How Zookeeper Elects Its Leader: A Human Election Analogy Explained

Basic Election Concepts

Candidate ability : a node with the most recent data (largest zxid) is considered strongest.

Switch to stronger : if a node learns of a candidate with a higher zxid, it changes its vote.

Ballot box : each node keeps an in‑memory map of its own vote and the votes received from peers.

Leader : the candidate that receives votes from a majority of the ensemble becomes the leader.

ZooKeeper Election Mechanism

ZooKeeper uses the same four concepts. A node's "ability" is measured by its transaction ID ( zxid). The larger the zxid, the newer the data and the higher the voting weight. Each vote also carries the node's unique identifier ( sid).

During an election a node initially votes for itself ( zxid + sid). When it receives a vote with a larger zxid, it updates its own vote to that candidate. Because votes are exchanged, all nodes eventually have identical ballot boxes. When a candidate gathers votes from more than half of the nodes, the election ends and the candidate assumes the LEADING state; the others transition to FOLLOWING.

When Does ZooKeeper Trigger an Election?

Server startup – the first time a server joins a cluster.

Leader failure – the current leader crashes or becomes unreachable.

Startup Election Example (5‑node cluster)

Server 1 starts : votes for itself ( LOOKING). No majority.

Server 2 starts : both vote for themselves; Server 1 sees a higher sid (2) and switches its vote to Server 2. Server 2 now has 2 votes.

Server 3 starts : all three vote for themselves; Servers 1 and 2 discover Server 3 has the highest sid and switch to it. Server 3 obtains 3 votes (majority) and becomes LEADER ( LEADING). Servers 1 and 2 become FOLLOWING.

Server 4 starts : votes for itself, sees the existing majority for Server 3, switches to Server 3 and becomes FOLLOWING.

Server 5 starts : same as Server 4, follows Server 3.

Result: Server 3 is the leader; the other four servers are followers.

Runtime Election Example (Leader Crash)

Assume the cluster originally had Server 3 as leader. The leader crashes and the remaining servers have the following zxid values:

Server 1: zxid=99
Server 2: zxid=102
Server 4: zxid=100
Server 5: zxid=101

All non‑observer servers change state to LOOKING and cast a vote for themselves.

Each server receives votes from peers; when a node sees a higher zxid, it updates its vote to that node.

Votes are tallied after each round. Once a node receives a majority of votes, it becomes the new leader.

The elected node switches to LEADING and announces the result; the others transition to FOLLOWING.

Key Concepts for the Election Algorithm

Server ID (sid) : unique numeric identifier; higher IDs have higher weight during startup comparison.

zxid : ZooKeeper Transaction ID; larger values indicate newer data and give higher voting weight during runtime elections.

Epoch : logical clock representing the election round; it increments each time a vote is cast.

Server states : LOOKING: candidate searching for a leader. FOLLOWING: follower synchronizing with the leader. OBSERVING: observer synchronizing with the leader but not voting. LEADING: the elected leader.

Summary

ZooKeeper performs leader elections both at cluster startup and when the current leader fails. Startup elections compare sid values; runtime elections compare zxid values. A node becomes leader when it secures a majority of votes. Understanding zxid, sid, epoch, and server states is essential for designing, operating, and troubleshooting ZooKeeper clusters.