Screaming Fast API Clients
summary
Uses async networking with Twisted to start all backend calls in parallel (example fan‑out of 10) instead of serial Flask-style calls, measures p50/p90/p99 histograms on a simulated laptop, adds short 0.1s timeout and a retry (simple second try, ~25% extra requests) and a final 0.4s give‑up that converts long tails into ~1% failures. Results show big p50/p90 wins and a flattened p99 when combined.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:56Acknowledgement of Country
- 1:58Microservice Architecture
- 4:20Lognormal Black Swans
- 6:46Averages Lie
- 8:04Your Backend is Slow
- 9:25Multiplicity Magnifies Outliers
- 10:55Measure
- 14:11Let's Write Some Code
- 18:49Timing Out and Retry
- 20:31Let's Simulate
- 25:39Summary
- 31:43Latency is the Site Killer
related talks
24mBuffalo: when Node and APIs become BFFs
23mµDjango, an Asynchronous Microservices Technique with Practical Examples
38mAdvanced API Design for Scalable Distributed Systems
01h 34mPython for Distributed Systems
30mOn scaling Machine Learning microservices
51mSimplifying Event-Driven Services with FastStream: A How-To Guide
17mHigh Performance Python Microservice Communication
55mGive it a REST - Tips for designing and consuming public API's