With the increased adoption of a microservices architecture, we're building more distributed applications than before. Although there are many benefits of a decoupled architecture, the complexity and the effort involved in proving the correctness of the overall system has dramatically increased. Jepsen provides much needed tooling to verify correctness in coordination of task schedulers, test eventual consistency, linearizability and serializability characteristics of distributed databases. We've used Jepsen in a few projects and we like the fact that we can test drive configurations, inject and correct faults, and verify the state of the system after recovery.
With the growth in usage of NoSQL data stores, and the growth in popularity of polyglot approaches to persistence, teams now have many choices when it comes to storing their data. While this has brought many advantages, product behavior with flaky networks can introduce subtle (and not so subtle) issues that are often not well understood, even in some cases by the product developers themselves. The Jepsen toolkit and accompanying blog have become the de-facto reference for anyone looking to understand how different database and queuing technologies react under adverse conditions. Crucially, the approach to testing, which includes clients in the transactions, shines a spotlight on possible failure modes for many teams building microservices.
