How We Build Production-Grade HPA: From Effective Algorithm to Risk-Free Autoscaling

Discover how to build a production-grade Horizontal Pod Autoscaler (HPA) in this conference talk from Ant Group engineers. Learn about an innovative "traffic-driven" replica prediction algorithm designed for workloads with complex resource usage patterns, and explore a unique multi-stage gray scaling strategy that minimizes risks during autoscaling. Gain insights into a methodology that has successfully saved approximately 100,000 CPU cores annually with high stability. See a practical demonstration of implementing these techniques in Kubernetes environments using Kapacity, a newly open-sourced cloud native capacity solution, without relying on vendor-specific features.

How We Build Production-Grade HPA: From Effective Algorithm to Risk-Free... - Ziqiu Zhu & Yiru Guo