jmarya/knowledge
jmarya/knowledge
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
knowledge/technology/tools/Raft.md

6.2 KiB
Raw Permalink Blame History

obj website wiki rev
concept https://raft.github.io https://en.wikipedia.org/wiki/Raft_(algorithm) 2024-02-07

Raft Consensus Algorithm

Raft is a consensus algorithm designed to manage replicated logs in distributed systems. Developed as an alternative to Paxos, Raft prioritizes understandability and simplicity, making it an attractive choice for building fault-tolerant distributed systems.

Concepts

Leader Election

Raft employs a leader election mechanism where nodes contend to become the leader.
Nodes transition through follower, candidate, and leader states based on timeouts and votes.

Log Replication

The leader manages log replication by sending heartbeat messages and append entries to followers.
Followers replicate the leader's log, ensuring consistency across the system.

Term and Commit Index

Raft introduces the concept of terms to track changes in leadership.
Each leader's term is associated with a commit index, indicating the highest log entry replicated to a majority of nodes.

How Raft Works

  • Follower State:
    Nodes start as followers, waiting for incoming leader messages.
    If a follower doesn't receive a message within a specified timeout, it transitions to the candidate state.

  • Candidate State:
    In the candidate state, a node requests votes from other nodes in the system.
    If a candidate receives votes from a majority, it transitions to the leader state.

  • Leader State:
    The leader manages log replication by sending heartbeat messages to maintain its leadership.
    It appends new log entries and sends append entries messages to followers for replication.

  • Log Replication:
    Followers replicate the leader's log, responding with acknowledgments to confirm successful replication.
    Log entries are committed when they have been replicated to a majority of nodes.

  • Handling Failures:
    Raft ensures fault tolerance by allowing nodes to detect leader failures and initiate new elections.
    Only the most up-to-date log entries are committed, preventing inconsistencies.

Use Cases

  • Distributed Databases: Raft is widely used in distributed databases to ensure data consistency and reliability.
  • Distributed Systems: Applications requiring consensus for critical decision-making leverage Raft for its simplicity and reliability.