Conflict Resolution and CRDTs — Distributed Systems for Backend Engineers

When multiple replicas accept writes (multi-region, offline-first apps, collaborative editing), concurrent updates conflict. Strategies: last-write-wins (simple, loses data), application merge (custom logic or user resolution), and CRDTs — data types that merge deterministically without coordination (G-Counter, PN-Counter, OR-Set, LWW-Register, sequence CRDTs for text). Libraries: Automerge, Yjs; databases: Riak, Redis Enterprise CRDBs, Azure Cosmos DB conflict policies.

Conflict strategies

Strategy	Data loss?	Complexity	Use for
Last-write-wins	Yes (concurrent writes)	Low	Caches, idempotent overwrites
Custom merge / user resolves	No (if done right)	Medium–high	Documents, records
CRDT counters/sets	No	Medium	Likes, carts, flags
Sequence CRDTs (Yjs/Automerge)	No	Library-provided	Collaborative text, offline apps

class GCounter:
    def __init__(self, node_id: str):
        self.node_id, self.counts = node_id, {}

    def increment(self, n: int = 1):
        self.counts[self.node_id] = self.counts.get(self.node_id, 0) + n

    def value(self) -> int:
        return sum(self.counts.values())

    def merge(self, other: "GCounter"):
        for node, c in other.counts.items():          # element-wise max: commutative, idempotent
            self.counts[node] = max(self.counts.get(node, 0), c)

import * as Y from "yjs";
const doc = new Y.Doc();
const todos = doc.getArray<string>("todos");
todos.push(["Buy milk"]);                         // offline edits on two devices…
const update = Y.encodeStateAsUpdate(doc);
Y.applyUpdate(otherDoc, update);                  // …merge deterministically when they sync

Common pitfalls

LWW on wall clocks for important data.
Ignoring conflicts in offline-first apps.
Rolling your own CRDTs for text editing.

Key points

Concurrent writes need a merge strategy
CRDTs merge without coordination
Use Yjs/Automerge for collaboration

Conflict strategies

Common pitfalls

Key points

Practice questions