Kyle Kingsbury, known as Aphyr on Twitter and for his blog by the same name, talks to Stefan Tilkov about consensus in distributed systems and about his experience in testing systems to see how they behave in case of failures. In addition to discussing some of the theoretical foundations, such as the CAP theorem, isolation levels, and consensus protocols, Kyle talks about some specific databases, including MongoDB, Riak, and Redis, and discusses how they maintain and achieve — or fail to achieve — a consistent state. Finally, there's some advice for practitioners on how to pick a solution and understand its properties.