I See What You Mean
summary
Argues that a time-aware query language called Dedalus makes distributed programs mean what they produce by treating facts as timestamped ephemeral knowledge, yielding confluence (unique ultimate models) for monotonic fragments and decidable analysis limits otherwise. Describes implementations and toolchain—Bloom (language), Blazes (coordination insertion), and LDFI (lineage-driven fault injection)—and reports LDFI applied at Netflix for automated fault experiments, showing practical debugging, analysis and repair of consistency and fault-tolerance.
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:47Abstraction: the good
- 3:58Abstraction: the bad
- 5:30Abstraction: the ugly
- 13:52Back-of-envelope calculations
- 17:31Hellerstein's inequality
- 22:47Semantics: a richer example
- 24:00Lineage and derivation
- 29:26(stateless) protocols
- 29:56(stateful) protocols
- 30:28Distributed language requirements
- 33:45A relativist model of time
- 36:11Dedalus: Syntax
- 37:25Syntax Sugar
- 38:08Dedalus: stateful patterns
- 42:47Ultimate Models
- 45:27CALM Theorem
- 46:32Dedalus: design choices
- 47:49Logic programming tricks From model theory
- 50:08Dedalus answers
related talks
01h 01mLogic and Algebras for Distributed Computing: Applications and Open Questions
46mCoordination-Free Computations
13mAccelerating Program Analyses in Datalog by Merging Library Facts
17mBring Your Own Data Structures to Datalog
32mReactive Datalog for Datomic
17mFixpoints for the Masses: Programming with First-Class Datalog Constraints
41mProgramming Distributed Systems
17mInteractive Debugging of Datalog Programs