The Zen of Akka
summary
A practical checklist: prefer one actor is no actor and design parent‑child hierarchies with meaningful names so logs and supervision point to domain concepts; isolate blocking on a dedicated dispatcher — never block without an IO dispatcher — and use backoff supervisors to avoid restart storms; avoid Java serialization (ping‑pong goes <1 µs locally but ~25 µs with Java serialization); switch to Kryo or protobuf (Kryo drops the 25× penalty to about 5×) and consider Akka Streams as a typed, lower‑power alternative.
AI-written summary · about
chapters (13)
- 0:00Intro
- 2:15The Akka landscape
- 3:09One actor is no Actor
- 6:00Structure your Actors
- 7:22Name your Actors
- 11:44"Matrix of mutability (Pain)"
- 14:15Blocking needs careful management
- 22:34Avoid Java Serialization
- 34:32Backoff Supervision
- 38:27Design using State Machines
- 40:18Cluster Convergence and Joining
- 42:57Cluster Partitions and "Down"
- 46:06Happy hAkking, Community!
related talks
48mThe Zen of Akka
40mTop Mistakes When Writing Reactive Applications
51mUp up and Out: Scaling Software with Akka
32mAdopting Actors... an epic tale of loss and learning
37mAkka Present and Future
01h 32mFrom Spring + Java to Spring + Akka - A Journey of Discovery
38mBest Practices in Reactive Application with Java 1.8 and Play
44mDeath to Latency: Building Reactive, Cloud Native Apps with Akka NET