Mastering Apache Zookeeper: Core Concepts and Real-World Big Data Applications

1. Overview

Apache Zookeeper is an open‑source server that provides highly reliable distributed coordination. It acts as a service for clients such as Hadoop, HBase, and Hive, enabling cluster management, master election, publish/subscribe, data storage, and distributed locks.

2. Basic Concepts

2.1 Data Structure

Zookeeper stores data in a hierarchical tree of nodes called znodes, similar to a file system where each znode can contain data and have children.

Through a visual client like ZooInspector you can view the znode hierarchy and see which components use Zookeeper.

2.2 Znode Types

2.2.1 Node Types

Znodes are either temporary (ephemeral) or persistent.

Ephemeral nodes exist only while the session that created them is active; they are automatically removed when the session ends and cannot have children.

Persistent nodes survive session termination and are removed only when explicitly deleted.

Zookeeper also assigns a sequential number to nodes when they are created with the -s flag, e.g., /diaozhatian0000000037 .

# create sequential node
[zk: localhost:2181(CONNECTED) 2] create -s /diaozhatian aaaaaa
Created /diaozhatian0000000037
[zk: localhost:2181(CONNECTED) 5] get /diaozhatian0000000037
aaaaaa

# create persistent node
[zk: localhost:2181(CONNECTED) 3] create /diaozhadi bbbbbbbb
Created /diaozhadi
[zk: localhost:2181(CONNECTED) 6] get /diaozhadi
bbbbbbbb

2.2.2 Node Attributes

Each znode has attributes such as cZxid, ctime, mZxid, mtime, version numbers, ACL version, and data length, which can be retrieved with the get command.

# create sequential ephemeral node and get attributes
[zk: localhost:2181(CONNECTED) 7] create -s -e /niubility 666
Created /niubility0000000039
[zk: localhost:2181(CONNECTED) 8] get /niubility0000000039
666
cZxid = 0x70159
ctime = Sun Apr 05 10:59:45 CST 2020
mZxid = 0x70159
mtime = Sun Apr 05 10:59:45 CST 2020
...

3. Core Functions

3.1 File System

Zookeeper’s file‑system‑like hierarchy allows each znode to store data, but unlike a traditional file system, every znode (even those with children) can hold data.

# create parent and child znodes with data
[zk: localhost:2181(CONNECTED) 2] create /father zhangsan
Created /father
[zk: localhost:2181(CONNECTED) 4] create /father/child lisi
Created /father/child
[zk: localhost:2181(CONNECTED) 5] get /father
zhangsan
[zk: localhost:2181(CONNECTED) 6] get /father/child
lisi

Paths are absolute, starting with /.

Zookeeper has no separate file and directory concepts; only znodes.

3.2 Cluster Management

3.2.1 Node Join/Leave

Clients create an ephemeral node under a persistent /GroupMembers parent; Zookeeper watches child changes to detect node liveness.

3.2.2 Master Election

Master nodes create an ephemeral znode; the first to succeed becomes the active master, while others watch for its removal to trigger a new election.

The mechanism relies on the fact that an ephemeral node disappears when its session ends, ensuring only one active master at a time.

3.3 Distributed Locks

Zookeeper’s strong consistency guarantees unique node creation, enabling both exclusive (write) locks and shared (read) locks.

3.3.1 Exclusive Lock

An exclusive lock allows only one transaction to hold the lock at a time; Zookeeper ensures global uniqueness of the lock node.

3.3.2 Shared Lock

Multiple transactions can hold a shared lock simultaneously for read‑only access.

3.4 Watcher Mechanism

A watcher is a one‑time trigger that notifies a client when the watched znode changes, used in master election, lock release, and other coordination tasks.

Watcher Characteristics

Watches are one‑time; they must be re‑registered for subsequent events.

Watcher notifications are sent asynchronously from server to client.

4. Practical Applications

4.1 Hadoop

Zookeeper is used for Hadoop HA via the zkfc mechanism to elect an active NameNode and ResourceManager, and to store application metadata for YARN fault tolerance.

4.2 HBase

RegionServers register themselves in Zookeeper; the HMaster watches these nodes to monitor server health and to locate the META table.

4.3 Kafka

Kafka uses Zookeeper for broker registration, topic metadata storage, and consumer group coordination.

4.4 Hive

HiveServer2 uses Zookeeper for leader election (shared lock) and Hive tables use Zookeeper locks to guarantee atomic read/write operations.

5. Conclusion

Zookeeper is the heart of many big‑data systems, providing reliable coordination, configuration, naming, and lock services that keep clusters running smoothly.