4.2 Leaderless Replication

1. What Is Leaderless Replication?

There is no single write leader for a key. A client or coordinator sends writes to multiple replicas and reads from multiple replicas, then reconciles the returned versions.

2. Quorum-Style Reads and Writes

Let:

• N = number of replicas;
• W = replicas that must acknowledge a write;
• R = replicas consulted for a read.

The familiar condition R + W > N creates overlap between a successful read set and write set under simplified assumptions. It does not by itself guarantee linearizability: sloppy quorums, concurrent writes, clock/version rules, network timing, and failure recovery still matter.

3. Concurrent Writes

Leaderless systems need version metadata to distinguish newer, older, and concurrent values. Depending on the system this may use version vectors, logical clocks, timestamps, or application-specific merge semantics.

4. Repair

4.1 Read Repair

When a read discovers an older replica, the system can write the newer or reconciled value back to that replica.

4.2 Anti-Entropy

Background processes compare replicas and repair divergent data even when a key is not actively read.

Leaderless designs can remain available under some failures, but conflict semantics are part of the data model rather than an implementation detail.