The efficient way of Autoscaling Your Workload in Kubernetes

on YouTube ↗40m

" summary

Baseline cost is about $9 for the demo app. Using k6 load tests, OpenCost and Istio/Envoy metrics, iterative tuning of requests and limits drops cost to about $2.4 but introduces CPU throttling and unstable response times. Adding autoscaling via HorizontalPodAutoscaler with a custom metric source and finally exposing observability through the Keptn Metric Server (and Keptn Metrics Operator) lets HPA use Istio/Envoy metrics and ultimately a CPU throttling metric to reach about $1.6 while keeping response time stable.

AI-written summary · about

" chapters (7)

1. 0:00Introduction and speaker intro
2. 3:11Story of optimizing a Kubernetes cluster
3. 7:47Understanding Kubernetes costs and workload resizing
4. 16:21Introducing Horizontal Pod Autoscaler (HPA)
5. 21:15Extending HPA with custom metrics
6. 26:09Key takeaways and recommendations
7. 28:16Q&A session begins

" related talks

• 29mKubernetes Autoscaling: Beyond One Size Fits AllJohn Platt · KubeDay Colombia 2024
• 34mIntroduction to AutoscalingGuy Templeton, Joe Burnett · KubeCon + CloudNativeCon North America 2020 Virtual
• 30mHorizontal Pod Autoscaler Reloaded - Scale on Custom MetricsMaciej Pytel, Solly Ross · KubeCon + CloudNativeCon Europe 2018
• 15mDemo: Resource(Cost) optimization on planet scale Kubernetes clusters.Rahul Mahale · Kubernetes Community Days Bengaluru 2023
• 51mHorizontal Autoscaling with KubernetesRob Richardson · KCD Los Angeles 2023
• 35mAutoscaling Kubernetes Deployments: A (Mostly) Practical GuideNatalie Serrino · KubeCon + CloudNativeCon Europe 2022
• 19mBuilding Cost Visibility For Your Kubernetes ClustersGarland Kan · DevOpsDays Baltimore 2019
• 44mFinOps at scale: Lessons from scaling Kubernetes workloadsRicardo Magalhães · JNation 2026