Stateful Functions: Flexible like a Microservice, Consistent like a Monolith
summary
Claims a runtime built on Apache Flink provides exactly-once state semantics and eliminates pay-for-wait time by inverting messaging so stateful storage routes messages to functions. Implements SDKs for Java, Python, Go, TypeScript; runs on AWS Lambda/Kubernetes with durability via synchronous snapshots to S3. Demo builds a fraud-alert pipeline (transaction coordinator, merchant-score caching, 30-day rolling counter, Python model) that runs end-to-end, scales without overflowing DB connections, and keeps messaging atomic and idempotent.
AI-written summary · about
chapters (8)
- 0:00Introduction to talk
- 0:25Evolution of application architectures
- 3:14Challenges with stateful microservices
- 6:20Introducing Stateful Functions
- 10:12Building with Stateful Functions
- 16:10Deployment flexibility
- 17:12Model serving use case
- 24:08Q&A and resources
related talks
55mStateful Functions: Polyglot Event-Driven Functions for Stateful Distributed Applications
27mStateful Functions 3.0: Towards a Cross-Platform Stateful Serverless Stack
42mInterview and AMA with the creators of Stateful Functions
34mHands-on Stateful Serverless Applications with Kubernetes and Stateful Functions
38mDeploying Stateful FaaS on Streaming Dataflows
43mStream Processing and Applications in the Modern Age
29mManaging State in Apache Flink
35mApache Flink 2.0: Unified Enterprise Data Processing System and Beyond