How three engineers at Antithesis shipped a production database in 14 months - safely!
summary
Antithesis builds and ships Pangolin, a tree‑based OLAP database written in Rust and deployed to S3 with V8 for query lambdas, in 14 months with three engineers. The team uses property-based testing (Hegel/QuickCheck-style swarm testing and shrinking) plus deterministic simulation testing to fuzz multiversal execution traces, catch a subtle small‑vector coalescing bug, and validate correctness of distributed behavior, compaction, and recovery before production rollout.
AI-written summary · about
chapters (10)
- 0:00Rolling your own database
- 2:49Correctness matters
- 7:30Design to production
- 11:00Property-based testing
- 13:07Impurum sandwich testing in Pangolin
- 14:34Primary index complexity
- 17:47Bug via property-based testing
- 23:44Simulation testing benefits
- 26:16Database testing takeaways
- 36:08Q&A
related talks
32mBuilding Reliable Infrastructure in Rust
06mLightning talk: Testing Distributed Databases
44mSchrodinger's Elephant
31mHelp! My Database Is The Bottleneck
56mLightning Talks
36mEmbrace the chaos: Composable data systems with fewer asterisks
30mSparkless Patterns: The New Data Stack and Modern Data Architecture
35mAn Ode to 17 Databases in 39 minutes
