Make dependency aggregate query strategy configurable - SELECT/JOIN

[mipo256]

When debugging the spring-data-jdbc queries, we found out, that for two entities, having OneToMany relationship between aggregates, like the two below (simplified for the sake of an example)

interface MyEntityRepo extends CrudRepository<MyEntity, Long> { }

@ToString
class MyEntity {

    @Id
    private Long id;

    @MappedCollection(idColumn = "my_entity_id", keyColumn = "my_entity_id")
    private List<Dependency> dependencies;
}

@ToString
class Dependency {

    @Id
    private Long id;

    private String type;

    private Long myEntityId;
}

if we will trigger an myEntityRepo.findById(), then there will be 2 selects issued, not one select with the join. I mean, we get:

SELECT "my_entity"."id" AS "id" FROM "my_entity" WHERE "my_entity"."id" = ?

and then

SELECT "dependency"."id" AS "id", "dependency"."type" AS "type", "dependency"."my_entity_id" AS "my_entity_id", "dependency"."my_entity_id" AS "my_entity_id" FROM "dependency" WHERE "dependency"."my_entity_id" = ? ORDER BY "my_entity_id"

instead of having something like:

SELECT ... FROM my_entity me INNER JOIN dependency d WHERE me.id = ?

That's a crucial thing. Because issuing 2 selects is generally, depends on a lot of conditions and particular DBMS worse. It is espesially the case when it comes to small selection subsets in the huge data set, which is a very common thing. There is an open discussion about that in DBA stackexchange.

The framework should at least provide a the way to choose per relation basis, like an additional attribute to MappedCollection annotation. That seems to be a quite large change, but still I think it is vital for the project in general

[schauder]

I consider this a duplicate of #592 which just generally asks for a better handling of this situation.

Adding configurability to this could be a later improvement on top of that, but we can decide that only ones we have the basic infrastructure in place.

[mipo256]

Ok, I think we can close it as a duplicate, but I will add the comment in the aforementioned ticket anyway to just keep this improvement in mind