Concurrent Writes — Detect and Resolve Diagram

What Concurrent Means

Two writes are concurrent if neither happens before the other
Physical clock time does not matter: a slow network can make writes concurrent
A causal dependency, one write building on another, orders them

Version Vectors

The server keeps one version counter per replica per key
Reads return the vector and later writes must send it back
Values covered by the vector are overwritten; the rest stay as siblings

Merge Algorithms

CRDTs give every character or item a stable ID and merge by ID
Operational transformation rewrites indexes of concurrent edits
A cart merged by set union can resurrect deleted items: track deletions