We Keep Coding

Implementing NULLS LAST with JPA CriteriaBuilder API

Spring/Hibernate/MySQL/JPA here. I have the following code:
public void setOrdering(
SearchRequest searchRequest,
CriteriaQuery query,
CriteriaBuilder builder,
Root<? extends MyEntity> root) {
String sortParam = "reportedOn";
Expression expression = builder.selectCase()
.when(builder.isNull(root.get(sortParam)), root.get(sortParam))
.otherwise(root.get(sortParam));
Order order = (searchRequest.isAscending())
? builder.asc(expression)
: builder.desc(expression);
query.orderBy(order);
}
Basically, I'm trying to implement the CriteriaBuilder/JPA equivalent of:
SELECT
*
FROM
mytable
WHERE
<lots of predicates here>
ORDER BY reported_on IS NULL, reported_on <ASC/DESC>
I already have the WHERE predicates added, I'm just struggling with the query.orderBy(...).
At runtime, when searchRequest.isAscending() is false, the results come back working just fine, with the records that contain a null reported_on value ordered at the end of the results.
But if searchRequest.isAscending() is true, the NULLS LAST attempt does not appear to work at all.

You're mixing up the Spring and JPA APIs, here query is from the JPA API so you need to sort using something like:
CriteriaBuilder cb = ...
Root root = ...
query.orderBy(cb.asc(root.get("reportedOn")));

It does not look like JPA's CriteriaBuilder supports NULLS LAST. I actually got this working using a SQL "hack":
String sortParam = "reportedOn";
Order order = (searchRequest.isAscending())
? builder.desc(builder.neg(root.get(sortParam)))
: builder.desc(root.get(sortParam));
query.orderBy(order);
Basically ORDER BY -reported_on DESC does the same thing as ORDER BY reported_on ASC but it sorts records with NULL reported_on values all the way to the bottom of the search results, which is what NULLS LAST is supposed to do.

SELECT DISTINCT + ORDER BY in JPA 2 Criteria API

I've a class Lawsuit, that contains a List<Hearing>, each one with a Date attribute.
I need to select all the Lawsuits ordered by the date of their Hearings
I've a CriteriaQuery like
CriteriaBuilder cb = em.getCriteriaBuilder();
CriteriaQuery<Lawsuit> cq = cb.createQuery(Lawsuit.class);
Root<Lawsuit> root = cq.from(Lawsuit.class);
I use distinct to flatten the results:
cq.select(root).distinct(true);
I then join Lawsuit with Hearing
Join<Lawsuit, Hearing> hearing = root.join("hearings", JoinType.INNER);
to create Predicates
predicateList.add(cb.isNotNull(hearing.<Date>get("date")));
and Orders:
orderList.add(cb.asc(hearing.<Date>get("date")));
Everything works fine if I avoid distinct, but if I use it, it complains about not being able to order based on fields that are not in the SELECT:
Caused by: org.postgresql.util.PSQLException: ERROR: for SELECT DISTINCT, ORDER BY expressions must appear in select list
The List<Hearing> is already accessible through the Lawsuit classes returned, so I'm confused: how should I add them to the select list ?

I've discovered the source of the problem somewhere else, and solving it has made unnecessary to do what asked in the question;
as described in other answers, it should be unnecessary to perform the distinct here.
The duplicate rows were originated by erroneous left joins that were performed on collections (attributes of the root object) even if the predicates were not been used:
Join<Lawsuit, Witness> witnesses = root.join("witnesses", JoinType.LEFT);
if (witnessToFilterWith!=null) {
predicateList.add(cb.equal(witnesses.<Long>get("id"),witnessToFilterWith.getId()));
}
The join should obviously be performed as inner and only if needed:
if (witnessToFilterWith!=null) {
Join<Lawsuit, Witness> witnesses = root.join("witnesses", JoinType.INNER);
predicateList.add(cb.equal(witnesses.<Long>get("id"),witnessToFilterWith.getId()));
}
So, if you're here because you're getting the same problem, search the problem in the joins.

You can also de-duplicate via group by based on primary key column of root table:
cq.groupBy(root.get("id")); // Assuming that Lawsuite.id is primary key column

Join cartesian product

Context
I send some files to some enterprises every week. I need to restitute for each week and each enterprise whether the file is sent or not.
Tables
ent (enterprise)
wek (week)
fil (file : references wek and ent)
Solution with pure SQL
Make a cartesian product between ent and wek then left outer join fil. This works :
select * from
(
select * from wek, ent e
) as t1
left join fil f
on f.ent_id = t1.ent_id
and f.wek_id = t1.wek_id
Problem :
How to translate this into JPA (in the CriteriaBuilder way)?
For example, if I try :
CriteriaBuilder cb = em.getCriteriaBuilder();
CriteriaQuery<ResultClass> query = cb.createQuery(ResultClass.class);
Root<Week> week = query.from(Week.class);
Root<Enterprise> query.from(Enterprise.class);
Expression<???> cartesianProduct = ??? //How?
cartesianProduct.leftJoin(???_.file);
query.where(
cb.equal(file.get(File_.wek_id), week.get(Week.week_id));
)
Using 2 "from" clauses gives me the cartesian product but how do I left join this result?
Unstatisfaying solution :
Create a view :
CREATE VIEW view_ent_wek AS
SELECT ent_id as ent_id, wek_id as wek_id, ent_id || '-' || ent_id as id
FROM ent, wek;
Map it to an entity :
#Entity
#Table(name = "view_ent_wek")
public class WeekEnterprise {
#Id
#Column(name = "id")
private String id;
#ManyToOne
#JoinColumn(name = "ent_id")
private Enterprise enterprise;
#ManyToOne
#JoinColumn(name = "wek_id")
private Week week;
[...]
Then I can use it in a query :
Root<WeekEnterprise> weekEnterprise = query.from(WeekEnterprise.class);
weekEnterprise.join(...)
I don't like this solution because it makes me create a view that is obviously not necessary. Any idea?

I don't see the point of doing the reporting-style SQL query with JPA, let alone with Criteria Query.
Of course, JPA (and its implementations) do support tuple queries (as opposed to entity queries, see also: Hibernate tuple criteria queries), but tuple queries is not what JPA is good at. Imagine you want to change that report and add more calculations. Wouldn't it be much easier to do that with SQL?
Using JPA for mapping to Java objects and for transaction modelling still makes sense, and thus, your SQL view solution is already quite good, or if you prefer not to store the SQL in a view, use JPA's native query API:
List<Object[]> list = em.createNativeQuery(sql).getResultList();
Or:
List<WeekEnterprise> list = em.createNativeQuery(sql, WeekEnterprise.class)
.getResultList();

I have some ideas:
With 2 JPA queries. First: fetch the cartesian product of wek and enterprises. Second: take an inner join between wek, enterprises and files. Uses a map (wek_id+ent_id => Tuple(Wek, Ent, File)) to quickly identify where to put the file.
Write plain SQL queries and execute them with the JPA API.
(Didn't think much about this one) Create a back reference from Wek and Ent to File (lazy loaded) and then you should be able to continue your first idea.

Don't know much about JPA, but about Postgres. And I would suggest this superior query:
SELECT *
FROM wek CROSS JOIN ent
LEFT JOIN fil USING (ent_id, wek_id);
You don't need the subselect.
The only difference in the result (besides being shorter and faster): You get the columns ent_id and wek_id once in the output (which might solve a problem with duplicated column names).
Should be easy to translate to JPA now.

Need help creating JPA criteria query

I'm building my first Java EE web application using Glassfish and JSF. I'm fairly new to the criteria query and I have a query I need to perform but the javaee6 tutorial seems a little thin on examples. Anyway, I'm having a hard time creating the query.
Goal: I want to pull the company with the most documents stored.
Companies have a OneToMany relationship with Documents.
Documents has a ManyToOne relationship with several tables, the "usertype" column distinguishes them.
MySQL query:
SELECT USERID, COUNT(USERID) AS CNT
FROM DOCUMENTS
WHERE USERTYPE="COMPANY"
GROUP BY USERID
ORDER BY CNT DESC
Thanks
--update--
Based on user feedback, here is what I have so far:
CriteriaBuilder cb = em.getCriteriaBuilder();
CriteriaQuery<Documents> cqry = cb.createQuery(Documents.class);
//Intersting Stuff
Root<Documents> root = cqry.from(Documents.class);
Expression userid = root.get("userID");
Expression usertype = root.get("userType");
Expression count = cb.count(userid);
cqry.multiselect(userid, count);
Predicate userType = cb.equal(usertype, "COMPANY");
cqry.where(userType);
cqry.groupBy(userid);
cqry.orderBy(cb.desc(count));
//more boilerplate
Query qry = em.createQuery(cqry);
List<Documents> results = qry.getResultList();
The error I get is:
Exception Description: Partial object queries are not allowed to maintain the cache or be edited. You must use dontMaintainCache().
Typical error, means nothing to me!

Your query doesn't return a complete entity object as you're only selecting two fields of the given table (this is why you're getting an error that says yadayadapartialyadayada).
Your solution is almost right, here's what you need to change to make it work—making it partial.
Instead of a plain CriteriaQuery<...> you have to create a tuple CriteriaQuery<..> by calling CriteriaBuilder.createTupleQuery(). (Basically, you can call CriteriaBuilder.createQuery(...) and pass Tuple.class to it as an argument. Tuple is a sort of wildcard entity class.)
CriteriaBuilder cb = em.getCriteriaBuilder();
CriteriaQuery<Tuple> cq= cb.createTupleQuery();
Root<Documents> root = cq.from(Documents.class);
Expression<Integer> userId = root.get("USERID");
Expression<String> userType = root.get("USERTYPE");
Expression<Long> count = cb.count(userId);
cq.multiselect(userId.alias("USERID"), count.alias("CNT"));
cq.where(cb.equal(userType, "COMPANY");
cq.groupBy(userId);
cq.orderBy(cb.desc(count));
TypedQuery<Tuple> tq = em.createQuery(cq);
for (Tuple t : tq.getResultsList()) {
System.out.println(t.get("USERID"));
System.out.println(t.get("CNT"));
}
(Accessing fields of a Tuple gave me an error if I didn't use aliases for them (in multiselect(...)). This is why I've used aliases, but this can be tackled more cleanly by using JPA 2's Metamodel API, which is described in the specification quite thoroughly. )
The documentation for CriteriaQuery.multiselect(...) describes the behaviour of queries using Tuple objects more deeply.

If you are using Hibernate, this should work:
ProjectionList pl = Projections.projectionList()
.add(Projections.groupProperty("userid"))
.add(Projections.property("userid"))
.add(Projections.count("userid"));
Criteria criteria = session.createCriteria(Document.class)
.add(Restrictions.eq("usertype",usertype))
.setProjection(pl)
.addOrder(Order.desc("cnt"));
Hope it helps!

Take a look into this easy tutorial. It uses JPA2 and Criteria
http://www.jumpingbean.co.za/blogs/jpa2-criteria-api
Regards!

You need to add a constructor to Documents with only userid and count because you will need it on:
cqry.multiselect(userid, count);

Referring to an earlier aliased field in a criteria query

In this query:
CriteriaBuilder cb = em.getCriteriaBuilder();
CriteriaQuery<Tuple> q = cb.createTupleQuery();
// FROM GamePlayedEvent gpe
Root<GamePlayedEvent> gpe = q.from(GamePlayedEvent.class);
// SELECT gameId, COUNT(*) AS count, AVG(duration)
// AS avDur, AVG(rewardCurrency) AS avCur, AVG(rewardXP) avXp
q.select(cb.tuple(
gpe.<String>get("gameId"),
cb.count(gpe).alias("count"),
cb.avg(gpe.<Double>get("duration")).alias("avDur"),
cb.avg(gpe.<Integer>get("rewardCurrency")).alias("avCur"),
cb.avg(gpe.<Integer>get("rewardXp")).alias("avXp")
));
// WHERE loginTime BETWEEN ...
q.where(cb.between(gpe.<Date>get("time"), fromTime, toTime));
// GROUP BY gameId
q.groupBy(gpe.<String>get("gameId"));
// ORDER BY count DESC
q.orderBy(cb.desc(???));
How can I add the ORDER BY count DESC, referring to the "count" defined in the SELECT clause?

What if you just captured the count expression, and used it directly?
Expression event_count = cb.count(gpe);
q.select(cb.tuple(
gpe.<String>get("gameId"),
event_count,
...
));
q.orderBy(cb.desc(event_count));

I came across the same problem today but none of the suggested solutions worked for me because I needed to reuse the expression not only in the order by clause but also in the group by clause.
One obvious solution would be to create a view on the database level but this is a bit clumsy, creates an unnecessary subquery and even not possible if the db user isn't granted enough privileges. A better option which I ended up implementing is to write something like this
q.select(cb.tuple(
gpe.<String>get("gameId"),
cb.count(gpe),
...
)).groupBy(cb.literal(2)).orderBy(cb.literal(2));
The first downside of this approach is that the code is errorprone. The other drawback is that the generated sql query contains ordinal position notation, which works on some databases (like postgresql or mariadb) but doesn't work on others (like sql server). In my case, however, I found this to be the best option.
Tested on jpa 2.1 with hibernate 5.2.3 as a provider and postgresql 9.6.

Even though the Pro JPA 2 book describes that the alias method can be used to generate a sql query alias (on page 251) I have had no success with making it work with neither EclipseLink or Hibernate. For your question I would say that your orderBy line should read:
q.orderBy(cb.desc(cb.count(gpe));
if it was supported by the different vendors.
As far as my research goes it seams that the alias method is only used for naming elements in the tuble used in the select (so only for projection).
I have one question though. Why would you want to use the JPA Criteria API for this query. It (the query) seams to be static in nature so why not use JPQL where you can define your query aliases directly.

Have you tried setting up a projection with an alias?
criteria.setProjection(Projections.projectionList()
.add(Projections.count("item.id"), "countItems"));
criteria.addOrder(Order.desc("countItems"));

For a sum aggregation field I have the following code which worked for me:
CriteriaBuilder cb = em.getCriteriaBuilder();
CriteriaQuery<T> cq = cb.createQuery(entity);
Root<T> root = cq.from(entity);
cq.orderBy(cb.desc(cb.sum(root.get(orderByString))));
// orderByString is string entity field that is being aggregated and which we want to put in orderby clause as well.