Tempering our Production Practices
summary
Built autoscaling for Kubernetes that uses a Prometheus exporter per pod, a custom-metrics API server, and HPA driven by RPS for foreground and Sidekiq queue size for background processing. Deployed HAProxy buffer servers toggled by RDS→SNS events to isolate problematic tenants, plus BG restriction routing bad jobs to SQS and a low‑concurrency queue to slow processing. Handles sudden 2x traffic spikes and 15x job surges, cuts off-hours costs and keeps blast radius narrowly contained.
AI-written summary · about
chapters (4)
- 0:00Introduction to Freshservice
- 2:34Autoscaling framework
- 16:17Blast radius reduction
- 31:46Q&A and closing
related talks
14mA Tale of Autoscale: Increasing Resiliency while Reducing On-Call Toil
42mMaking Resilient, Self-Adaptive, And Autonomous Kubernetes Clusters
08mGetting Creative with HPA Scaling in Kubernetes
29mKubernetes Autoscaling: Beyond One Size Fits All
34mEfficient Resource Management
31mBattle of Black Friday: Proactively Autoscaling StockX
21mAutoscaling and Progressive Delivery - A Match Made in Heaven
28mApplication Autoscaling Made Easy With Kubernetes Event-Driven Autoscaling