High Performance API on Kubernetes
summary
Naive REST API in Go reads a single Redis and collapses past about 8,000 RPS; bottleneck is the single DB node. Using Kubernetes pods with a local Redis sidecar, asynchronous updates, and autoscaled replicas pushes stable traffic into the 10k–12k RPS range in demos and supports near-linear scaling in theory. Demos run on Google Cloud, containers built with Docker, deployed via kubectl, and stress-tested with Vegeta; live demo is flaky but the architectural fix proves effective.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:05Kubernetes scaling outline
- 5:00Kubernetes core concepts
- 12:05Naive API setup
- 23:37First scaling failure
- 31:54Redesign with sidecars
- 43:03Testing scalable architecture
- 48:38Scaling lessons
- 50:03Q&A
related talks
20mUsing Kubernetes API Effectively with Golang
27mRails Orbiting the cloud: Mastering the scalability with Kubernetes
48mScale and Performance Testing of Kubernetes: Answers for Specific Applications
20mOperating High Traffic Websites on Kubernetes
34mPanel: Improving and Managing Kubernetes at Scale
47mKEDA, HPA and VPA? An Introduction to Scaling
32mEnhancing the Performance Testing Process for gRPC Model Inferencing at Scale
33mKubernetes API Server Performance Clinic: Auditing, and Priority...