EJB3 bean 可以“自我注入”并通过 EJB 容器调用自己的方法吗？

Question

Is it possible to "self inject" an EJB in order to call local methods as bean methods? There are some cases where this could be favorable, for example if container managed transactions are used and something should be accomplished in a new transaction.

An example how this could work:

Foo.java:

@Local
public interface FoO {
    public void doSomething();
    public void processWithNewTransaction(); // this should actually be private
}

FooBean.java:

@Stateless
public class FooBean implements Foo {

    @EJB
    private Foo foo;

    public void doSomething() {
        ...
        foo.processWithNewTransaction();
        ...
    }

    @TransactionAttribute(TransactionAttributeType.REQUIRES_NEW)
    public void processWithNewTransaction() {
        ...
    }
}

If I extract processWithNewTransaction() to another bean, it would need to be exposed as a public method in the interface, even though it should be called only by FooBean . (The same issue is with my code above, that's why there is a comment in the interface definition.)

One solution would be to switch to bean managed transactions. However this would require changing the whole bean to manage its own transactions, and would add a lot of boiler plate to all methods.

Answer 1

Self injection of a EJB is indeed possible. The reason why an infinite recursion won't happen in this case is pretty simple: the container doesn't inject an actual bean instance from the pool. Instead, it injects a proxy object. When you call a method on the injected proxy (foo), the container gets a bean instance from its pool or creates one, if there are no available instances.

Answer 2

It is possible to do a self injection . You need to use SessionContext .

SessionContext sc = ...
sc.getBusinessObject(FooBean.class).processWithNewTransaction()

Answer 3

Update : As noted by other answers, this is indeed technically possible. Please see the answers of Csaba and Michael on how and why it works despite the apparend endless recursion.

---

I cannot give a 100% accurate answer but I'm pretty sure that this is not possible.

I think so because for injecting the Foo bean into the Foo bean itself the container would initially have to create a Foo instance which he can inject afterwards. But to create this he has to inject an already existing instance of Foo into the Foo to be created... which leads to an infinite recursion.

If you need separate transactions I'd suggest to keep things easy and create two independend beans/interfaces.

Answer 4

I'd tend to disagree, often it's useful for transaction management to call a local bean method via the container.

Exactly as the example if you had to call a local bean method inside a loop you'd be better having a transaction per iteration than for all iterations. (provided business logic wasn't "all or nothing" like despatching a delivery, or issuing stock)

Answer 5

Interesting question. I never create a method with a different transaction attribute in the same bean, that is, this begs for refactoring. This usually makes it hard to spot bugs in the application when it evolves.

EDIT: fixed typos