Battle of Black Friday: Proactively Autoscaling StockX
summary
StockX converts manual scaling into an automated workflow by combining EKS, Helm, HorizontalPodAutoscaler, Cluster Autoscaler and Datadog metrics and then adding a tiny event-driven warming service called barricade. Barricade accepts scheduled events, runs a Lambda/cronjob pipeline to pre-scale pods, then cools down after the window. The platform handles a ~533% spike in under 30 seconds, reaches 80,000 requests per minute with sub-300ms responses, and shows 93% fewer errors versus reactive-only scaling using Cluster Autoscaler.
AI-written summary · about
chapters (9)
- 0:00Intro
- 1:00StockX Overview
- 1:39The Problem
- 3:05The Problems
- 4:20Reactive Scaling
- 12:47Proactive Autoscaling
- 17:24Warming System
- 23:32Takeaways
- 30:29Outro
related talks
38mTempering our Production Practices
17mScaling Made Simple: Implementing Autoscaling for Optimal K8s Performance
36mAvoiding the Pitfalls of Autoscaling with Constant Work
29mKubernetes Autoscaling: Beyond One Size Fits All
21mHandling Traffic Spikes Through Automated Prioritized Load Shedding
42mMaking Resilient, Self-Adaptive, And Autonomous Kubernetes Clusters
29mProactive cluster autoscaling in Kubernetes
36mIntro: Autoscaling SIG