As outlined in this discussion (https://hibernate.atlassian.net/browse/HHH-10485?page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel&showAll=true) when using the EntityGraph class to query for data a JOIN is always used to construct the query to retrieve the data rather than a secondary select (the @Fetch annotation is ignored).
In the later part of the discussion the following question was raised but not answered
“Will we ever consider that if we annotate an association with @Fetch(FetchMode.SELECT) and the user wants to fetch it via Entity Graphs, then we should use secondary SELECTs rather than JOINs?”
Not allowing the @Fetch annotation to be considered limits the usefulness of using an EntityGraph in the case where there is more than one collection as discussed here Way to solve n+1 & cartesian product problem.
One of the pain points of using an ORM is often the need to translate from an Entity to DTO (and back). I think the addition of dynamic EntityGraphs with a combination of FetchMode.SUBSELECT and Hibernate filters
promises a 80-90% reduction in the need for a DTO layer in many use cases as the desired graph of entities can be simply specified and then populated at run time using the EntityGraph class only. However this requires a secondary select to populate the collections. Admittedly currently a Set could be used to house the collections but the underlying query will still produce a cartesian join.
Is anyone able to answer the question “Will we ever consider that if we annotate an association with @Fetch(FetchMode.SELECT) and the user wants to fetch it via Entity Graphs, then we should use secondary SELECTs rather than JOINs?” Other comments?