And other friends

1. Consensus Algorithm that is easier to understand than Paxos
2. Leader Election:
   • Every node can become a leader
   • Every node has an election timeout (randomised value within a set interval)
   • Once the election timeout occurs, the node becomes a candidate
   • Candidates send vote requests to all known nodes
   • Once the candidate receives the majority of votes, it becomes the Leader
   • The Leader sends heartbeat ("Append Log") messages to followers
   • If the replica has not received a heartbeat from the leader in its election timeout, it becomes a candidate, etc...
3. Change Propagation:
   • Client sends a change to the leader
   • The change is appended to the leaders log
   • The change is sent to the followers in the next heartbeat
   • An entry is "committed" once the majority of followers acknowledge it
   • Once the entry is committed, the response is sent to the client
4. Resilience:
   • In case of network partitions, raft separates into sub-clusters with their own leaders
   • A sub-cluster with the smaller number of nodes will be unable to commit transactions
   • When the network partition is healed:
     • The leader with the lower follower count (smaller) will step down
     • The smaller and all its followers will rollback their logs to match the bigger one
     • This guarantees preservation of log consistency in the face of network failures

Good guys cite sources

1. Raft homepage: https://raft.github.io/
2. Rust bay Area Meetup talks on Raft: https://onlinexperiences.com/scripts/Server.nxp
3. Raft paper: https://raft.github.io/raft.pdf