Verifying Transactional Consistency with Jepsen and FaunaDB

I’d like to give an update on our efforts to verify FaunaDB’s transactional correctness guarantees.

FaunaDB is a mission critical NoSQL database. Most database technologies sound like a joke version of the “good, fast, or cheap” mantra: usable, scalable, or correct, choose two, but more often just choose one, and sometimes you get zero. Instead, FaunaDB endeavors to be all three at once. Today we will talk about becoming correct.

FaunaDB passes these tests.

Up until this point, correctness in most distributed systems was purely aspirational, potentially excepting Tandem NonStop. It turned out that essentially no systems did what they said on the tin, due to both implementation defects and because they made claims that physics and information science unfortunately do not allow.

Jepsen has since become capable of simulating a much larger variety of machine and network failure scenarios, and of checking a system’s more formal isolation characteristics, such as its ability to maintain sequential consistency, snapshot isolation, linearizability, and liveness in the presence of these failures.

Thanks to Brandon Mitchell, Nathan Taylor, Jeff Smick, Attila Szegedi, and the whole Fauna engineering team for their hard work on the Jepsen analysis. Evan Weaver and Dhruv Gupta contributed to this post.

Related posts:

Developing an app today is no rocket science. However, for that app to be successful, you need to have a unique idea along with excellent execution. What we essentially mean is that if your app idea…