

Autoscaling in KEDA continued

Scaling based on schedule

Hands-on lab: Scaling to zero during non-working hours

KEDA's HTTP add-on

What if a KEDA scaler is not available?

Caching metrics

Pausing autoscaling

Fallback scaling actions

Advanced autoscaling features

Scaling modifiers

Important considerations with scalingModifiers

Hands-on lab: Pausing autoscaling when resources are constrained

Extending KEDA with external scalers

KEDA with cloud providers

KEDA on Amazon EKS

KEDA on Azure Kubernetes Service

KEDA on Google Kubernetes Engine

Chapter 6: Workload Autoscaling Operations

Workload autoscaling operations

Troubleshooting workload autoscaling

Troubleshooting HPA

Reviewing HPA conditions and events

Checking Metrics Server logs

Troubleshooting VPA

Reviewing the events and status of the VPA objects

Reading the logs from VPA components

Verifying why VPA can't apply recommendations

Troubleshooting KEDA

Reviewing events from ScaledObject or ScaledJob objects

Reading the logs from KEDA's components

Monitoring KEDA

Hands-on lab: Deploying KEDA's Grafana dashboard

Upgrading KEDA

Best practices for workload efficiency

Part 3: Node Autoscaling and Karpenter

Chapter 7: Data Plane Autoscaling Overview

What is data plane autoscaling?

Data plane autoscalers

CAS

Karpenter

CAS on AWS

Hands-on lab: CAS on AWS.

Step 1: Deploy CAS

Step 2: Remove VPA rules that could resize the sample application

Step 3: Deploy a sample application to see how CAS adds new nodes

Step 4: Remove the sample application to see how CAS removes unnecessary nodes

Step 5: Uninstall CAS

CAS best practices

Use homogeneous instance types

Limit the number of node groups

Configure a sane scaling speed

Use a priority expander with multiple nodes

Use a single Availability Zone for persistent volumes

Relevant autoscalers

Descheduler

CPA

CPVA

Chapter 8: Node Autoscaling with Karpenter - Part 1

What is Karpenter?

How does Karpenter work in AWS?

History of Karpenter

Karpenter resources

NodeClass

NodePool

NodePool requirements

General recommendations

NodeClaim

Launching nodes

1. Scheduling

2. Batching

3. Bin-packing

4. Launching the request

Deploying Karpenter

Hands-on lab: Autoscaling with Karpenter

Creating a fresh EKS cluster

Deploying a sample application

Deploying a default NodePool and EC2NodeClass

Removing the sample application

Workload scheduling constraints

Node selector

Node affinity

Pod affinity and pod anti-affinity

Tolerations and taints

Topology spread constraints

Scheduling nuances and minDomains for topology spread

PersistentVolume topology and zonal placement

Requesting specific hardware

minValues

NodePool weights

Chapter 9: Node Autoscaling with Karpenter - Part 2

Removing nodes

Control flow architecture

Disruption controller

Termination controller

Disruption

Consolidation

What happens when Karpenter skips consolidation?.

Hands-on lab: Consolidating nodes after scaling events

Drift

What causes drift?

How drift is detected

Managing drift carefully

Hands-on lab: Drifted nodes after updating AMI

Expiring nodes

Graceful node termination

NodePool disruption budgets

Disruptions by schedule

Disruption by reasons

Multiple disruptions budgets

Hands-on lab: Disruption budgets

Working with PDBs

Karpenter best practices

NodePools

Workloads

Chapter 10: Karpenter Management Operations

Advanced operations and troubleshooting

Troubleshooting Karpenter

Getting Karpenter events

Events reference table

Getting Karpenter logs

Exploring status conditions

Common operational challenges

Cost spikes

Consolidation churn

Having extra capacity due to scheduling preferences

DaemonSet taints cause the launch of extra nodes

Node provisioning challenges

Subnet and security group selection

Invalid launch templates or block device mappings

IP address exhaustion in subnets

Insufficient Instance Capacity error (ICE)

Quota limits

Consolidation failures

Pods remain pending

Node Auto Repair for unhealthy node handling

Preventing cascading failures

Enabling NAR

Hands-on lab: NAR in action

1. Enable the NodeRepair feature

2. Check that the Node Monitoring Agent is running

3. Inject a failure into the node

4. Wait for Karpenter to replace the node

Upgrading Karpenter

1. Pre-upgrade checks

2. Stage the upgrade

3. Upgrade process

4. Post-upgrade verification

Observability and metrics

Cluster resource efficiency metrics

Provisioning reliability and performance metrics

Hands-on lab: Deploy the Karpenter Grafana dashboard

Summary.

Get This Book's PDF Version and Exclusive Extras.

Notes:

Description based on publisher supplied metadata and other sources.

ISBN:

1-83664-382-9

9781836643821

OCLC:

1550453854

Publisher Number:

CIPO000303198

CIPO000303199