Add a Group by to a Hibernate Criteria-Query without Projection

Question

I have a Criteria-Query, which joins a second table B, to select entities from table A. The problem is, that this query returns some entities from table A multiple times. But I need the results to be distinct.

Using Criteria.DISTINCT_ROOT_ENTITY is useless, becaus this filters out the multiple occurences after the SQL-Query was executed. So, when I limit my results to 20 hits, I end up with only 4, though there are more entries, that match my query.

In pure SQL I simply can add a "GROUP BY ID" to the query and everything is fine, because the join of table B is only used, to select the entities from table A. But with the Criteria-API I cannot do this. The only way to add a "GROUP BY" is by using Projections. But then, I end up with scalar values, not with a real instance of my class. Using a SQL-restriction does not work either, because hibernate adds a bogous "1=1" after my "GROUP BY"-clause. :(

Any ideas?

Answer 1

Have you tried to use something like this?

    ICriteria criteria = dc.GetExecutableCriteria(RepositoryInstance.Session)
            .SetProjection(Projections.distinct(Projections.projectionList()
                    .add(Projections.property("Prop1"), "Prop1")
                    .add(Projections.property("Prop2"), "Prop2")
                    .add(Projections.property("Prop3"), "Prop3")
                    .add(Projections.property("Prop4"), "Prop4")));
    result = criteria.List();

You can dynamically add properties through reflection of the class.

This creates SQl like this: select distinct prop1,prop2,prop3,prop4 from yourClass

I did not include DetachedCriteria dc since that is irrelevant.

Answer 2

GROUP BY WITHOUT PROJECTION: Its not possible as it make sense, in many answers you may found, But most people don't want to use projection, because it require them to project each and every attribute, but requirement is that a bean must be projected. (and returned as a result). In example below I have tried to project the required bean as resultant object.

I have achieved the same result with a little bit of trick I believe, First I was trying to apply group by without projection but I have found no solution, so I have to rely on Projection.

Here is what I wanted to achieve

select p.* FROM parent p INNER JOIN child c ON p.id_parent=c.id_father
WHERE c.child_name like '%?%' AND p.parent_name like '%?%' 
group by p.id_parent

In Java code I wanted p.* to be a Parent class which is my entity bean and I wanted it be unique, one way is get the result list in a Set, but i dont like this way due many reasons :)

So I created a Criteria from Child.class instead of Parent.class , and this trick worked for me.

Criteria c = session.createCriteria(Child.class,"c");// starting from Child
    c.add(Restrictions.like("childName",   "abc", MatchMode.ANYWHERE));
    c.createAlias("parent", "p"); //remember parent is an attribute in Child.class
    c.add(Restrictions.like("p.parentName",   "xyz", MatchMode.ANYWHERE));
    c.setProjection( Projections.projectionList().add(Projections.groupProperty("parent"))); //projecting parent which is an attribute of Child.class

    List<Parent> result = c.list(); //get the result
    for (Parent p: result) {
        System.out.println(p);
    }

If you still haven't got the idea here are my mapped Entity Bean classes.

package com.mazhar.beans;

import static javax.persistence.GenerationType.IDENTITY;

import java.util.List;

import javax.persistence.CascadeType;
import javax.persistence.Column;
import javax.persistence.Entity;
import javax.persistence.FetchType;
import javax.persistence.GeneratedValue;
import javax.persistence.Id;
import javax.persistence.OneToMany;
import javax.persistence.Table;

@Entity
@Table(name = "parent")
public class Parent {
    private Integer idParent;
    private String parentName;
    private List<Child> childs;

    @Id
    @GeneratedValue(strategy = IDENTITY)
    @Column(name = "id_parent")
    public Integer getIdParent() {
        return idParent;
    }
    public void setIdParent(Integer idParent) {
        this.idParent = idParent;
    }

    @Column(name = "parent_name")
    public String getParentName() {
        return parentName;
    }
    public void setParentName(String parentName) {
        this.parentName = parentName;
    }

    @OneToMany(fetch=FetchType.LAZY, mappedBy="parent", cascade=CascadeType.ALL)
    public List<Child> getChilds() {
        return childs;
    }
    public void setChilds(List<Child> childs) {
        this.childs = childs;
    }

}

and my child class

package com.mazhar.beans;

import static javax.persistence.GenerationType.IDENTITY;

import javax.persistence.CascadeType;
import javax.persistence.Column;
import javax.persistence.Entity;
import javax.persistence.FetchType;
import javax.persistence.GeneratedValue;
import javax.persistence.Id;
import javax.persistence.JoinColumn;
import javax.persistence.ManyToOne;
import javax.persistence.Table;

@Entity
@Table(name = "child")
public class Child {
    private Integer idChild;
    private String childName;
    private Parent parent; //this actually we projected in criteria query.

    @Id
    @GeneratedValue(strategy = IDENTITY)
    @Column(name = "id_city", unique = true, nullable = false)
    public Integer getIdChild() {
        return idChild;
    }

    public void setIdChild(Integer idChild) {
        this.idChild = idChild;
    }

    @Column(name = "city_name", nullable = false)
    public String getChildName() {
        return childName;
    }

    public void setChildName(String cName) {
        this.childName = cName;
    }

    @ManyToOne(fetch = FetchType.LAZY, cascade = CascadeType.ALL)
    @JoinColumn(name = "id_father")
    public Parent getParent() {
        return parent;
    }

    public void setParent(Parent parent) {
        this.parent = parent;
    }
}

Answer 3

It is possible to write actual SQL Queries which Hibernate can use to return entities. So if you really need to, you can bypass HQL and write exactly the query you want with your GROUP BY in it.

See here for details.

For example you can define a query something like this in your hbm.xml file:

<sql-query name="exampleQuery">
<query-param name="param" type="int"/>
<return alias="x" class="foo.bar.X"/>
<return alias="y" class="foo.bar.Y"/>
    <![CDATA[
        select {x.*}, {y.*}
        from XEntity x
        inner join YEntity y on y.xId = x.id
        where y.value = :param
    ]]>
</sql-query>

Note the {x. } and {y. } shorthand syntax for selecting all the properties of entity X and entity Y

Answer 4

The main problem of grouping by without projecting, in the way you want, is that in some DBMS like Oracle it will not work, the Oracle will return an error.

If you group a select, you have to group by all non-aggregation fields that you are selecting. The MySQL for example don't have this restriction.

The approach that i've been using is selecting only the id as groupProperty projection with all filters, orderings and number of results limit. Then a execute other query filtering with these retrieved ids. That way the implementation will be independent of the DBMS.

Answer 5

The workaround I used was to add sqlRestriction("TRUE GROUP BY this_.some_field") as the last criteria builder call. query builder will not put any warnings and will just put it at the end of WHERE clause which will end up as "where a=1 and b=2....and true group by this_.some_field" Do not forget to use 'this_' prefix to group by root entity field or you can launch the application and see how alias is named inside the query if this is joined table property you want to use for grouping.

this approach helped to solve n+1 rows problem for a query due to joins and delay full refactoring until the time is available.

tested on Hibernate 3.4 + MySQL 5.6