To invoke a method on a local or remote stateless session bean, client code must normally perform a JNDI lookup to obtain the (local or remote) EJB Home object, then use a 'create' method call on that object to obtain the actual (local or remote) EJB object. One or more methods are then invoked on the EJB.

To avoid repeated low-level code, many EJB applications use the Service Locator and Business Delegate patterns. These are better than spraying JNDI lookups throughout client code, but their usual implementations have significant disadvantages. For example:

The Spring approach is to allow the creation and use of proxy objects, normally configured inside a Spring container, which act as codeless business delegates. You do not need to write another Service Locator, another JNDI lookup, or duplicate methods in a hand-coded Business Delegate unless you are actually adding real value in such code.

Assume that we have a web controller that needs to use a local EJB. We’ll follow best practice and use the EJB Business Methods Interface pattern, so that the EJB’s local interface extends a non EJB-specific business methods interface. Let’s call this business methods interface MyComponent.

public interface MyComponent {
    ...
}

One of the main reasons to use the Business Methods Interface pattern is to ensure that synchronization between method signatures in local interface and bean implementation class is automatic. Another reason is that it later makes it much easier for us to switch to a POJO (plain old Java object) implementation of the service if it makes sense to do so. Of course we’ll also need to implement the local home interface and provide an implementation class that implements SessionBean and the MyComponent business methods interface. Now the only Java coding we’ll need to do to hook up our web tier controller to the EJB implementation is to expose a setter method of type MyComponent on the controller. This will save the reference as an instance variable in the controller:

private MyComponent myComponent;

public void setMyComponent(MyComponent myComponent) {
    this.myComponent = myComponent;
}

We can subsequently use this instance variable in any business method in the controller. Now assuming we are obtaining our controller object out of a Spring container, we can (in the same context) configure a LocalStatelessSessionProxyFactoryBean instance, which will be the EJB proxy object. The configuration of the proxy, and setting of the myComponent property of the controller is done with a configuration entry such as:

<bean id="myComponent"
      class="org.springframework.ejb.access.LocalStatelessSessionProxyFactoryBean">
  <property name="jndiName" value="ejb/myBean"/>
  <property name="businessInterface" value="com.mycom.MyComponent"/>
</bean>

<bean id="myController" class="com.mycom.myController">
  <property name="myComponent" ref="myComponent"/>
</bean>

There’s a lot of work happening behind the scenes, courtesy of the Spring AOP framework, although you aren’t forced to work with AOP concepts to enjoy the results. The myComponent bean definition creates a proxy for the EJB, which implements the business method interface. The EJB local home is cached on startup, so there’s only a single JNDI lookup. Each time the EJB is invoked, the proxy invokes the classname method on the local EJB and invokes the corresponding business method on the EJB.

The myController bean definition sets the myComponent property of the controller class to the EJB proxy.

Alternatively (and preferably in case of many such proxy definitions), consider using the <jee:local-slsb> configuration element in Spring's "jee" namespace:

<jee:local-slsb id="myComponent" jndi-name="ejb/myBean"
      business-interface="com.mycom.MyComponent"/>

<bean id="myController" class="com.mycom.myController">
  <property name="myComponent" ref="myComponent"/>
</bean>

This EJB access mechanism delivers huge simplification of application code: the web tier code (or other EJB client code) has no dependence on the use of EJB. If we want to replace this EJB reference with a POJO or a mock object or other test stub, we could simply change the myComponent bean definition without changing a line of Java code. Additionally, we haven’t had to write a single line of JNDI lookup or other EJB plumbing code as part of our application.

Benchmarks and experience in real applications indicate that the performance overhead of this approach (which involves reflective invocation of the target EJB) is minimal, and is typically undetectable in typical use. Remember that we don’t want to make fine-grained calls to EJBs anyway, as there’s a cost associated with the EJB infrastructure in the application server.

There is one caveat with regards to the JNDI lookup. In a bean container, this class is normally best used as a singleton (there simply is no reason to make it a prototype). However, if that bean container pre-instantiates singletons (as do the various XML ApplicationContext variants) you may have a problem if the bean container is loaded before the EJB container loads the target EJB. That is because the JNDI lookup will be performed in the init() method of this class and then cached, but the EJB will not have been bound at the target location yet. The solution is to not pre-instantiate this factory object, but allow it to be created on first use. In the XML containers, this is controlled via the lazy-init attribute.

Although this will not be of interest to the majority of Spring users, those doing programmatic AOP work with EJBs may want to look at LocalSlsbInvokerInterceptor.

Accessing remote EJBs is essentially identical to accessing local EJBs, except that the SimpleRemoteStatelessSessionProxyFactoryBean or <jee:remote-slsb> configuration element is used. Of course, with or without Spring, remote invocation semantics apply; a call to a method on an object in another VM in another computer does sometimes have to be treated differently in terms of usage scenarios and failure handling.

Spring's EJB client support adds one more advantage over the non-Spring approach. Normally it is problematic for EJB client code to be easily switched back and forth between calling EJBs locally or remotely. This is because the remote interface methods must declare that they throw RemoteException, and client code must deal with this, while the local interface methods don't. Client code written for local EJBs which needs to be moved to remote EJBs typically has to be modified to add handling for the remote exceptions, and client code written for remote EJBs which needs to be moved to local EJBs, can either stay the same but do a lot of unnecessary handling of remote exceptions, or needs to be modified to remove that code. With the Spring remote EJB proxy, you can instead not declare any thrown RemoteException in your Business Method Interface and implementing EJB code, have a remote interface which is identical except that it does throw RemoteException, and rely on the proxy to dynamically treat the two interfaces as if they were the same. That is, client code does not have to deal with the checked RemoteException class. Any actual RemoteException that is thrown during the EJB invocation will be re-thrown as the non-checked RemoteAccessException class, which is a subclass of RuntimeException. The target service can then be switched at will between a local EJB or remote EJB (or even plain Java object) implementation, without the client code knowing or caring. Of course, this is optional; there is nothing stopping you from declaring RemoteExceptions in your business interface.

Accessing EJB 2.x Session Beans and EJB 3 Session Beans via Spring is largely transparent. Spring's EJB accessors, including the <jee:local-slsb> and <jee:remote-slsb> facilities, transparently adapt to the actual component at runtime. They handle a home interface if found (EJB 2.x style), or perform straight component invocations if no home interface is available (EJB 3 style).

Note: For EJB 3 Session Beans, you could effectively use a JndiObjectFactoryBean / <jee:jndi-lookup> as well, since fully usable component references are exposed for plain JNDI lookups there. Defining explicit <jee:local-slsb> / <jee:remote-slsb> lookups simply provides consistent and more explicit EJB access configuration.

Spring provides convenience classes to help you implement EJBs. These are designed to encourage the good practice of putting business logic behind EJBs in POJOs, leaving EJBs responsible for transaction demarcation and (optionally) remoting.

To implement a Stateless or Stateful session bean, or a Message Driven bean, you need only derive your implementation class from AbstractStatelessSessionBean, AbstractStatefulSessionBean, and AbstractMessageDrivenBean/AbstractJmsMessageDrivenBean, respectively.

Consider an example Stateless Session bean which actually delegates the implementation to a plain java service object. We have the business interface:

public interface MyComponent {
    public void myMethod(...);
    ...
}

We also have the plain Java implementation object:

public class MyComponentImpl implements MyComponent {
    public String myMethod(...) {
        ...
    }
    ...
}

And finally the Stateless Session Bean itself:

public class MyFacadeEJB extends AbstractStatelessSessionBean
        implements MyFacadeLocal {

    private MyComponent myComp;

    /**
     * Obtain our POJO service object from the BeanFactory/ApplicationContext
     * @see org.springframework.ejb.support.AbstractStatelessSessionBean#onEjbCreate()
     */
    protected void onEjbCreate() throws CreateException {
        myComp = (MyComponent) getBeanFactory().getBean(
            ServicesConstants.CONTEXT_MYCOMP_ID);
    }

    // for business method, delegate to POJO service impl.
    public String myFacadeMethod(...) {
        return myComp.myMethod(...);
    }
    ...
}

The Spring EJB support base classes will by default create and load a Spring IoC container as part of their lifecycle, which is then available to the EJB (for example, as used in the code above to obtain the POJO service object). The loading is done via a strategy object which is a subclass of BeanFactoryLocator. The actual implementation of BeanFactoryLocator used by default is ContextJndiBeanFactoryLocator, which creates the ApplicationContext from a resource locations specified as a JNDI environment variable (in the case of the EJB classes, at java:comp/env/ejb/BeanFactoryPath). If there is a need to change the BeanFactory/ApplicationContext loading strategy, the default BeanFactoryLocator implementation used may be overridden by calling the setBeanFactoryLocator() method, either in setSessionContext(), or in the actual constructor of the EJB. Please see the Javadocs for more details.

As described in the Javadocs, Stateful Session beans expecting to be passivated and reactivated as part of their lifecycle, and which use a non-serializable container instance (which is the normal case) will have to manually call unloadBeanFactory() and loadBeanFactory from ejbPassivate and ejbActivate, respectively, to unload and reload the BeanFactory on passivation and activation, since it can not be saved by the EJB container.

The default behavior of the ContextJndiBeanFactoryLocator classes which is to load an ApplicationContext for the use of the EJB is adequate for some situations. However, it is problematic when the ApplicationContext is loading a number of beans, or the initialization of those beans is time consuming or memory intensive (such as a Hibernate SessionFactory initialization, for example), since every EJB will have their own copy. In this case, the user may want to override the default ContextJndiBeanFactoryLocator usage and use another BeanFactoryLocator variant, such as the ContextSingletonBeanFactoryLocator which can load and use a shared container to be used by multiple EJBs or other clients. Doing this is relatively simple, by adding code similar to this to the EJB:

   /**
    * Override default BeanFactoryLocator implementation
    * @see javax.ejb.SessionBean#setSessionContext(javax.ejb.SessionContext)
    */
   public void setSessionContext(SessionContext sessionContext) {
       super.setSessionContext(sessionContext);
       setBeanFactoryLocator(ContextSingletonBeanFactoryLocator.getInstance());
       setBeanFactoryLocatorKey(ServicesConstants.PRIMARY_CONTEXT_ID);
   }

You would then need to create a bean definition file named beanRefContext.xml. This file defines all bean factories (usually in the form of application contexts) that may be used in the EJB. In many cases, this file will only contain a single bean definition such as this (where businessApplicationContext.xml contains the bean definitions for all business service POJOs):

<beans>
    <bean id="businessBeanFactory" class="org.springframework.context.support.ClassPathXmlApplicationContext">
        <constructor-arg value="businessApplicationContext.xml" />
    </bean>
</beans>

In the above example, the ServicesConstants.PRIMARY_CONTEXT_ID constant would be defined as follows:

public static final String ServicesConstants.PRIMARY_CONTEXT_ID = "businessBeanFactory";

Please see the respective Javadocs for the BeanFactoryLocator and ContextSingletonBeanFactoryLocator classes for more information on their usage.

For EJB 3 Session Beans and Message-Driven Beans, Spring provides a convenient interceptor that resolves Spring 2.5's @Autowired annotation in the EJB component class: org.springframework.ejb.interceptor.SpringBeanAutowiringInterceptor. This interceptor can be applied through an @Interceptors annotation in the EJB component class, or through an interceptor-binding XML element in the EJB deployment descriptor.

@Stateless
@Interceptors(SpringBeanAutowiringInterceptor.class)
public class MyFacadeEJB implements MyFacadeLocal {

    // automatically injected with a matching Spring bean
    @Autowired
    private MyComponent myComp;

    // for business method, delegate to POJO service impl.
    public String myFacadeMethod(...) {
        return myComp.myMethod(...);
    }
    ...
}

SpringBeanAutowiringInterceptor by default obtains target beans from a ContextSingletonBeanFactoryLocator, with the context defined in a bean definition file named beanRefContext.xml. By default, a single context definition is expected, which is obtained by type rather than by name. However, if you need to choose between multiple context definitions, a specific locator key is required. The locator key (i.e. the name of the context definition in beanRefContext.xml) can be explicitly specified either through overriding the getBeanFactoryLocatorKey method in a custom SpringBeanAutowiringInterceptor subclass.

Alternatively, consider overriding SpringBeanAutowiringInterceptor's getBeanFactory method, e.g. obtaining a shared ApplicationContext from a custom holder class.