Join Fetch applied twice for ElementCollection

[vokinpirks]

Hi!
I'm playing around with this library and encountered an issue.
I have pretty simple JPA model. There is element collection defined as follows:

@ElementCollection(fetch = FetchType.LAZY)
@GraphQLDescription("A set of user-defined tags")
@BatchSize(size = 16)
private Set<String> tags = new HashSet<>();

When I do NOT include this field in query everything works fine:

query {
  Tracks() {
    select {
      addedAt
      playCount
    }
  }
}

But when I do so, I get an empty result:

query {
  Tracks() {
    select {
      addedAt
      playCount
      tags
    }
  }
}

'tags' field is empty for all entities.
I checked generated JPQL/SQL queries and found out that there are two joins for tags, left and inner one. I guess I'm getting the empty result because of that inner join.
I started debugging and figured out that the second join is added at
com.introproventures.graphql.jpa.query.schema.impl.QraphQLJpaBaseDataFetcher.java:234
(root.fetches becomes [2] there).

And..join is a very inefficient way of fetching *ToMany/ElemectCollection relations. Batch fetching is a way better for that. Theoretically, is it possible to implement batch fetching in this library?

[igdianov]

@vokinpirks Thank you for reporting this problem. :) There should be only one fetch join generated for element collections here:

graphql-jpa-query/graphql-jpa-query-schema/src/main/java/com/introproventures/graphql/jpa/query/schema/impl/QraphQLJpaBaseDataFetcher.java

Line 233 in 9854252

from.fetch(selectedField.getName());

The first join is needed to fetch the data from the graphql query. The inner join should not be there unless you want to filter elements in your collection. It is probably a side effect from the predicate builder somewhere. I will try to create a test to find out and fix it.

The batch fetching is not part of JPA spec, so it cannot be used directly. I am also not sure what effect @BatchSize annotation has on the Hibernate to fetch element collections on top JPA query.

Graphql-java has something called DataLoader: https://github.com/graphql-java/java-dataloader. I have not figured out how to use it, but may be your case is a good candidate to do it.

[vokinpirks]

@igdianov the first join is left whereas the second one is inner.
FYI the first join appears right after this call:

graphql-jpa-query/graphql-jpa-query-schema/src/main/java/com/introproventures/graphql/jpa/query/schema/impl/QraphQLJpaBaseDataFetcher.java

Line 224 in 9854252

fetch = reuseFetch(from, selectedField.getName(), isOptional);

The line of code you are referring to always produces inner join despite the fact isOptional set to true. The method javax.persistence.criteria.FetchParent#fetch(java.lang.String) has an overloaded signature which takes JoinType as a parameter.
Shouldn't it be like
from.fetch(selectedField.getName(), isOptional ? JoinType.LEFT : JoinType.INNER);
?

Thanks for pointing me to DataLoader. Will take a look at that.

[igdianov]

@vokinpirks Yes, this is a bug. It should not be apply to fetch joins for element collections. It will be easy to fix.

[igdianov]

@vokinpirks I have fixed the issue. Let me know if it works for you.

[vokinpirks]

@igdianov thanks! I'll try this out.