Conflict-Free Replicated Datatypes (CRDT)
summary
Explains concrete CRDT designs with working examples: a grow-only counter across three devices converges to 3 despite different delivery orders, an add-only set that merges by union, and a composition that fixes remove-by-using two add-sets. Introduces using a per-client counter (Lamport timestamp) to build a last-write-wins set that allows remove‑then‑readd to succeed. Notes op‑based vs state‑based tradeoffs, duplicate messages, unbounded tombstones, and that GC and rich ordering are harder problems.
AI-written summary · about
chapters (12)
- 0:00Start
- 2:15Having a Single Source of Truth
- 4:18Merge Conflict
- 4:47Google Docs
- 5:12Operational Transformation
- 6:09Conflict Free Replicated Data Types
- 7:43Grow Only Counter
- 10:13An Add Only Set
- 12:20Strong Eventual Consistency
- 24:35Lamport Time Stamp
- 33:07Garbage Collection
- 35:26Questions
related talks
48mCRDTs: From sequential to concurrent executions
39mEventually Consistent Data Structures
40mPractical Demystification of CRDT
36mCRDT Architecture and Group Systems
29mAbout Combining Functional Reactive Programming and Replicated Data Types
34mUsing CRDTs Beyond Text Editors
43mAn Approach For The Internet Of Things
50mUsing CRDTs to Build a Highly Available Decentralized Service