hibernate - QueryDSL generates cross join -

i looking offer content filtering results. (edited brevity) entities following:

node:

@entity @inheritance(strategy = inheritancetype.joined) public abstract class node {     @id     @generatedvalue     public integer id; } 

scene:

@entity public class scene extends node {     @onetomany(mappedby = "scene", /*fetch = fetchtype.eager, */cascade = cascadetype.all)     public list<source> sources; } 

source:

@entity public class source extends node {     @manytoone     public scene scene;      @enumerated(enumtype.ordinal)     public sourcetype type; } 

what follows example of filter wish implement.

given collection of sourcetypes wish select scenes such scene referred source of each 1 of types. achieve using querydsl following predicate:

private predicate filterbysourcetype(collection<sourcetype> it) {     booleanbuilder bb = new booleanbuilder();      (sourcetype st : it)         bb.and(qscene.scene.sources.any().type.eq(st));      return bb.getvalue(); } 

a series of these predicates combined give overall query. when selecting 1 sourcetype resulting query looks so:

hibernate: select count(scene0_.id) col_0_0_ scene scene0_ inner join node scene0_1_ on scene0_.id=scene0_1_.id exists (select 1 source source1_ inner join node source1_1_ on source1_.id=source1_1_.id (source1_.id in (select sources2_.id source sources2_ inner join node sources2_1_ on sources2_.id=sources2_1_.id scene0_.id=sources2_.scene_id)) , source1_.type=?)

i believe happens above cross join , result (2k scenes, 1k sources) query takes multiple seconds.

i tried switching concrete class polymorphism eliminate node joins no noticeable performance improvement.

how optimise predicate?

turns out can use jpasubquery collectionexpression , write out query how write plain old sql.

        bb.and(qscene.scene.in(                 new jpasubquery()                         .from(source)                         .where(source.type.eq(st))                         .list(source.scene)         ));