Skip to main content

🚀 What is Raft Consensus?

Raft is a consensus algorithm used in distributed systems to make a cluster of servers agree on a single sequence of operations (a log).

  • Imagine you have 5 servers.

  • They must all agree on the same log of events


    [SET x=1] → [SET y=3] → [DEL x]

  • Even if some servers crash or messages get lost, Raft ensures that:

    • All surviving servers eventually agree on the same log.
    • The log is applied in the same order everywhere.

This makes Raft ideal for databases, blockchain, and distributed systems.

⚖️ Why Consensus Is Hard

  • Network messages can be delayed, reordered, or lost.
  • Servers can crash and restart.
  • Without consensus, two servers might think they are the "truth," leading to data divergence.

Raft solves this by electing one leader and having all followers trust it.

🏗️ The Core Ideas of Raft

1. Leader Election​

  • At any time, one server is the leader, others are followers.
  • Followers start an election if they don’t hear from a leader within a timeout.
  • Votes are requested; the one with the majority becomes leader.
  • Guarantee: only one leader at a time.

2. Log Replication​

  • Clients send requests (commands) to the leader.
  • The leader appends the command to its log and replicates it to followers.
  • Once a majority acknowledges, the leader commits the entry.
  • Followers also commit that entry.

✅ This ensures that all nodes eventually have the same ordered log.

3. Safety Rules​

  • Election restriction: Candidates must have the latest log to win leadership.
  • Log matching: If two logs have an entry at the same index and term, they’re identical before that point.
  • Commitment: Only entries known by a majority are committed.

🔄 Raft States

Each server cycles between 3 roles:

  1. Follower → passive, just waits for leader’s messages.
  2. Candidate → requests votes during election.
  3. Leader → handles client requests, replicates logs.

📊 Timeline Example

Initial state: 5 servers, no leader
↓
Timeout (no heartbeat received)
↓
One follower becomes Candidate, requests votes
↓
Majority vote → Leader elected
↓
Client: "SET x=10"
↓
Leader appends log [SET x=10], replicates to followers
↓
Majority responds OK → Leader commits entry
↓
Followers commit entry → State is consistent

🛠️ Where Raft Is Used

  • etcd (Kubernetes metadata store)
  • Consul (HashiCorp service discovery)
  • TiKV (Rust-based distributed KV store, uses raft-rs)
  • Databend (uses openraft)

🎯 Key Takeaways

  • Raft = Consensus via Leadership.
  • Only leader talks to clients.
  • Majority (quorum) ensures fault tolerance.
  • Handles crashes, restarts, partitions.
  • Simpler than Paxos, but equally powerful.