How ZooKeeper Leader Election Ensures High Availability in Distributed Systems

Leader Election is used to ensure high availability of a critical service by selecting one node out of three to serve externally, and switching to a standby node if the leader fails.

ZooKeeper provides a primary use case for leader election. The implementation idea is as follows:

Implementation Idea

Assume three services: service_node1, service_node2, service_node3. Under ZooKeeper’s root a persistent node /ELECTION is created.

When each service starts, it creates an identical temporary sequential node under /ELECTION, e.g., /ELECTION/n_. ZooKeeper appends a sequence number, producing nodes like n_000000, n_000001, n_000002.

Each service reads all child nodes of /ELECTION and checks whether its own node has the smallest sequence number. The service with the smallest number becomes the leader.

Other services watch the node that precedes theirs. When the leader’s node is deleted (because the leader process crashes), the next service receives a notification, re‑reads the child list, and if its node is now the smallest, it assumes the leader role.

For example, if service_node1 (node n_000000) fails, service_node2 (watching n_000000) is notified, reads the remaining nodes [n_000001, n_000002], finds its node is now the smallest, and becomes the new leader. Similar logic applies when other nodes fail.

Summary of the Approach

All clients create a temporary sequential node under a fixed ZooKeeper path.

If a client’s node has the smallest sequence number, it becomes the leader.

Otherwise, the client watches the node that directly precedes its own.

When a watched node is deleted, the client re‑examines the child list to determine if it should become the leader.

Open‑Source Example

An example implementation is available on GitHub:

https://github.com/SainTechnologySolutions/allprogrammingtutorials/tree/master/apache-zookeeper/leader-election

The dist directory contains a compiled JAR that can be run directly.

Running the Example

Start three terminals and execute:

java -jar leader-election-0.0.1-SNAPSHOT.jar 1 192.168.31.204:2181

java -jar leader-election-0.0.1-SNAPSHOT.jar 2 192.168.31.204:2181

java -jar leader-election-0.0.1-SNAPSHOT.jar 3 192.168.31.204:2181

When terminal 1 is closed, terminal 2 receives the deletion event, becomes the smallest node, and assumes leadership.

Another reference implementation can be found at:

https://github.com/perezrathke/zookeeper-leader-election