What ZooKeeper Is (and Isn't)

A Coordinator, Not a Database

ZooKeeper is a coordination service for distributed systems.

Its job is helping many machines agree: who is the leader, which nodes are alive, who holds a lock.

It is not a place to store your application data. That mistake causes most ZooKeeper failures in production.

The Znode Tree

ZooKeeper stores everything in a tree of znodes, much like a filesystem.

Each znode has a path, such as /services/api/leader, and can hold a small piece of data.

A znode can also have children, so the tree gives you natural grouping: all locks under one branch, all live members under another.

The 1MB Limit

Every znode carries a hard size limit: 1MB of data.

That sounds restrictive until you remember what belongs there. A hostname, a lock owner, a config flag, a leader identifier.

All of those fit in a few hundred bytes. The limit is a guardrail, telling you when you have started storing the wrong kind of thing.

Why Small Data Matters

ZooKeeper copies every write to every server in the ensemble before confirming it.

That agreement is what makes the data trustworthy, and it is also what makes it expensive.

Large values would slow every member down at once. Small, coordination-only data keeps that agreement fast.

Redis and Cassandra Are Different

Redis and Cassandra are built to hold application data at volume.

ZooKeeper is built to hold tiny facts that many machines must agree on.

Putting user sessions or event records in ZooKeeper is using a coordination service as a storage engine, and it will not end well.

A Test Before You Write

Before putting anything in a znode, ask one question. Does every machine need to agree on this fact?

If yes, and it fits in a few hundred bytes, ZooKeeper is the right home.

If it is application data, or if it is pushing toward 1MB, send it to Redis or Cassandra instead.

Is ZooKeeper intended to store user sessions and event records as application data?

• Yes
• No

Answer: No

No. ZooKeeper is a coordination service for shared facts such as leaders, locks, and live nodes. User sessions and event records belong in an application-data store such as Redis or Cassandra.

ZooKeeper is a _____ for distributed systems; Redis and Cassandra are built to hold _____ at volume.

Answer: ZooKeeper is a coordination service for distributed systems; Redis and Cassandra are built to hold application data at volume.

ZooKeeper stores tiny facts that machines must agree on, while Redis and Cassandra are designed for high-volume application data.

Why does ZooKeeper keep coordination values small?

• Every write is copied to every server before confirmation.
• Large values cannot be represented in a znode tree.
• Redis and Cassandra reject values larger than a few hundred bytes.
• Small values prevent znodes from having child nodes.

Answer: Every write is copied to every server before confirmation.

ZooKeeper copies every write to every server in the ensemble before confirming it. Larger values would therefore slow all members during agreement, while small values keep coordination fast.