Plan to scale or plan to fail: an evidence-based approach for improving systems performance
summary
Methodology replicates a Rails production stack on real infra with terraform, pushes synthetic traffic via a custom Go generator, and measures throughput and bottlenecks. Full test runs five days ingesting 190,000 chats/hr and 634,000 emails/hr; a 240k-record workflow averages about 2.5s per record but queue latency spikes to 6.5 hours as DB connections double from ~500 to 1,000 causing ActiveRecord connection failures until the database is scaled.
AI-written summary · about
chapters (10)
- 0:00Scaling challenges
- 3:13Load testing essentials
- 5:02Replicating production
- 8:18Terraform test infrastructure
- 9:41Simulating workloads
- 13:12Analyzing test results
- 15:27Full system performance case
- 20:09Sharing team insights
- 21:39Avoiding premature optimization
- 22:44Summary
related talks
31mA RAILS PERFORMANCE GUIDEBOOK
29mScaling Rails to 2M MySQL requests per second
33mHow much pressure can your app handle - Performance testing with Ruby
01h 03mRuby at Scale
44mMo' Jobs Mo' Problems: Lessons Learned Scaling to Millions of Jobs an Hour
34mHow to scale a Ruby webservice
39mWhere is my scalable API?
26mHow to Performance