Hyperdex might seem like just another kid on the street of open-source distributed key-value stores. On a second look, however, it appears to achieve magic: it provides strong consistency (all clients always see the same order of events), fault tolerance up to a definable number of simultaneous failures, and queries on secondary attributes that are practically free. Next to these much stronger guarantees, it significantly outperforms MongoDB and Cassandra across a wide variety of workloads. In this talk, we will look at the key concepts behind this "magic": "hyperspace hashing" and "value-dependent chaining"; learn why this set of features does not violate the infamous CAP theorem, and look at a few practical concerns. Finally, a live session will show how to setup Hyperdex and how to work with its client.