Caching half a billion user transactions

on YouTube ↗43m

" summary

A homegrown Clojure cache backs a Memcached cluster to serve ~half a billion transactions by decoupling invalidation from eviction, using logical timestamps stored persistently (initially DynamoDB, later moved off after ~$10k/month costs), protocol-buf + gzip for serialization, and a three-tier generational/local/remote design. A delay-based local cache ensures each DAG of dependent computations runs once per generation, cutting repeated 20s recomputes to milliseconds; ~600 lines across three files implement it and it works in production across ~170 EC2 instances.

AI-written summary · about

" chapters (13)

1. 0:00Start
2. 0:33LEVEL APP + PLATFORM
3. 4:59EVICTION & INVALIDATION
4. 6:58NAMING
5. 10:13LEVEL CACHE-DEPENDENCIES
6. 13:51RACE CONDITIONS
7. 16:00MEM CACHE - PROBLEMS
8. 16:36CORE CACHE
9. 26:33A GENERTION CACHE, WITH LOCAL AND REMOTE PROVIDERS
10. 30:00WATCH YOUR DASHBOARDS
11. 30:43CLOJURE PERFORMANCE
12. 33:50HELLO, AGAIN
13. 36:25THE FUTURE

" related talks

• 31mSolid Cache: A disk backed Rails cacheDonal McBreen · Rails World 2023
• 33mWhat the Cache?!Simon Kröger · RailsConf 2014
• 47mCaching beyond simple put and getsJakub Marchwicki · GeeCON Prague 2022
• 23mScaling MemcacheHerman Lee, Rajesh Nishtala, Hans Fugal, Steven Grimm, Marc Kwiatkowski, Harry C. Li, Ryan McElroy, Mike Paleczny, Daniel Peek, Paul Saab, David Stafford, Tony Tung, Venkateshwaran Venkataramani · NSDI 2013
• 24mCaching Strategy for High Volume Personalized Content RequestsEmily Voytek, Jamie F · GraphQL Summit 2023
• 38mSecuring Distributed Cache: Achieving Secure-by-Default with Key Challenges &...Akashdeep Goel, Sriram Rangarajan, Samuel Fu · SREcon25 Americas
• 30mCaching patterns for lazy devs for lazy loadingLuigi Fugaro · Voxxed Days Ticino 2023
• 38mLazy Defenses: Using Scaled TTLs to Keep Your Cache CorrectBonnie Eisenman · Strange Loop 2017