Reliable Actors with Retry Orchestration
summary
Proposes retry orchestration that guarantees failed actor invocations are retried, completed invocations never repeat, and strict happen‑before ordering across failures; formalizes semantics and implements it on Apache Kafka. Uses virtual actors plus tail calls and call‑stack awareness to minimize re‑execution. Evaluates with a 5,000‑line Java demo, injects thousands of node restarts, and reports ~22s median outage (runtime) with a workflow latency rising from ~100ms to ~24s during recovery.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:46Programming model
- 3:32RPC and tail calls
- 6:17Formal semantics
- 8:04Fault tolerance
- 12:58Implementation
- 15:46Summary
- 16:07Q&A
related talks
37mBeyond Fault Tolerance
48mReliable messaging in the world of actors
48mLost in transaction?
33mA Reliability Benchmark for Actor-Based Server Languages
27mFour Principles of Reactive Architectures
52mDistributed Computing Dealing with Time and Failure in the Wild
27mThe simplest way to build resilient applications
53mActors and Workflows: Different, Yet the Same