Chapter 8. Seam and Object/Relational Mapping - JBoss Seam 1.1.6 正式版英文参考手册

Seam grew out of the frustration of the Hibernate team with the statelessness typical of the previous generation of Java application architectures. The state management architecture of Seam was originally designed to solve problems relating to persistence—in particular problems associated with optimistic transaction processing. Scalable online applications always use optimistic transactions. An atomic (database/JTA) level transaction should not span a user interaction unless the application is designed to support only a very small number of concurrent clients. But almost all interesting work involves first displaying data to a user, and then, slightly later, updating the same data. So Hibernate was designed to support the idea of a persistence context which spanned an optimistic transaction.

Unfortunately, the so-called "stateless" architectures that preceded Seam and EJB 3.0 had no construct for representing an optimistic transaction. So, instead, these architectures provided persistence contexts scoped to the atomic transaction. Of course, this resulted in many problems for users, and is the cause of the number one user complaint about Hibernate: the dreaded LazyInitializationException. What we need is a construct for representing an optimistic transaction in the application tier.

EJB 3.0 recognizes this problem, and introduces the idea of a stateful component (a stateful session bean) with an extended persistence context scoped to the lifetime of the component. This is a partial solution to the problem (and is a useful construct in and of itself) however there are two problems:

Seam solves the first problem by providing conversations, and stateful session bean components scoped to the conversation. (Most conversations actually represent optimistic transactions in the data layer.) This is sufficient for many simple applications (such as the Seam booking demo) where persistence context propagation is not needed. For more complex applications, with many loosly-interacting components in each conversation, propagation of the persistence context across components becomes an important issue. So Seam extends the persistence context management model of EJB 3.0, to provide conversation-scoped extended persistence contexts.

EJB session beans feature declarative transaction management. The EJB container is able to start a transaction transparently when the bean is invoked, and end it when the invocation ends. If we write a session bean method that acts as a JSF action listener, we can do all the work associated with that action in one transaction, and be sure that it is committed or rolled back when we finish processing the action. This is a great feature, and all that is needed by some Seam applications.

However, there is a problem with this approach. A Seam application may not perform all data access for a request from a single method call to a session bean.

The more transactions per request, the more likely we are to encounter atomicity and isolation problems when our application is processing many concurrent requests. Certainly, all write operations should occur in the same transaction!

Hibernate users developed the "open session in view" pattern to work around this problem. In the Hibernate community, "open session in view" was historically even more important because frameworks like Spring use transaction-scoped persistence contexts. So rendering the view would cause LazyInitializationExceptions when unfetched associations were accessed.

This pattern is usually implemented as a single transaction which spans the entire request. There are several problems with this implementation, the most serious being that we can never be sure that a transaction is successful until we commit it—but by the time the "open session in view" transaction is committed, the view is fully rendered, and the rendered response may already have been flushed to the client. How can we notify the user that their transaction was unsuccessful?

Seam solves both the transaction isolation problem and the association fetching problem, while working around the problems with "open session in view". The solution comes in two parts:

In the next section, we'll tell you how to set up a conversation-scope persistence context. But first we need to tell you how to enable Seam transaction management. Note that you can use conversation-scoped persistence contexts without Seam transaction management, and there are good reasons to use Seam transaction management even when you're not using Seam-managed persistence contexts. However, the two facilities were designed to work together, and work best when used together.

<lifecycle>
     <phase-listener>
        org.jboss.seam.jsf.TransactionalSeamPhaseListener
    </phase-listener>
</lifecycle>

If you're using Seam outside of a Java EE 5 environment, you can't rely upon the container to manage the persistence context lifecycle for you. Even if you are in an EE 5 environment, you might have a complex application with many loosly coupled components that collaborate together in the scope of a single conversation, and in this case you might find that propagation of the persistence context between component is tricky and error-prone.

In either case, you'll need to use a managed persistence context (for JPA) or a managed session (for Hibernate) in your components. A Seam-managed persistence context is just a built-in Seam component that manages an instance of EntityManager or Session in the conversation context. You can inject it with @In.

Seam-managed persistence contexts are extremely efficient in a clustered environment. Seam is able to perform an optimization that EJB 3.0 specification does not allow containers to use for container-managed extended persistence contexts. Seam supports transparent failover of extended persisence contexts, without the need to replicate any persistence context state between nodes. (We hope to fix this oversight in the next revision of the EJB spec.)

<core:managed-persistence-context name="bookingDatabase" 
                           auto-create="true"
            persistence-unit-jndi-name="java:/EntityManagerFactories/bookingData"/>

<property name="jboss.entity.manager.factory.jndi.name" 
          value="java:/EntityManagerFactories/bookingData"/>

@In EntityManager bookingDatabase;

<core:hibernate-session-factory name="hibernateSessionFactory"/>

<core:managed-hibernate-session name="bookingDatabase" 
                         auto-create="true"
           session-factory-jndi-name="java:/bookingSessionFactory"/>

<session-factory name="java:/bookingSessionFactory">
    <property name="transaction.flush_before_completion">true</property>
    <property name="connection.release_mode">after_statement</property>
    <property name="transaction.manager_lookup_class">org.hibernate.transaction.JBossTransactionManagerLookup</property>
    <property name="transaction.factory_class">org.hibernate.transaction.JTATransactionFactory</property>
    <property name="connection.datasource">java:/bookingDatasource</property>
    ...
</session-factory>

@In Session bookingDatabase;

@In EntityManager em; //a Seam-managed persistence context

@Begin(flushMode=MANUAL)
public void beginClaimWizard() {
    claim = em.find(Claim.class, claimId);
}

public void addPartyToClaim() {
    Party party = ....;
    claim.addParty(party);
}

@End
public void commitClaim() {
    em.flush();
}

The EntityManager interface lets you access a vendor-specific API via the getDelegate() method. Naturally, the most interesting vendor is Hibernate, and the most powerful delegate interface is org.hibernate.Session. You'd be nuts to use anything else. Trust me, I'm not biased at all.

But regardless of whether you're using Hibernate (genius!) or something else (masochist, or just not very bright), you'll almost certainly want to use the delegate in your Seam components from time to time. One approach would be the following:

@In EntityManager entityManager;

@Create
public void init() {
    ( (Session) entityManager.getDelegate() ).enableFilter("currentVersions");
}

But typecasts are unquestionably the ugliest syntax in the Java language, so most people avoid them whenever possible. Here's a different way to get at the delegate. First, add the following line to components.xml:

<factory name="session" 
         scope="STATELESS" 
         auto-create="true" 
         value="#{entityManager.delegate}"/>

Now we can inject the session directly:

@In Session session;

@Create
public void init() {
    session.enableFilter("currentVersions");
}

The coolest, and most unique, feature of Hibernate is filters. Filters let you provide a restricted view of the data in the database. You can find out more about filters in the Hibernate documentation. But we thought we'd mention an easy way to incorporate filters into a Seam application, one that works especially well with the Seam Application Framework.

Seam-managed persistence contexts may have a list of filters defined, which will be enabled whenever an EntityManager or Hibernate Session is first created. (Of course, they may only be used when Hibernate is the underlying persistence provider.)

<core:filter name="regionFilter">
    <core:name>region</core:name>
    <core:parameters>
        <key>regionCode</key>
        <value>#{region.code}</value>
    </core:parameters>
</core:filter>

<core:filter name="currentFilter">
    <core:name>current</core:name>
    <core:parameters>
        <key>date</key>
        <value>#{currentDate}</value>
    </core:parameters>
</core:filter>

<core:managed-persistence-context name="personDatabase"
    persistence-unit-jndi-name="java:/EntityManagerFactories/personDatabase">
    <core:filters>
        <value>#{regionFilter}</value>
        <value>#{currentFilter}</value>
    </core:filters>
</core:managed-persistence-context>