Kubernetes Autoscaling: Beyond One Size Fits All

on YouTube ↗29m

" summary

Kubernetes clusters average about 20% efficiency, so horizontal scaling on CPU often thrashes; switching to queue depth cut costs roughly 5x in the demo. HPA v2 supports custom and external metrics but needs a Prometheus adapter; KEDA provides a prewired alternative with 50+ scalers and built‑in auth and the ability to scale to zero. Cluster autoscaling moves from Cluster Autoscaler to faster Carpenter; vertical autoscaling (VPA) sees ~1% adoption despite solving requests/limits pain.

AI-written summary · about

" chapters (6)

1. 0:00Introduction to autoscaling
2. 1:03Autoscaling dimensions
3. 2:58Horizontal Pod Autoscaler
4. 7:52KEDA and custom metrics demo
5. 21:36Cluster autoscaling
6. 24:13Vertical scaling and conclusion

" related talks

• 35mSIG Autoscaling IntroMarcin Wielgus, Solly Ross · KubeCon + CloudNativeCon Europe 2018
• 51mHorizontal Autoscaling with KubernetesRob Richardson · KCD Los Angeles 2023
• 30mHorizontal Pod Autoscaler Reloaded - Scale on Custom MetricsMaciej Pytel, Solly Ross · KubeCon + CloudNativeCon Europe 2018
• 34mIntroduction to AutoscalingGuy Templeton, Joe Burnett · KubeCon + CloudNativeCon North America 2020 Virtual
• 17mScaling Made Simple: Implementing Autoscaling for Optimal K8s PerformanceEli Birger · KubeDay Israel 2023
• 35mAutoscaling Kubernetes Deployments: A (Mostly) Practical GuideNatalie Serrino · KubeCon + CloudNativeCon Europe 2022
• 35mIntro: SIG AutoscalingMarcin Wielgus, Maciej Pytel · KubeCon + CloudNativeCon China 2018
• 24mSIG-Autoscaling: Introduction + HPA/VPA UpdatesMarcin Wielgus, Joe Burnett · KubeCon + CloudNativeCon North America 2021