AntidoteDB: highly available, transactional database
summary
AntidoteDB targets global low-latency, highly available transactions by combining CRDTs (counters, sets, maps, registers), causal consistency and selective coordination. It ships as open source on GitHub, is implemented in Erlang, and has ~40 contributors; tooling includes the Reckless verifier, an automated tester, Docker images, a benchmark suite, an Antidote query language and an interactive playground. Concrete claims: prevents Cassandra-style delete/update anomalies (e.g. null authors in high-volume chat), supports AP operations unless an operation is CAP-sensitive (e.g. limited resources use reservations).
AI-written summary · about
chapters (18)
- 0:00Introduction
- 0:40Agenda
- 1:33Largescale distributed systems
- 2:35Twitter
- 3:22Pokemon Go
- 4:43Availability
- 8:59Data Model
- 13:12How did this happen
- 15:18Cloud vendor lockin
- 17:20What is AntidoteDB
- 18:30IDTs
- 20:50Chat
- 24:05Consistency model
- 25:38Invariants
- 29:43Access control
- 32:01Access control semantics
- 33:58Antidote
- 37:08Conclusion
related talks
44mAntidote - a Scalable and Consistent Transactional Data Store
50mCRDTs in Practice
16mScaleDB: A Scalable, Asynchronous In-Memory Database
35mData Types for Eventually Consistent Databases
46mCoordination-Free Computations
29mThe Design and Evolution of Event-Driven Databases
52mTransactions: Myths, Surprises and Opportunities
17mJust-Right Consistency