Order by count using Spring Data JpaRepository

I've solved the puzzle using hints and inspirations from:

1. Limiting resultset using @Query anotations by Koitoer
2. How to order by count() in JPA by MicSim
3. Exhaustive experiments on my own

The first and most important thing I've not been aware of about spring-data is that even using @Query custom methods one can still create paging queries by simply passing the Pageable object as parameter. This is something that could have been explicitely stated by spring-data documentation as it is definitely not obvious though very powerful feature.

Great, now the second problem - how do I actually sort the results by size of associated collection in JPA? I've managed to come to a following JPQL:

select new package.AwithBCount(count(b.id) as bCount,c) from A a join a.bes b group by a

where AwithBCount is a class that the query results are actually mapped to:

public class AwithBCount{
    private Long bCount;
    private A a;

    public AwithBCount(Long bCount, A a){
        this.bCount = bCount;
        this.a = a;
    }
    //getters
}

Excited that I can now simply define my repository like the one below

public interface ARepository extends JpaRepository<A, Long> {
    @Query(
        value = "select new package.AwithBCount(count(b.id) as bCount,c) from A a join a.bes b group by a",
        countQuery = "select count(a) from A a"
    )
    Page<AwithBCount> findAllWithBCount(Pageable pageable);
}

I hurried to try my solution out. Perfect - the page is returned but when I tried to sort by bCount I got disappointed. It turned out that since this is a ARepository (not AwithBCount repository) spring-data will try to look for a bCount property in A instead of AwithBCount. So finally I ended up with three custom methods:

public interface ARepository extends JpaRepository<A, Long> {
    @Query(
        value = "select new package.AwithBCount(count(b.id) as bCount,c) from A a join a.bes b group by a",
        countQuery = "select count(a) from A a"
    )
    Page<AwithBCount> findAllWithBCount(Pageable pageable);

    @Query(
        value = "select new package.AwithBCount(count(b.id) as bCount,c) from A a join a.bes b group by a order by bCount asc",
        countQuery = "select count(a) from A a"
    )
    Page<AwithBCount> findAllWithBCountOrderByCountAsc(Pageable pageable);

    @Query(
        value = "select new package.AwithBCount(count(b.id) as bCount,c) from A a join a.bes b group by a order by bCount desc",
        countQuery = "select count(a) from A a"
    )
    Page<AwithBCount> findAllWithBCountOrderByCountDesc(Pageable pageable);
}

...and some additional conditional logic on service level (which could be probably encapsulated with an abstract repository implementation). So, although not extremely elegant, that made the trick - this way (having more complex entities) I can sort by other properties, do the filtering and pagination.

One option, which is much simpler than the original solution and which also has additional benefits, is to create a database view of aggregate data and link your Entity to this by means of a @SecondaryTable or @OneToOne.

For example:

create view a_summary_view as
select
   a_id as id, 
   count(*) as b_count, 
   sum(value) as b_total, 
   max(some_date) as last_b_date 
from b 

Using @SecondaryTable

@Entity
@Table
@SecondaryTable(name = "a_summary_view", 
       pkJoinColumns = {@PrimaryKeyJoinColumn(name = "id", referencedColumnName= "id")})
public class A{

   @Column(table = "a_summary_view")
   private Integer bCount;

   @Column(table = "a_summary_view")
   private BigDecimal bTotal;

   @Column(table = "a_summary_view")
   private Date lastBDate;
}

You can now then sort, filer, query etc purely with reference to entity A.

As an additional advantage you have within your domain model data that may be expensive to compute in-memory e.g. the total value of all orders for a customer without having to load all orders or revert to a separate query.