ZooKeeper is Hadoops distributed coordination service. It can’t make partial failures go away but it can help you handle partial failures.

• Its simple, a stripped down filesystem that exposes a few simple operations
• Its expressive, it can be used to build coordination data structures and protocols
• Its available, applications can depend on it
• It helps loosely coupled interactions, helps computers find other finders
• its a library, It lets open source use tried and tested protocols

Example

How can we have an actively maintained list of active servers?

• With zookeeper you can
• You have znodes, that form a hierarchical namespace. e.g /zoo and /zoo/duck

ZooKeeper service

Data Model

• Data access is atomic. In reads and writes. No partial failures.
• Znodes can be ephemeral or persistent. Ephemeral nodes are deleted when creating client session ends.
• You can have sequential znodes, where file paths are generated with numbers. This gives you clear global ordering!
• You can create watches. One off znode change notification alerts!

Zookeeper service runs on a cluster. Gets high availability with replication.

• If 2/5 nodes fail its fine.
• Uses Zab protocol to do this (leader election, atomic broadcast)

All updates to znode tree have a globally unique identifier.

Building applications with ZooKeeper

Zookeeper comes with prebuilt distributed data structures and protocols you can use.

• barriers
• queues
• two phase commits