@Entity
public class A {
@GeneratedValue
@Id
private long id;
public long getId() {
return id;
}
public void setId(final long id) {
this.id = id;
}
@OneToMany(mappedBy = "a")
List<B> bs;
public List<B> getBs() {
return bs;
}
public void setBs(final List<B> bs) {
this.bs = bs;
}
}
@Entity
public class B {
@GeneratedValue
@Id
private long id;
public long getId() {
return id;
}
public void setId(final long id) {
this.id = id;
}
@ManyToOne
@JoinTable
A a;
public A getA() {
return a;
}
public void setA(final A a) {
this.a = a;
}
}
To establish the relationship, I have to call
b.setA(a);
a.getBs().add(b);
Why is both necessary, Why is it not sufficient to do only
b.setA(a);
or
a.getBs().add(b);
?
The relationship is stored in a join table, and b.setA(a)
will update that join table.
But when I do a query afterwards, a.getBs()
is empty. Why is that?
Here is a Test case that illustrates the question. Note that the very last assert fails.
public class QuickTestAB2 {
private static String dbUrlBase = "jdbc:derby:testData/db/test.db";
private static String dbUrlCreate = dbUrlBase + ";create=true";
private static String dbUrlDrop = dbUrlBase + ";drop=true";
private EntityManagerFactory factory;
private EntityManager em;
public Map<String, String> createPersistenceMap(final String dbUrl) {
final Map<String, String> persistenceMap = new HashMap<>();
persistenceMap.put("javax.persistence.jdbc.url", dbUrl);
return persistenceMap;
}
public void dropDatabase() throws Exception {
if (em != null && em.isOpen()) {
em.close();
}
if (factory != null && factory.isOpen()) {
factory.close();
}
try (Connection conn = DriverManager.getConnection(dbUrlDrop)) {
} catch (final SQLException e) {
// always
}
}
public void deleteDatabase() throws Exception {
dropDatabase();
final File file = new File("testData/db/test.db");
if (file.exists()) {
FileUtils.forceDelete(file);
}
}
public void createNewDatabase() throws SQLException, IOException {
FileUtils.forceMkdir(new File("testData/db"));
try (Connection conn = DriverManager.getConnection(dbUrlCreate)) {
}
}
@BeforeClass
public static void setUpBeforeClass01() throws Exception {
Tests.enableLog4J();
JPATests.enableJPA();
}
@AfterClass
public static void tearDownAfterClass01() throws Exception {
}
@Before
public void setUp01() throws Exception {
deleteDatabase();
createNewDatabase();
final Map<String, String> map = createPersistenceMap(dbUrlCreate);
factory = Persistence.createEntityManagerFactory("pu", map);
}
@After
public void tearDown01() throws Exception {
if (em != null && em.isOpen()) {
em.close();
}
em = null;
if (factory != null && factory.isOpen()) {
factory.close();
}
factory = null;
}
@Test
public void test01() throws Exception {
em = factory.createEntityManager();
final A a = new A();
final B b = new B();
b.setA(a);
try {
em.getTransaction().begin();
em.persist(a);
em.persist(b);
em.getTransaction().commit();
} finally {
em.close();
}
em = factory.createEntityManager();
B b2;
A a2;
try {
em.getTransaction().begin();
Query q = em.createQuery("SELECT b FROM B b");
b2 = (B) q.getSingleResult();
q = em.createQuery("SELECT a FROM A a");
a2 = (A) q.getSingleResult();
em.getTransaction().commit();
} finally {
em.close();
}
assertThat(a2, is(not(nullValue())));
assertThat(b2, is(not(nullValue())));
assertThat(b2.getA(), is(not(nullValue())));
assertThat(a2.getBs().isEmpty(), is(false));
}
}
Motivation: It can be useful to change a bidirectional relationship by changing "only one side", when the number of a.Bs
gets large. In this case, an UPDATE SELECT
query on the owning side is much faster than to call a.getBs().remove(b)
See also here.
As stated in Nikos Paraskevopoulos' answer below, JPA and java objects require you to set both sides of a relationship. As long as you set the owning side, the database will be updated with the relationship changes, but the non-owning side will only reflect what is in the database if you manually set it, or you force it to be refreshed or reloaded. Reading the entity from a separate context does not force reloading, as your JPA provider can use a second level cache; this is the default in EclipseLink. Your other read is returning the A from the shared cache, which like your original object, does not have B added to its list of Bs.
The easiest solution is to just set B into A's list upfront. Other options here though would be to force the refresh of A using em.refresh(a) or with a query hint, or disable the shared cached.
There are 2 "facets" in the question: The Java side and the JPA side.
Java side
A more complete listing of the code might be:
The JPA entites are still Java objects. If you do instruct Java explicitly to, e.g. "add the B in the collection of Bs, when its
a
property is set" it has no reason to do it automatically. Having said that, I have often seen patterns like (skipping null checking for brevity):JPA side
JPA 2.1. specs, ch. 2.9 "Entity Relationships":
In the setup of the question,
B.a
is the owning side becauseA.bs
specifiesmappedBy="a"
. The specifications says that the relation will be updated (i.e. an entry in the join table will be inserted) only when the owning side is updated. That is why doingb.setA(a)
updates the join table.After doing the above and successfully updating the DB, reading the related A object fresh from the DB should fetch the correct
bs
collection. To be sure, first try merging the B, committing the transaction, and fetching A (or refreshing it) in a different transaction. If you want the state of the Java objects to be reflected immediately in the same transaction, you have no other option but to set bothb.a
anda.getBs().add(b)
.