↓Skip to main content
  1. References/
  2. Architecture Design Basics/
  3. Pattern Taxonomy/
  4. Scaling & Performance/

Leaderless Replication

·· 176 words· 1 min
Reference Scaling Distributed Systems P1
Table of Contents

🟠 P1 — Dynamo-style replication; no single leader, quorum-based

Problem #

Leader-based replication has a single point of write failure (the leader). Leaderless eliminates this by allowing any replica to accept writes.

Mechanism #

  • Client writes to multiple replicas simultaneously (or via a coordinator)
  • Reads also query multiple replicas and resolve conflicts
  • Consistency via quorum: write to W nodes, read from R nodes, where W + R > N (total replicas)
N=3 replicas, W=2, R=2

Write: client sends to all 3, succeeds when 2 acknowledge
Read:  client reads from 2, takes the most recent value
W + R = 4 > N = 3 → guaranteed overlap → read sees latest write

Key Trade-offs #

  • Availability: No leader = no leader failover. Any node can serve reads and writes.
  • Conflict resolution: Concurrent writes to different replicas create conflicts. Need a resolution strategy (last-write-wins, vector clocks, CRDTs).
  • Complexity: Significantly more complex than leader-follower. Conflict resolution is the hard part.

Instinct #

Leaderless is for systems that prioritise availability over consistency (AP in CAP terms). DynamoDB, Cassandra, Riak use this model. For most applications, leader-follower is simpler and sufficient. Leaderless shines in multi-datacenter deployments where you can’t afford cross-region writes to a single leader. See also: Quorum Reads & Writes.

DDIA 2e Reference #

  • Chapter 5: Leaderless Replication, quorums, conflict resolution