Living in eventually consistent reality
summary
Describes how Conflict-free Replicated Data Types power multi-master systems like CosmosDB, DynamoDB, Riak, Cassandra and Redis and shows concrete implementations: a G-counter for global view counts, a PN-counter for likes/unlikes, a conflict‑free shopping cart built from counters and maps, and follower lists using observed-remove sets with timestamps or vector clocks. Explains dotted version vectors and delta-state replication to shrink metadata and network cost, cites Automerge, Akka Distributed Data and Redis Enterprise as usable libraries, and gives offline/geo-scale tradeoffs and merge rules.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 0:57Agenda
- 2:20Example
- 7:10Use cases
- 13:29Why and how
- 15:43Distributed systems
- 18:43Growing only counter
- 21:06Positive negative counter
- 22:24Voting system
- 24:12Distributed shopping cart
- 28:01List of followers
- 28:41Still face set
- 34:41Vector clocks
- 40:47Deltas
- 44:46Enforce consistency
- 47:06Dotted version vectors
- 48:45Disjoint dots
- 55:13Summary
related talks
49mDistributed Data: Conflict-free Data Structures in JS
29mHow to Read the Exact Time When You Have a Billion Clocks
35mData Types for Eventually Consistent Databases
43mDYNAMO: the paper that changed the database world
47mDatabase Resiliency
41mEverything you always wanted to know about highly available distributed databases
57mScaling the CAP Theorem for Modern Distributed Systems
41mTransactions: myths, surprises and opportunities