From Request To Response And Everything In Between

on YouTube ↗36m

" summary

Follows a real Rails request from load balancer to response, explains how Puma processes (one request per thread) and multithreading interact with the Global VM Lock and why a blocking binding.pry can stall everything. Derives a capacity formula: processes ≈ arrival_rate × average_total_response_time × 4 (25% operating load), and demonstrates with 5 req/s → ~5 processes keeping queues from growing. Calls out per-thread connection pool sizing and database connection limits (example: 20 max connections) as the real scaling constraint.

AI-written summary · about

" chapters (12)

1. 0:00Introduction
2. 1:03About the talk
3. 3:19Requests in a Rails app
4. 7:06Single-threaded example
5. 9:15Scaling with processes
6. 12:30Optimal process count
7. 17:25Buffering traffic variability
8. 21:44Multi-threaded design
9. 25:46Global VM lock limits
10. 29:14Database connection pooling
11. 33:32Request to response
12. 35:41Closing notes

" related talks

• 32mLook Ma, No Sidekiq: A peek into async futureManu Janardhanan · RubyConf India 2023
• 27mInside Rails: The Lifecycle of a ResponseKrystan Huffmenne · RailsConf 2020 CE
• 41mIn The LoopLourens Naudé · LoneStarRuby Conf 2011
• 56mRails 4 and the Future of WebAaron Patterson · Aloha RubyConf 2012
• 41mRequest and ResponseAaron Patterson · Full Stack Fest 2015
• 38mA Practical Look at Performance TheoryKavya Joshi · Strange Loop 2018
• 26mRails Performance Monitoring 101: A Primer for DevelopersRishi Jain · RubyConf TH 2023
• 30mRails Performance monitoring: A primer for developersRishi Jain · RubyConf India 2023