I know exactly what you mean. I've realised that when you're solving a hard problem (complex data logic, extreme performance requirements) with SQL, you end up systematically experimenting and thinking very deeply within a rigidly defined system. With no-sql databases, you're monkeying with obscure APIs that cover up obscure, idiosyncratic, and hard-to-reason about database engines, and patching strange behavior with rickety app-layer code.

Both kinds of work are hard, but with SQL you can usually expect the problem to yield to intense engagement and produce a solution that's complete, robust, and satisfying.