QuerydslDataFetcher to specify which properties and associations to fetch

[JamesPeters98]

I've just been having a play with the library testing it with my current GraphQL implementation using https://github.com/leangen/graphql-spqr

I've tried using the QuerydslDataFetcher like this by passing in the DataFetchingEnvironment:

var job =
        QuerydslDataFetcher.builder(owningJobDataFetcher)
            .many()
            .get(env.dataFetchingEnvironment);

Just to clarify should this be joining entities requested in the dataFetchingEnvironment.

e.g if the query requested a job with a customer like this:

job {
  customer {
    name
    id
  }
}

From a Job Entity with a Customer Entity mapped ManyToOne, should it be joining the customer entity in the query?

Because currently I'm seeing N+1 behaviour, thought I'd clarify if Joining is supposed to happen before I investigate if it's a Hibernate mapping issue instead.

[schauder]

At least for *-1 relationships that is very reasonable.
I'll try to make it happen.

[rstoyanchev]

Scheduling tentatively.

[schauder]

After a little investigation I realised that under the hood the QuerydslDataFetcher is not just using some Spring Data infrastructure but directly calling a repository.
Yeah, I know this should have been clear to start with.

Anyway, repositories deliberately do not have a way to specify an EntityGraph or similar dynamically. The discussion about this is here: spring-projects/spring-data-jpa#951

But it makes a lot of sense for GraphQL.
Calling @mp911de to discuss how to get out of this conflict.

[mp911de]

That isn't something that we can solve already. We face demand for various types of projections (interface/DTO projections, limiting/expressing the properties to materialize) for Spring Data's Querydsl and Query-by-Example programming models so we should investigate how we can approach these requirements.

[mp911de]

We're investigating means to express eager-loading/how to prevent loading relationships using Spring Data JPA with GraphQL.

[austinarbor]

Fellow SPQR user here. I was able to solve this problem in my own application with the help of the Cosium/spring-data-jpa-entity-graph library. I can't post the exact code I wrote, but the general algorithm is this:

1. Get all the query fields from the data fetching environment, so you get something like [id, name, child/id, child/name, child/subchild/id, ...] and then split each entry on /
2. Get the Hibernate ManagedType for the root entity class the query begins with
3. Fetch the Hibernate ManagedType for the root entity's class
4. For each path array in step 1
   1. If metamodel does not contain path[0], ignore the path
   2. If the metamodel contains the path , but Attribute::isAssociation is false, ignore the path (we only care about including assocations in the entity graph)
   3. Otherwise, if the metamodel contains and node and it is an assocation, fetch the child managed entity from the metamodel and repeat for path[1], etc until you reach the end. If all steps return true, add the path to a list of entries
5. In the end, you will get a list of association nodes [child, child/subchild]. Replace each / with .. so you get [child, child.subchild]
6. com.cosium.spring.data.jpa.entity.graph.domain.EntityGraph = com.cosium.spring.data.jpa.entity.graph.domain.EntityGraphUtils.fromAttributePaths(paths)
7. Use this in your JPA method

[rstoyanchev]

This issue is superseded by PR #168 which implements the required changes.