NoSQL is the umbrella label for a wave of non-relational database technologies that gained prominence around 2009. The name is often read as “not only SQL,” reflecting that these systems were a reaction to, rather than a wholesale rejection of, the relational model that had dominated for decades. On his own site, Martin Fowler describes NoSQL as “a wave of interest in alternative database technologies” and notes that the databases involved “were a wide-ranging bunch, including Mongo, Neo4j, Cassandra, and Riak.”
Despite the single name, NoSQL is not one thing. Fowler observes that “most sources I’ve looked at mention at least four groups of data model: key-value, document, column-family, and graph.” Key-value stores treat data as opaque values fetched by a key; document databases store self-describing records such as JSON; column-family or wide-column stores group columns for sparse, large-scale tables; and graph databases store entities and their connections. The first three, Fowler notes, tend to share a “rich structure of closely related data” as their unit of storage.
The motivation behind these systems was largely about scale and flexibility. Where a traditional relational database emphasizes a fixed schema and strong ACID transactional guarantees, many NoSQL systems relaxed those guarantees to spread data across many commodity machines, accept schema-less or flexible records, and stay available even when parts of the cluster fail. This is closely tied to ideas like the CAP theorem and eventual consistency, which formalize the trade-offs distributed databases must make.
Fowler is careful not to overstate the movement. He writes that he has “never seen these databases as supplanting relational databases as the dominant data storage approach,” but does see them “playing a significant role in any data architecture.” The lasting effect of NoSQL was less the death of SQL and more the normalization of choosing a data store to fit a problem, drawing on his and Pramod Sadalage’s book NoSQL Distilled.