Having spent the bulk of this book exploring Cassandra using CQL, we've now peeked under the hood to see how the robust data structures offered by CQL are in fact abstractions upon a much more rudimentary column family data structure. If nothing else, this is an opportunity to appreciate modern technology; as recently as Cassandra 1.1, the CQL data structures we take for granted were unheard of. Before Cassandra 0.7, CQL did not exist at all; the Thrift interface we explored in this chapter was state-of-the-art.

Our deep dive into Cassandra's internal data structures also helps us understand the underlying reasons behind facts that in previous chapters we'd taken as first principles. You learned in Chapter 3, Organizing Related Data that partition keys group related data, but now we know that in fact all the data under a partition key, or RowKey, coexists in a single wide row. Similarly, we already knew that clustering columns determine row ordering, but now we know that the sorting is actually happening to cells in a wide row, and the use of the clustering column as a prefix for wide row cell names generates the familiar behavior at the CQL level.

In the final appendix, we will do a brief survey of Cassandra's authentication and authorization functionality, and discuss the various tools Cassandra gives you for protecting your deployments against security vulnerabilities.