Use actors, Luke
summary
Company runs payments in ~20 countries and builds a distributed payments coordinator using virtual actors (Orleans/.NET) that host a tiny in-memory virtual queue per payment, drive stepwise sagas for authorization/capture/refund, and persist only tails to durable storage. The migration cuts DB connector contention and smooths spike handling (reduced DB contention), but raises CPU from many timers and requires careful placement across two clouds. Implementation works in production with tradeoffs: simpler concurrency and better throughput for command handling, at cost of timer overhead and added operational complexity.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:31Actor model
- 8:11Multithreading problems
- 18:48Actor examples
- 23:42Payment system solution
- 30:12When not to use actors
- 43:00Q&A
related talks
47mUsing the Actor Model in Financial Transaction Systems
49mDDD, Event Sourcing and Actors
48mReliable messaging in the world of actors
45mSo you want to be an actor? Lessons from implementing the actor model
46mThe Actor Model and the Queue or “Batch is the New Black”
40mCAPtivating architecture: Navigating Distributed Systems and Microservices
22mCustom Pipeline Parallelism for Frontier-Scale Model Serving
19mScaling to 500k concurrent customers with orleans