Kubernetes Autoscaling - HPA vs VPA vs KEDA

Author

Govardhana M K
January 07, 2025

TechOps Examples

Hey — It's Govardhana MK 👋

Along with a use case deep dive, we identify the remote job opportunities, top news, tools, and articles in the TechOps industry.

👋 Before we begin... a big thank you to today's sponsor PERFECT SCALE

Join us for a hands-on webinar featuring Zbyněk Roubalík, CTO of Kedify and KEDA Project Maintainer, as we explore the advantages (as well as challenges) of event-driven autoscaling for Kubernetes.

You will learn:

→ An introduction to KEDA and its role in Kubernetes event-driven autoscaling.

→ Key scaling triggers and how they optimize resource allocation.

→ Real-world advantages and solutions to common challenges in event-driven scaling.

→ Insights into the future of KEDA and event-driven scaling.

→ A live demonstration of KEDA in action.

IN TODAY'S EDITION

🧠 Use Case

  • Kubernetes Autoscaling - HPA vs VPA vs KEDA

🚀 Top News

👀 Remote Jobs

  • Stakefish is hiring a DevOps Engineer

    Remote Location: USA, UK, Singapore, Taiwan, UAE

📚️ Resources

📢 Reddit Threads

👋 They say 2025 is the year of AI agents, and to prep you for that, I can’t resist recommending.

Your daily AI dose

Mindstream is your one-stop shop for all things AI.

How good are we? Well, we become only the second ever newsletter (after the Hustle) to be acquired by HubSpot. Our small team of writers works hard to put out the most enjoyable and informative newsletter on AI around.

It’s completely free, and you’ll get a bunch of free AI resources when you subscribe.

🛠️ TOOL OF THE DAY

killercoda -  A hands on learning platform to master Kubernetes, Linux, and DevOps tools.

  • Perfect for certifications like CKA, CKAD, CKS, LFCS, and ICA.

  • Right in your browser. No setup needed, just start learning.

  • Gain hands on skills for real world DevOps challenges.

🧠 USE CASE

Kubernetes Autoscaling - HPA vs VPA vs KEDA

Which type of autoscaling should I use for my workload?

Can I scale based on message queues or external triggers?

What happens if I combine multiple autoscaling strategies?

No seasoned Kubernetes professional can skip these questions in their career.

Since there is already tons of information available on how to set up autoscaling, let’s not go there.

For someone yet to know, you can refer to the Autoscaling Workloads guide here.

And I have tried to simplify the behind the scenes as a self-explanatory illustration below:

Despite being powerful, Kubernetes autoscaling has its pitfalls, understanding how and when to use each autoscaler is critical.

1. Horizontal Pod Autoscaler (HPA)

HPA is the default choice for scaling Kubernetes workloads horizontally by adding or removing pods based on resource utilization or custom metrics.

How HPA Works:

  • Continuously monitors metrics like CPU, memory, or custom metrics (e.g., request rates).

  • Adjusts the number of pod replicas in a deployment based on predefined thresholds.

  • Formula: desiredReplicas = ceil(currentReplicas * (currentMetricValue / targetValue)).

2. Vertical Pod Autoscaler (VPA)

VPA optimizes pod resource requests (CPU and memory) by learning from historical and real-time usage patterns.

How VPA Works:

  • Analyzes resource utilization and suggests or directly applies changes to resource requests/limits.

  • Modes:

    • Auto: Automatically applies resource recommendations.

    • Initial: Sets resource requests at pod creation only.

    • Off: Provides recommendations without applying changes.

3. Kubernetes Event-Driven Autoscaling (KEDA)

KEDA extends autoscaling to handle event-driven workloads by scaling deployments based on external triggers like message queues, HTTP requests, or custom metrics.

How KEDA Works:

  • Integrates with external systems (e.g., Kafka, RabbitMQ) to fetch metrics and decide scaling.

  • Uses ScaledObjects to define scaling rules and event sources.

To be honest, we can’t sum up the depth of KEDA in a short phrase, so we are bringing a hands-on webinar with live exploration and demonstration from the KEDA Project Maintainer - Zbyněk Roubalík himself.

For many workloads, a hybrid approach works best:

  1. HPA + VPA: Use HPA to scale pods based on CPU/memory usage while VPA adjusts pod resource requests for efficient utilization.

  2. HPA + KEDA: Use HPA for resource-based scaling and KEDA for event-driven scaling.

  3. HPA + VPA + KEDA: Combine all three for workloads that are both resource-intensive and event-driven, ensuring cost efficiency and performance.

You may even like:

Looking to promote your company, product, service, or event to 30,000+ TechOps Professionals? Let's work together.