Explore the challenges and solutions of building financial systems on eventually consistent databases in this conference talk from Strange Loop. Dive into Netflix's approach to managing billing for millions of global customers across multiple AWS regions. Learn about the tradeoffs between availability and consistency when using multi-master databases, and discover how the Billing Engineering platform developed a scalable, available, and strongly consistent solution. Examine techniques such as lightweight transactions in Cassandra, update consistency with optimistic locking, and fault tolerance strategies. Gain insights into building a financial system on top of Cassandra, including object modeling, creating a relational schema, and understanding performance implications. This talk provides valuable lessons for engineers working on large-scale financial systems in distributed environments.
Overview
Syllabus
Intro
Billing Engineering
Netflix operates out of multiple AWS regions to ensure availability
Single Master Single Point of Failure
Read replicas
Slave Promotion
We needed something better...
But it's not perfect, especially for a financial system
So how do you build a financial system on top of it?
Lightweight Transactions in Cassandra
Update Consistency With Optimistic Locking
Two Scenarios
With Cassandra, last write wins
How does Cassandra achieve fault tolerance?
Local Quorum vs Each Quorum
Insert Consistency
Some things to consider
Minimizing inconsistencies
Building a financial system on top of Cassandra
Our object model needed relations
We built a poor man's relational schema
Understand performance implications before using collection types
Querying across rows
Recap
Taught by
Strange Loop Conference
