Durable Functions: Semantics for Stateful Serverless
summary
Durable Functions extends FaaS with actors, workflows and entities and claims to give strong, fault free semantics while running on ephemeral workers and external storage. Paper models an ideal semantics and two implementations: a compute storage model that relies on atomic commit to simulate exactly once effects and a replay-based history that persists progress for arbitrary languages (C#, JavaScript, Python) without runtime checkpointing. Proofs show both implementations equal the idealized model and a SQL backend is used in practice.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:35What is Durable Functions?
- 0:53Serverless/Faas: What is it?
- 2:49The "Serverless Experience" is About Productivity
- 4:10Expanding the Serverless Experience
- 6:46Development Challenges
- 8:36Example: DF Thumbnail Service
- 9:23In the paper
- 10:190 Compute-Storage Implementation o
- 11:25Atomic Commit
- 12:31Replay-based implementation
- 13:27State = History of Events
- 14:21Summing up
related talks
28m8 Keys for Successful Serverless Architectures
28mStateful Applications in the Age of Durable Execution
58mAdvanced Serverless Workflows with Durable Functions
14mOvercoming Serverless challenges with Durable Functions
27mStateful Functions 3.0: Towards a Cross-Platform Stateful Serverless Stack
35mServerless is Interesting But FaaS is Not Enough
48mDelightful Durable Function Patterns
24mStateful Functions: Flexible like a Microservice, Consistent like a Monolith