Part 2 of 5 · 1 min read
Leaderless replication
Any node can accept a write or serve a read; a client sends the operation to several nodes directly.
Think of a small team with no single designated note-taker. Instead, anyone who hears an announcement writes it down on their own copy of the shared notes. Everyone compares notes later to make sure they agree.
In plain words: leaderless replication has no designated leader or follower roles at all; any node can accept a write directly from a client, and a client typically sends a write to several nodes at once rather than to one single leader. This removes the single point that both single-leader and multi-leader replication still rely on for each individual write, at the cost of needing a different mechanism entirely to decide when enough nodes agree for an operation to be considered successful, which is exactly what a quorum, covered next, provides.
A worked example. A client writing an updated price sends that write directly to three different nodes at once, rather than to one designated leader, with no single node acting as the authoritative first recipient.
The key lines. No new SQL; which nodes a write or read is sent to, and how many responses are required, is a client or coordinator-level configuration in a leaderless system, not a single SQL statement's concern.
Leaderless replication
Leaderless replication has no designated leader or follower roles; a client sends a write or read directly to several nodes at once, with success determined by how many nodes respond, not by any single node's role.
Example: a team with no single note-taker, where anyone can write down an announcement on their own copy, mirrors leaderless replication's lack of a designated writer.