Providing Flexible Database Consistency Levels with Manhattan at Twitter

Explore Twitter's primary distributed key-value store, Manhattan, in this 40-minute conference talk from GOTO Chicago 2016. Dive into the general architecture of the storage system, focusing on motivations, tradeoffs, and flexible database consistency levels. Learn about Manhattan's adoption, data model, partitioning, messaging, and consistency architecture. Understand the benefits of eventual consistency and when it may not be sufficient. Discover how Twitter adapted the architecture to address challenges, and examine different approaches to consistency models, topology transitions, and time definitions. Gain insights into the user experience and the results of implementing these changes in a large-scale distributed system.

Introduction
USING MANHATTAN
ADOPTION
BUILDING NEW FEATURES
ARCHITECTURE: DATA MODEL
ARCHITECTURE: PARTITIONING
ARCHITECTURE: MESSAGING
ARCHITECTURE: CONSISTENCY
BENEFITS OF EVENTUAL CONSISTENCY
WHEN IT'S NOT ENOUGH
ADAPTING ARCHITECTURE
RESULTS
A DIFFERENT CONSISTENCY MODEL
A DIFFERENT TOPOLOGY TRANSITION
A DIFFERENT DEFINITION OF TIME
WHAT THE USER SEES