Polyglot Data
summary
Argues for an append-only event log as the durable source of truth so nothing is lost, with projections that replay from event zero to build read models; shows operational tradeoffs of polyglot persistence and why reads and writes diverge. Uses concrete examples: 100+ GB/day event volumes, Twitter outgrowing MySQL, and neo4j solved a recursive graph query in three days versus ~10 engineers for a year. Describes producer-driven queues failing for replay and recommends client-controlled subscriptions to avoid reconciliation nightmares.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:42What is event sourcing
- 2:16Structural models
- 7:49Projections
- 10:16Databases
- 11:08Object Databases
- 13:24Wrong Models
- 18:34CQRS
- 22:02Data Separation
- 24:52Multiple Models
- 30:42Event Stores
- 38:07Graph Databases
- 38:27Operational Complexity
- 39:45Summary
related talks
51mThings I Wish I knew Before I Started with Event Sourcing
31mSQL to NoSQL to NewSQL and the rise of polyglot persistence Paul Dix
32mPolyglot Persistence
48mEvent Sourcing - what could possibly go wrong?
45mData Modeling in an Event Centric World
55mRoad to polyglot persistence
29mThe Design and Evolution of Event-Driven Databases
38mPostmortem: Evolution of an Event recording solution