Skip to content
scrambled
Software Engineering Concept

Conflict-free Replicated Data Type

JourneySoftware engineering Statuslearning Tags
  • collaboration
On this page

A collection of data structures designed for replication across multiple computers in a distributed system.

Properties ​

  • Each replica can be updated independently and concurrently.

  • An algorithm automatically resolves inconsistencies.

  • All replicas are guaranteed to eventually converge (eventual consistency).

Common Types ​

  • Grow-only set: Only operation is "add". Adding twice is a no-op. Apply all updates in any order.

  • Last-writer-wins register: A single value. Updates carry a value, timestamp, and peer ID. Latest wins (peer ID breaks ties).

  • G-Counter: A counter that only increments. Each node maintains its own count. The total is the sum of all nodes.

Strengths ​

  • Mathematical guarantees of eventual consistency.

  • No central authority needed.

  • Well-studied foundation for distributed systems.

Weaknesses ​

  • Designed for decentralized systems, so they carry performance and memory overhead for tracking causality.

  • In a centralized system (client-server), this overhead is unnecessary.

Practical Insight ​

Even in a client-server setup, CRDTs are worth studying. They build intuition for creating correct systems. You can then relax requirements based on your needs (as Figma did).