Building an Event-Driven Data Mesh

Chapter 10. Eventual Consistency

Eventual consistency is one of the main concerns that people have with distributed systems and event-driven data products. But eventual consistency can mean different things to different people. For example, an application developer may be using a database that doesn’t offer consistent read-after-writes, such as in the case of a large distributed database or when using event sourcing to build up state. In an event-driven data mesh, we’re more concerned with the effects of multiple consumer systems subscribing to event-driven data products and how to work with individual consumers each reading data at their own rate.

There are people who have been looking at, working on, and thinking about eventual consistency for quite a long time. Pat Helland is just such a person and has written an excellent piece that collates insights and opinions from numerous thought leaders on the subject.

Since Doug [Terry] coined the phrase eventual consistency in the Bayou paper in 1995, I was interested in his perspective. When he defined eventual consistency, it meant that for each object in a collection of objects, all the replicas of each object will eventually have the same value. Then, he said: “Yeah, I should have called it eventual convergence.”

Pat Helland

Helland goes on to discuss a definition by Peter Alvaro, from his 2015 PhD thesis “Data-Centric Programming for Distributed Systems”:

A system is convergent or “eventually consistent” if, when all messages ...