Consensus & Leader Election

Consensus is the problem of getting multiple servers to agree on a value. Leader election is a common way to achieve consensus — one server is the leader, others follow.

Visual: Raft Leader Election

sequenceDiagram
    participant L as 🏆 Leader (Server 1)
    participant F2 as 📋 Follower (Server 2)
    participant F3 as 📋 Follower (Server 3)
    participant F4 as 📋 Candidate (Server 4 — disconnected)

    Note over L,F4: Term 1 — Normal Operation
    L->>F2: Heartbeat (append entries)
    L->>F3: Heartbeat (append entries)
    F2-->>L: ✅ Ack
    F3-->>L: ✅ Ack

    Note over L,F4: Term 2 — Leader Fails ❌
    F2->>F2: No heartbeat from leader (timeout)
    F2->>F2: Start election
    F2->>F3: Request vote (term 2)
    F2->>F4: Request vote (term 2) — no response
    F3-->>F2: ✅ Vote granted
    Note over F2: ✅ Elected leader for term 2!

How Consensus Works (Raft Simplified)

Leader election: Servers vote for a leader. The candidate with majority votes wins.
Log replication: The leader accepts writes and replicates them to followers.
Commit: When a majority of followers acknowledge a write, it’s committed.
Safety: If the leader fails, a new leader is elected with the latest committed data.

Quorum

A quorum is the minimum number of nodes that must agree for a decision to be valid.

With 3 nodes: quorum = 2 (majority)
With 5 nodes: quorum = 3 (majority)

Why quorum matters:

Leader needs quorum votes to be elected
Writes need quorum acks to be committed
Reads need quorum responses to be consistent

Raft vs Paxos

Aspect	Raft	Paxos
Understandability	✅ Designed for understandability	❌ Famous for being hard to understand
Leader	Strong leader (elected)	Multiple proposers possible
Implementation	Used in etcd, Consul, MongoDB	Used in Google Chubby, Spanner
Log consistency	Sequential log replication	Multiple values can be proposed
Real-world usage	Very common	Common (often hidden behind abstractions)

Use Cases

System	Consensus Algorithm	Purpose
etcd / Consul	Raft	Service discovery, distributed config
Apache ZooKeeper	Zab (similar to Paxos)	Coordination, leader election
MongoDB (replica set)	Raft	Primary election, write consistency
Kafka	KRaft (Raft-based)	Controller election, metadata management

In Simple Words

Consensus = getting multiple servers to agree on who the leader is and what data is correct.
Raft is the most popular algorithm — it’s designed to be understandable.
A quorum (majority) must agree for things to happen. 3-node cluster needs 2/3 votes.