Introduction

Introduction

• coordinate processes of distributed applications
• guarantee of FIFO execution per-client
• expose an API that lets developers create their own primitives
  • coordination kernel
• two important properties
  • FIFO client ordering
  • linearizable writes
• use “watches” to “subscribe” to updates on particular objects

ZooKeeper Service

• Abstract all data as znodes
• filesystem hierarchy
  • exploit to split different applications which may user zK
  • store small amounts of metadata info
• ephemeral and regular znodes
  • only regular znodes may have children
• linearizability is actually A-linearizability
  • (Asynchronous)
• primitives implementable on ZK
  • configuration management
  • rendezvous
  • group membership
  • simple locks

ZK Implementation

• assume faulty servers
• metadata tree held in-memory
• broadcast updates to all ZK nodes through Zab, an atomic broadcast protocol

Zab provides stronger order guarantees than regular atomic broadcast. More specifically, Zab guarantees that changes broadcast by a leader are delivered in the order they were sent and all changes from previous leaders are delivered to an established leader before it broadcasts its own changes

• use a “fuzzy” snapshot which doesn’t lock state while taking checkpoint
• can do this because state changes are idempotent

• timeouts to detect client failures