Personally, I find that SQL is hard to generate. It's hard to reason about the creation of dynamic queries, say, for some types of searches. It's not reasonable

Personally, I find that SQL is hard to generate. It's hard to reason about the creation of dynamic queries, say, for some types of searches. It's not reasonable to create a generic function that can operate on any same-shaped piece of data. And it's ridiculously hard to manage the concept of a pointer to a piece of data which could be one of many types, a problem I'm sure most people have faces. Stored procedures can help with some of this, but then you've got to remember to update them every time you need to access data in a slightly different way, and it's hard to figure out when you can get rid of old stored procedures. Tools for managing them aren't up to scratch with tools for managing "real" programs. Personally, I'm keen on RethinkDB. It's a document-oriented NoSQL database which doesn't lose your data, has built-in clustering, and an extremely strong query language built on chaining method calls. While it doesn't have transactions, the query language is powerful enough that you can easily model most forms of deterministic data transformation within a single query.