Scaling EDA Workloads with Kubernetes, KEDA & Karpenter
summary
Claims that combining KEDA and Karpenter lets Kubernetes react to event volume (not CPU) to survive ticket-sale spikes: example numbers cited — 3.5 million pre‑sale registrations and ~3.5 billion system requests (≈4× prior peak). KEDA uses ScaledObjects to poll queues (SQS/Kinesis) and drive HPAs; Karpenter calls the EC2 Fleet API synchronously, picks right‑sized instances (spot or on‑demand) and cuts node‑pool complexity. Benchmarks show faster scaleups, e.g. to 10,000 pods, with cost and scheduling improvements.
AI-written summary · about
chapters (9)
- 0:00Intro
- 2:59Agenda
- 3:20The Taylor Swift problem
- 16:02Scaling with Kubernetes
- 23:46KEDA overview
- 32:52Application scaling with Karpenter
- 39:59Solving our Taylor Swift problem
- 42:42Key takeaways
- 45:16Outro
related talks
42mScaling Kubernetes-based Event-driven Workloads with Keda & Karpenter
29mCutting Costs and Boosting Efficiency with Karpenter and KEDA on Kubernetes
46mKarpenter * Keda: The Winning FinOps Duo
36mAutoscaling Isn't Set-and-Forget: Tuning KEDA & Karpenter for Real Workloads.
40mOptimize EKS Cost and Scaling with Karpenter
37mKEDA: Unlocking Advanced Event-Driven Scaling for Kubernetes
22mKarpenter Efficient scaling of Kubernetes clusters
35mKEDA - Real Time And Serverless Scaling In Kubernetes