IMHO, there are very few cases when SQL is NOT the answer, when dealing with data. When dealing with structured data, NoSQL is for sure not the answer.

Agreed, standard SQL engines are not (yet) as easy to scale as Cassandra and friends. But they’re catching up quickly, I suggest taking a look at engines like FaunaDB, CockroachDB, YugabyteDB. And then Google Spanner.

But standard engines have something else, which is more subtle. Take some time to think about this: how many times did you wrote some code that behaves differently based on the quantity and quality of the input data ? Guess what, that’s exactly how engines like Oracle, SQL server or Postgres have on the table for you… Try to do on-the-fly histogram collection and programming paths accordingly, and see how far you can go.