71

Consider the following JPQL query:

SELECT foo FROM Foo foo INNER JOIN FETCH foo.bar bar WHERE bar.baz = :baz

I'm trying to translate this into a Criteria query. This is as far as I have gotten:

var cb = em.getCriteriaBuilder(); var query = cb.createQuery(Foo.class); var foo = query.from(Foo.class); var fetch = foo.fetch(Foo_.bar, JoinType.INNER); var join = foo.join(Foo_.bar, JoinType.INNER); query.where(cb.equal(join.get(Bar_.baz), value);

The obvious problem here is that I am doing the same join twice, because the Fetch class doesn't seem to have a method to get a Path. Is there any way to avoid having to join twice? Or do I have to stick with good old JPQL with a query as simple as that?

Improve this question
3
  • 1
    Well, thanks, but I'd much rather stick with the standard APIs and try to avoid additional third party libraries. If what I want to do is not possible with the JPA Criteria API, I'll probably just stick to plain JPQL. Commented Apr 28, 2011 at 9:40
  • Did you solved your problem? I have same issue. Fetch cannot be casted to Join, and I cannot get Path from Fetch. It's practically unusable. Only solution is to have two same joins, which is not acceptable. Commented Apr 22, 2015 at 7:17
  • 4
    Well, James' answer describes the root of the issue quite nicely. You just can't do it that way, and that is a sound design decision. If I remember correctly, I ended up actually joining twice. That being said, I'll never use JPA again in the first place if I have the choice, because I think it's a useless abstraction layer that adds unnecessary complexity while castrating the underlying implementation. Commented Apr 22, 2015 at 12:22

3 Answers 3

Reset to default
102

In JPQL the same is actually true in the spec. The JPA spec does not allow an alias to be given to a fetch join. The issue is that you can easily shoot yourself in the foot with this by restricting the context of the join fetch. It is safer to join twice.

This is normally more an issue with ToMany than ToOnes. For example, 

Select e from Employee e join fetch e.phones p where p.areaCode = '613'

This will incorrectly return all Employees that contain numbers in the '613' area code but will left out phone numbers of other areas in the returned list. This means that an employee that had a phone in the 613 and 416 area codes will loose the 416 phone number, so the object will be corrupted.

Granted, if you know what you are doing, the extra join is not desirable, some JPA providers may allow aliasing the join fetch, and may allow casting the Criteria Fetch to a Join.

Improve this answer
Sign up to request clarification or add additional context in comments.

6 Comments

Brilliant answer, thanks. Hadn't thought of that. Hibernate never complained to me about aliasing a fetch join, I wasn't aware that this actually violates the spec.
See also, java-persistence-performance.blogspot.com/2012/04/…
I was doing exactly this kind of query and wondering what the results would be like in this exact case. This solves my confusion.
I find this answer incorrect. In my case query like this returns only employees that have all phones with areaCode = '613'. It does not filter the phones collection like you said.
Join fetch is the whole point of Join. Object does not get corrupted when its nested collections are filtered out - regardless whether it's user code or JPA join fetch.
|
39

I will show visually the problem, using the same example from James answer and adding an alternative solution.

When you do the follow query, without the FETCH:

Select e from Employee e join e.phones p where p.areaCode = '613'

You will have the follow results from Employee as you expected:

EmployeeId EmployeeName PhoneId PhoneAreaCode
1 James 5 613
1 James 6 416

But when you add the FETCH clause on JOIN (FETCH JOIN), this is what happens:

EmployeeId EmployeeName PhoneId PhoneAreaCode
1 James 5 613

The generated SQL is the same for the two queries, but the Hibernate removes on memory the 416 register when you use WHERE on the FETCH join.

So, to bring all phones and apply the WHERE correctly, you need to have two JOINs: one for the WHERE and another for the FETCH. Like:

Select e from Employee e join e.phones p join fetch e.phones //no alias, to not commit the mistake where p.areaCode = '613'

Maybe in the newest versions of Hibernate you will need to use SELECT DISTINCT to avoid duplicate results.

Improve this answer

6 Comments

Can you a little explain it more with SQL query, why the first query will return the employees with areaCode 613 and 416 Because there is a where clause, and if I understand it correctly then there will be an inner join between Employee and Phone. Then why it will return the 613 and 416? Shouldn't it return only 613?
@FaizanAhmad the first query will return only the employee entity "James", because one of his two addresses has the 613 code. When you use fetch join, some developers expect to also receive these two addresses, because one of them is 613, but hibernate filter in memory and exclude the 416 from the result, returning only the 613. So, you need one join for the fetch (to bring all addresses) and another one for the where condition (return any employee that has some address with 613 code).
Sorry but from a performance point of view this solution is just terrible as it will produce n^2 rows with n being the size of the collection.
@FaizanAhmad You would be right if we thought in pure SQL, but here we are selecting entities, which can "embed" relations. The Employee entity contains the list of all phones of the employee (usually OneToMany association). So as we are selecting the whole employee entity (e), whatever the where clause is, it is indeed supposed to contain all the phones of that employee. The problem here is that when using fetch, hibernate internally excludes from the e.phones list the values that or not in the where clause.
With latest versions, we get a duplicate result row, with this solution. You can fix it by using SELECT DISTINCT .... Would be awesome if you could update your answer
|
3

I may answer late this but from my point of view.

Select e from Employee e join e.phones p join fetch e.phones //no alias, to not commit the mistake where p.areaCode = '613'

This could be translated to the following SQL Query

Select e.id, e.name, p.id ,p.phone From Employe e inner join Phone p on e.id = p.emp_id where exists( select 1 from Phone where Phone.id= p.id and Phone.area ='XXX' )

This will get all phones of an employee that belongs to an area.

BUT

Select e from Employee e join fetch e.phones p //no alias, to not commit the mistake where p.areaCode = '613'

could be translated to the following SQL Queries

Select e.id, e.name, p.id ,p.phone From Employe e inner join Phone p on e.id = p.id Where p.area ='XXX'

or

Select e.id, e.name, p.id ,p.phone From Employe e inner join Phone p on e.id = p.emp_id and p.area ='XXX'

this will restrict row selection to only rows where employees phone is of area XXX

And finally writing this

Select e from Employee e join e.phones p where p.areaCode = '613'

Could be seen as

Select e.id, e.name from Employe e where exists ( select 1 from phone p where p.emp_id = e.id and p.area = 'XXX' )

Where we are only getting employee data that have a phone number in some area

This should help get the idea after each query.

Improve this answer

Comments

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.