..

Raft Consensus Alogorithm

Raft is a consensus algorithm designed to help a group of machines agree on the same sequence of actions (a log) even if some machines fail. It’s widely used in systems like distributed databases, key-value stores, and Raft storage.

Step-by-step mental model of how Raft works.

1 Core Idea (One-Line)

Raft keeps multiple servers in sync by electing one leader, and only the leader decides what gets written.

2 Roles in Raft

Every node is always in one of these roles:

Role What it does
Leader Handles all client requests and log replication
Follower Passive, just listens and applies updates
Candidate Tries to become leader during elections

At any time, there is at most one leader.

3 Leader Election (How a Leader Is Chosen)

Image

Image

Step-by-step

  1. All nodes start as followers

  2. If a follower doesn’t hear from a leader (heartbeat timeout):

    • It becomes a candidate

  3. Candidate:

    • Increments its term
    • Asks others for votes

  4. If it gets majority votes:

    • It becomes the leader

  5. If not:

    • Election restarts with randomized timeouts

Majority is mandatory → prevents split-brain.

4 Log Replication (How Data Is Safely Written)

Image

Image

Writing data works like this:

  1. Client sends request → Leader

  2. Leader:

    • Appends entry to its log
    • Sends AppendEntries RPC to followers

  3. Followers:

    • Write entry to their logs
    • Send ACK back

  4. Once majority ACKs:

    • Leader commits the entry
    • Followers apply it to their state machine

Only committed entries are considered durable.

5 Terms (Raft’s Logical Clock)

Raft uses terms to maintain order.

Term Meaning
Integer counter Increases monotonically
New election New term
Higher term wins Old leaders step down

If a node sees a higher term, it immediately becomes a follower.

6 Safety Guarantees (Why Raft Is Reliable)

Raft guarantees:

Leader Completeness

  • If an entry is committed, all future leaders will have it

Log Matching

  • If two logs have same index + term → logs before it are identical

No Split Brain

  • Only one leader per term (majority rule)

7 Failure Scenarios (What Happens If…)

Leader crashes

  • Followers stop receiving heartbeats
  • New election starts
  • New leader takes over

Follower crashes

  • Leader continues (as long as majority exists)
  • Follower catches up later

Network partition

  • Side with majority continues
  • Minority side becomes read-only

8 Why Raft Is Easier Than Paxos

Raft was designed to be understandable, unlike Paxos.

Paxos Raft
Hard to reason about Easy mental model
Abstract Practical roles
Leader implicit Leader explicit

9 One Diagram Mental Model

Client
  |
  v
Leader  --->  Follower
  |            Follower
  |            Follower
  |
(commit after majority)

TL;DR

  • Raft = Leader-based consensus
  • Leader is elected by majority
  • All writes go through leader
  • Logs are replicated and committed safely
  • Designed to be simple and predictable