How can I abstract database repository from data service?

Question

I was reading at this article https://www.infoq.com/articles/spring-data-intro to understand how can a data service layer be independent of database(RDBMS /NoSQL). It looks like there's no way to design entity and repository to be independent of database. This article was written on 2012. Do we've any other technologies since then that has implemented this feature?

Answer 1

Before actually answering your question I have to ask: Why do you want to do that? Think twice because abstraction comes at a cost and just doing it in order to have a "clean" design is most certainly not worth it.

Now to your question:

1. There is no library or framework that does this out of the box.

2. Probably the thing getting closest to it is spring data which you are obviously aware of. If you stick to the persistent storage independent interfaces your repositories will abstract over the persistence store use at least to some extent. BUT: you will have to provide different kinds of metadata (typically as annotations on the entities) to make it work. So in this sense, it is a really leaky abstraction.

3. Of course, you can roll your own: create an interface with the operations you need and provide implementations for the different data stores you want to use. Also, include a storage independent way of providing metadata.

So the question becomes: Why hasn't anybody done that yet? And why you probably shouldn't do it either?

1. It is hard: Just writing SQL in a way that all (relevant) SQL databases understand is difficult, see this for an example.

2. You loose a lot of power of your stores. For example, RDBMSes are great at joining stuff. But joining is basically a no go for many No-SQL databases. So your API probably shouldn't offer this feature. This basically dumbs everything down to the common denominator, which is going to be really small when you have very different data stores.

3. It is not worth it. This leads back to my opening question: Why would you want to do it anyway? I certainly see the use of switching different RDBMSes. Some companies only want certain vendors in their datacenter, it is nice to have an in memory variant for testing and so on. But switching a store from for example Oracle to Hazelcast and then to MongoDB to CSV? Why would you want to do that? What is the business value of that?