Consensus: Raft / Paxos — Agreement Under Failure

Algorithm for a cluster of nodes to agree on a single value (or log entry) even when some nodes fail.

When to use

• Leader election in distributed systems (etcd, ZooKeeper)
• Replicated log for consistent state across nodes (Kafka KRaft, CockroachDB)

Tradeoffs

• Requires N/2+1 quorum — minority partition becomes unavailable
• Latency penalty for every write (must wait for quorum acknowledgment)

Go

type NodeState int

const (
    Follower  NodeState = iota // default, receives log entries from leader
    Candidate                  // seeking votes after election timeout
    Leader                     // sends heartbeats, accepts writes
)

type RaftNode struct {
    state       NodeState
    currentTerm int
    votedFor    *string
    log         []LogEntry
}

// Transition: Follower → Candidate on election timeout
func (n *RaftNode) startElection() {
    n.state = Candidate
    n.currentTerm++
    n.votedFor = &n.id
    // broadcast RequestVote RPCs to all peers
}

Python

from enum import Enum, auto

class NodeState(Enum):
    FOLLOWER  = auto()  # default, receives log entries from leader
    CANDIDATE = auto()  # seeking votes after election timeout
    LEADER    = auto()  # sends heartbeats, accepts writes

class RaftNode:
    def __init__(self, node_id: str):
        self.state = NodeState.FOLLOWER
        self.current_term = 0
        self.voted_for: str | None = None
        self.log: list = []
        self.id = node_id

    def start_election(self) -> None:
        # Transition: FOLLOWER → CANDIDATE on election timeout
        self.state = NodeState.CANDIDATE
        self.current_term += 1
        self.voted_for = self.id
        # broadcast RequestVote RPCs to all peers

Gotcha: Raft is Paxos made understandable. etcd uses Raft. Kafka replaced ZooKeeper with KRaft (also Raft). If you're building on these, you're already using consensus.