Object Relational Mapping (ORM) Data Access

原文链接：https://docs.spring.io/spring/docs/current/spring-framework-reference/data-access.html#orm

4.1. Introduction to ORM with Spring

The Spring Framework supports integration with the Java Persistence API (JPA) as well as native Hibernate for resource management, data access object (DAO) implementations, and transaction strategies. For example, for Hibernate there is first-class support with several convenient IoC features that address many typical Hibernate integration issues. You can configure all of the supported features for O/R (object relational) mapping tools through Dependency Injection. They can participate in Spring’s resource and transaction management, and they comply with Spring’s generic transaction and DAO exception hierarchies. The recommended integration style is to code DAOs against plain Hibernate or JPA APIs. The older style of using Spring’s DAO templates is no longer recommended; however, coverage of this style can be found in the Classic ORM usage in the appendices.

Spring adds significant enhancements to the ORM layer of your choice when you create data access applications. You can leverage as much of the integration support as you wish, and you should compare this integration effort with the cost and risk of building a similar infrastructure in-house. You can use much of the ORM support as you would a library, regardless of technology, because everything is designed as a set of reusable JavaBeans. ORM in a Spring IoC container facilitates configuration and deployment. Thus most examples in this section show configuration inside a Spring container.

Benefits of using the Spring Framework to create your ORM DAOs include:

• Easier testing. Spring’s IoC approach makes it easy to swap the implementations and configuration locations of Hibernate SessionFactory instances, JDBC DataSource instances, transaction managers, and mapped object implementations (if needed). This in turn makes it much easier to test each piece of persistence-related code in isolation.

• Common data access exceptions. Spring can wrap exceptions from your ORM tool, converting them from proprietary (potentially checked) exceptions to a common runtime DataAccessException hierarchy. This feature allows you to handle most persistence exceptions, which are non-recoverable, only in the appropriate layers, without annoying boilerplate catches, throws, and exception declarations. You can still trap and handle exceptions as necessary. Remember that JDBC exceptions (including DB-specific dialects) are also converted to the same hierarchy, meaning that you can perform some operations with JDBC within a consistent programming model.

• General resource management. Spring application contexts can handle the location and configuration of Hibernate SessionFactory instances, JPA EntityManagerFactory instances, JDBC DataSource instances, and other related resources. This makes these values easy to manage and change. Spring offers efficient, easy, and safe handling of persistence resources. For example, related code that uses Hibernate generally needs to use the same Hibernate Session to ensure efficiency and proper transaction handling. Spring makes it easy to create and bind a Session to the current thread transparently, by exposing a current Session through the Hibernate SessionFactory. Thus Spring solves many chronic problems of typical Hibernate usage, for any local or JTA transaction environment.

• Integrated transaction management. You can wrap your ORM code with a declarative, aspect-oriented programming (AOP) style method interceptor either through the @Transactional annotation or by explicitly configuring the transaction AOP advice in an XML configuration file. In both cases, transaction semantics and exception handling (rollback, and so on) are handled for you. As discussed below, in Resource and transaction management, you can also swap various transaction managers, without affecting your ORM-related code. For example, you can swap between local transactions and JTA, with the same full services (such as declarative transactions) available in both scenarios. Additionally, JDBC-related code can fully integrate transactionally with the code you use to do ORM. This is useful for data access that is not suitable for ORM, such as batch processing and BLOB streaming, which still need to share common transactions with ORM operations.

For more comprehensive ORM support, including support for alternative database technologies such as MongoDB, you might want to check out the Spring Data suite of projects. If you are a JPA user, the Getting Started Accessing Data with JPA guide from https://spring.io provides a great introduction.

4.2. General ORM integration considerations

This section highlights considerations that apply to all ORM technologies. The Hibernate section provides more details and also show these features and configurations in a concrete context.

The major goal of Spring’s ORM integration is clear application layering, with any data access and transaction technology, and for loose coupling of application objects. No more business service dependencies on the data access or transaction strategy, no more hard-coded resource lookups, no more hard-to-replace singletons, no more custom service registries. One simple and consistent approach to wiring up application objects, keeping them as reusable and free from container dependencies as possible. All the individual data access features are usable on their own but integrate nicely with Spring’s application context concept, providing XML-based configuration and cross-referencing of plain JavaBean instances that need not be Spring-aware. In a typical Spring application, many important objects are JavaBeans: data access templates, data access objects, transaction managers, business services that use the data access objects and transaction managers, web view resolvers, web controllers that use the business services,and so on.

4.2.1. Resource and transaction management

Typical business applications are cluttered with repetitive resource management code. Many projects try to invent their own solutions, sometimes sacrificing proper handling of failures for programming convenience. Spring advocates simple solutions for proper resource handling, namely IoC through templating in the case of JDBC and applying AOP interceptors for the ORM technologies.

The infrastructure provides proper resource handling and appropriate conversion of specific API exceptions to an unchecked infrastructure exception hierarchy. Spring introduces a DAO exception hierarchy, applicable to any data access strategy. For direct JDBC, the JdbcTemplate class mentioned in a previous section provides connection handling and proper conversion of SQLException to the DataAccessException hierarchy, including translation of database-specific SQL error codes to meaningful exception classes. For ORM technologies, see the next section for how to get the same exception translation benefits.

When it comes to transaction management, the JdbcTemplate class hooks in to the Spring transaction support and supports both JTA and JDBC transactions, through respective Spring transaction managers. For the supported ORM technologies Spring offers Hibernate and JPA support through the Hibernate and JPA transaction managers as well as JTA support. For details on transaction support, see the Transaction Management chapter.

4.2.2. Exception translation

When you use Hibernate or JPA in a DAO, you must decide how to handle the persistence technology’s native exception classes. The DAO throws a subclass of a HibernateException or PersistenceException depending on the technology. These exceptions are all runtime exceptions and do not have to be declared or caught. You may also have to deal with IllegalArgumentExceptionand IllegalStateException. This means that callers can only treat exceptions as generally fatal, unless they want to depend on the persistence technology’s own exception structure. Catching specific causes such as an optimistic locking failure is not possible without tying the caller to the implementation strategy. This trade-off might be acceptable to applications that are strongly ORM-based and/or do not need any special exception treatment. However, Spring enables exception translation to be applied transparently through the @Repository annotation:

@Repositorypublic class ProductDaoImpl implements ProductDao {        // class body here...}

<beans>        <!-- Exception translation bean post processor -->        <bean class="org.springframework.dao.annotation.PersistenceExceptionTranslationPostProcessor"/>        <bean id="myProductDao" class="product.ProductDaoImpl"/></beans>

The postprocessor automatically looks for all exception translators (implementations of the PersistenceExceptionTranslatorinterface) and advises all beans marked with the @Repository annotation so that the discovered translators can intercept and apply the appropriate translation on the thrown exceptions.

In summary: you can implement DAOs based on the plain persistence technology’s API and annotations, while still benefiting from Spring-managed transactions, dependency injection, and transparent exception conversion (if desired) to Spring’s custom exception hierarchies.

4.3. Hibernate

We will start with a coverage of Hibernate 5 in a Spring environment, using it to demonstrate the approach that Spring takes towards integrating O/R mappers. This section will cover many issues in detail and show different variations of DAO implementations and transaction demarcation. Most of these patterns can be directly translated to all other supported ORM tools. The following sections in this chapter will then cover the other ORM technologies, showing briefer examples there.

As of Spring Framework 5.0, Spring requires Hibernate ORM 4.3 or later for JPA support and even Hibernate ORM 5.0+ for programming against the native Hibernate Session API. Note that the Hibernate team does not maintain any versions prior to 5.0 anymore and is likely to focus on 5.2+ exclusively soon.

4.3.1. SessionFactory setup in a Spring container

To avoid tying application objects to hard-coded resource lookups, you can define resources such as a JDBC DataSource or a Hibernate SessionFactory as beans in the Spring container. Application objects that need to access resources receive references to such predefined instances through bean references, as illustrated in the DAO definition in the next section.

The following excerpt from an XML application context definition shows how to set up a JDBC DataSource and a Hibernate SessionFactory on top of it:

<beans>        <bean id="myDataSource" class="org.apache.commons.dbcp.BasicDataSource" destroy-method="close">                <property name="driverClassName" value="org.hsqldb.jdbcDriver"/>                <property name="url" value="jdbc:hsqldb:hsql://localhost:9001"/>                <property name="username" value="sa"/>                <property name="password" value=""/>        </bean>        <bean id="mySessionFactory" class="org.springframework.orm.hibernate5.LocalSessionFactoryBean">                <property name="dataSource" ref="myDataSource"/>                <property name="mappingResources">                        <list>                                <value>product.hbm.xml</value>                        </list>                </property>                <property name="hibernateProperties">                        <value>                                hibernate.dialect=org.hibernate.dialect.HSQLDialect                        </value>                </property>        </bean></beans>

Switching from a local Jakarta Commons DBCP BasicDataSource to a JNDI-located DataSource (usually managed by an application server) is just a matter of configuration:

<beans>        <jee:jndi-lookup id="myDataSource" jndi-name="java:comp/env/jdbc/myds"/></beans>

You can also access a JNDI-located SessionFactory, using Spring’s JndiObjectFactoryBean / <jee:jndi-lookup> to retrieve and expose it. However, that is typically not common outside of an EJB context.

4.3.2. Implementing DAOs based on plain Hibernate API

Hibernate has a feature called contextual sessions, wherein Hibernate itself manages one current Session per transaction. This is roughly equivalent to Spring’s synchronization of one Hibernate Session per transaction. A corresponding DAO implementation resembles the following example, based on the plain Hibernate API:

public class ProductDaoImpl implements ProductDao {        private SessionFactory sessionFactory;        public void setSessionFactory(SessionFactory sessionFactory) {                this.sessionFactory = sessionFactory;        }        public Collection loadProductsByCategory(String category) {                return this.sessionFactory.getCurrentSession()                                .createQuery("from test.Product product where product.category=?")                                .setParameter(0, category)                                .list();        }}

This style is similar to that of the Hibernate reference documentation and examples, except for holding the SessionFactory in an instance variable. We strongly recommend such an instance-based setup over the old-school static HibernateUtil class from Hibernate’s CaveatEmptor sample application. (In general, do not keep any resources in static variables unless absolutelynecessary.)

The above DAO follows the dependency injection pattern: it fits nicely into a Spring IoC container, just as it would if coded against Spring’s HibernateTemplate. Of course, such a DAO can also be set up in plain Java (for example, in unit tests). Simply instantiate it and call setSessionFactory(..) with the desired factory reference. As a Spring bean definition, the DAO would resemble the following:

<beans>        <bean id="myProductDao" class="product.ProductDaoImpl">                <property name="sessionFactory" ref="mySessionFactory"/>        </bean></beans>

The main advantage of this DAO style is that it depends on Hibernate API only; no import of any Spring class is required. This is of course appealing from a non-invasiveness perspective, and will no doubt feel more natural to Hibernate developers.

However, the DAO throws plain HibernateException (which is unchecked, so does not have to be declared or caught), which means that callers can only treat exceptions as generally fatal - unless they want to depend on Hibernate’s own exception hierarchy. Catching specific causes such as an optimistic locking failure is not possible without tying the caller to the implementation strategy. This trade off might be acceptable to applications that are strongly Hibernate-based and/or do not need any special exception treatment.

Fortunately, Spring’s LocalSessionFactoryBean supports Hibernate’s SessionFactory.getCurrentSession() method for any Spring transaction strategy, returning the current Spring-managed transactional Session even withHibernateTransactionManager. Of course, the standard behavior of that method remains the return of the current Sessionassociated with the ongoing JTA transaction, if any. This behavior applies regardless of whether you are using Spring’sJtaTransactionManager, EJB container managed transactions (CMTs), or JTA.

In summary: you can implement DAOs based on the plain Hibernate API, while still being able to participate in Spring-managed transactions.

4.3.3. Declarative transaction demarcation

We recommend that you use Spring’s declarative transaction support, which enables you to replace explicit transaction demarcation API calls in your Java code with an AOP transaction interceptor. This transaction interceptor can be configured in a Spring container using either Java annotations or XML. This declarative transaction capability allows you to keep business services free of repetitive transaction demarcation code and to focus on adding business logic, which is the real value of your application.

Prior to continuing, you are strongly encouraged to read Declarative transaction management if you have not done so.

You may annotate the service layer with @Transactional annotations and instruct the Spring container to find these annotations and provide transactional semantics for these annotated methods.

public class ProductServiceImpl implements ProductService {        private ProductDao productDao;        public void setProductDao(ProductDao productDao) {                this.productDao = productDao;        }        @Transactional        public void increasePriceOfAllProductsInCategory(final String category) {                List productsToChange = this.productDao.loadProductsByCategory(category);                // ...        }        @Transactional(readOnly = true)        public List<Product> findAllProducts() {                return this.productDao.findAllProducts();        }}

All you need to set up in the container is the PlatformTransactionManager implementation as a bean as well as a "<tx:annotation-driven/>" entry, opting into @Transactional processing at runtime.

<?xml version="1.0" encoding="UTF-8"?><beans xmlns="http://www.springframework.org/schema/beans"        xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"        xmlns:aop="http://www.springframework.org/schema/aop"        xmlns:tx="http://www.springframework.org/schema/tx"        xsi:schemaLocation="                http://www.springframework.org/schema/beans                http://www.springframework.org/schema/beans/spring-beans.xsd                http://www.springframework.org/schema/tx                http://www.springframework.org/schema/tx/spring-tx.xsd                http://www.springframework.org/schema/aop                http://www.springframework.org/schema/aop/spring-aop.xsd">        <!-- SessionFactory, DataSource, etc. omitted -->        <bean id="transactionManager"                        class="org.springframework.orm.hibernate5.HibernateTransactionManager">                <property name="sessionFactory" ref="sessionFactory"/>        </bean>        <tx:annotation-driven/>        <bean id="myProductService" class="product.SimpleProductService">                <property name="productDao" ref="myProductDao"/>        </bean></beans>

4.3.4. Programmatic transaction demarcation

You can demarcate transactions in a higher level of the application, on top of such lower-level data access services spanning any number of operations. Nor do restrictions exist on the implementation of the surrounding business service; it just needs a SpringPlatformTransactionManager. Again, the latter can come from anywhere, but preferably as a bean reference through a setTransactionManager(..) method, just as the productDAO should be set by a setProductDao(..) method. The following snippets show a transaction manager and a business service definition in a Spring application context, and an example for a business method implementation:

<beans>        <bean id="myTxManager" class="org.springframework.orm.hibernate5.HibernateTransactionManager">                <property name="sessionFactory" ref="mySessionFactory"/>        </bean>        <bean id="myProductService" class="product.ProductServiceImpl">                <property name="transactionManager" ref="myTxManager"/>                <property name="productDao" ref="myProductDao"/>        </bean></beans>

public class ProductServiceImpl implements ProductService {        private TransactionTemplate transactionTemplate;        private ProductDao productDao;        public void setTransactionManager(PlatformTransactionManager transactionManager) {                this.transactionTemplate = new TransactionTemplate(transactionManager);        }        public void setProductDao(ProductDao productDao) {                this.productDao = productDao;        }        public void increasePriceOfAllProductsInCategory(final String category) {                this.transactionTemplate.execute(new TransactionCallbackWithoutResult() {                        public void doInTransactionWithoutResult(TransactionStatus status) {                                List productsToChange = this.productDao.loadProductsByCategory(category);                                // do the price increase...                        }                });        }}

Spring’s TransactionInterceptor allows any checked application exception to be thrown with the callback code, while TransactionTemplate is restricted to unchecked exceptions within the callback. TransactionTemplate triggers a rollback in case of an unchecked application exception, or if the transaction is marked rollback-only by the application (via TransactionStatus). TransactionInterceptor behaves the same way by default but allows configurable rollback policies per method.

4.3.5. Transaction management strategies

Both TransactionTemplate and TransactionInterceptor delegate the actual transaction handling to a PlatformTransactionManager instance, which can be a HibernateTransactionManager (for a single Hibernate SessionFactory, using a ThreadLocal Session under the hood) or a JtaTransactionManager (delegating to the JTA subsystem of the container) for Hibernate applications. You can even use a custom PlatformTransactionManager implementation. Switching from native Hibernate transaction management to JTA, such as when facing distributed transaction requirements for certain deployments of your application, is just a matter of configuration. Simply replace the Hibernate transaction manager with Spring’s JTA transaction implementation. Both transaction demarcation and data access code will work without changes, because they just use the generic transaction management APIs.

For distributed transactions across multiple Hibernate session factories, simply combine JtaTransactionManager as a transaction strategy with multiple LocalSessionFactoryBean definitions. Each DAO then gets one specific SessionFactoryreference passed into its corresponding bean property. If all underlying JDBC data sources are transactional container ones, a business service can demarcate transactions across any number of DAOs and any number of session factories without special regard, as long as it is using JtaTransactionManager as the strategy.

Both HibernateTransactionManager and JtaTransactionManager allow for proper JVM-level cache handling with Hibernate, without container-specific transaction manager lookup or a JCA connector (if you are not using EJB to initiate transactions).

HibernateTransactionManager can export the Hibernate JDBC Connection to plain JDBC access code, for a specific DataSource. This capability allows for high-level transaction demarcation with mixed Hibernate and JDBC data access completely without JTA, if you are accessing only one database. HibernateTransactionManager automatically exposes the Hibernate transaction as a JDBC transaction if you have set up the passed-in SessionFactory with a DataSource through the dataSource property of theLocalSessionFactoryBean class. Alternatively, you can specify explicitly the DataSource for which the transactions are supposed to be exposed through the dataSource property of the HibernateTransactionManager class.

4.3.6. Comparing container-managed and locally defined resources

You can switch between a container-managed JNDI SessionFactory and a locally defined one, without having to change a single line of application code. Whether to keep resource definitions in the container or locally within the application is mainly a matter of the transaction strategy that you use. Compared to a Spring-defined local SessionFactory, a manually registered JNDI SessionFactory does not provide any benefits. Deploying a SessionFactory through Hibernate’s JCA connector provides the added value of participating in the Java EE server’s management infrastructure, but does not add actual value beyond that.

Spring’s transaction support is not bound to a container. Configured with any strategy other than JTA, transaction support also works in a stand-alone or test environment. Especially in the typical case of single-database transactions, Spring’s single-resource local transaction support is a lightweight and powerful alternative to JTA. When you use local EJB stateless session beans to drive transactions, you depend both on an EJB container and JTA, even if you access only a single database, and only use stateless session beans to provide declarative transactions through container-managed transactions. Also, direct use of JTA programmatically requires a Java EE environment as well. JTA does not involve only container dependencies in terms of JTA itself and of JNDI DataSource instances. For non-Spring, JTA-driven Hibernate transactions, you have to use the Hibernate JCA connector, or extra Hibernate transaction code with the TransactionManagerLookup configured for proper JVM-level caching.

Spring-driven transactions can work as well with a locally defined Hibernate SessionFactory as they do with a local JDBC DataSource if they are accessing a single database. Thus you only have to use Spring’s JTA transaction strategy when you have distributed transaction requirements. A JCA connector requires container-specific deployment steps, and obviously JCA support in the first place. This configuration requires more work than deploying a simple web application with local resource definitions and Spring-driven transactions. Also, you often need the Enterprise Edition of your container if you are using, for example, WebLogic Express, which does not provide JCA. A Spring application with local resources and transactions spanning one single database works in any Java EE web container (without JTA, JCA, or EJB) such as Tomcat, Resin, or even plain Jetty. Additionally, you can easily reuse such a middle tier in desktop applications or test suites.

All things considered, if you do not use EJBs, stick with local SessionFactory setup and Spring’s HibernateTransactionManageror JtaTransactionManager. You get all of the benefits, including proper transactional JVM-level caching and distributed transactions, without the inconvenience of container deployment. JNDI registration of a Hibernate SessionFactory through the JCA connector only adds value when used in conjunction with EJBs.

4.3.7. Spurious application server warnings with Hibernate

In some JTA environments with very strict XADataSource implementations — currently only some WebLogic Server and WebSphere versions — when Hibernate is configured without regard to the JTA PlatformTransactionManager object for that environment, it is possible for spurious warning or exceptions to show up in the application server log. These warnings or exceptions indicate that the connection being accessed is no longer valid, or JDBC access is no longer valid, possibly because the transaction is no longer active. As an example, here is an actual exception from WebLogic:

java.sql.SQLException: The transaction is no longer active - status: 'Committed'. Nofurther JDBC access is allowed within this transaction.

You resolve this warning by simply making Hibernate aware of the JTA PlatformTransactionManager instance, to which it will synchronize (along with Spring). You have two options for doing this:

• If in your application context you are already directly obtaining the JTA PlatformTransactionManager object (presumably from JNDI through JndiObjectFactoryBean or <jee:jndi-lookup>) and feeding it, for example, to Spring’s JtaTransactionManager, then the easiest way is to specify a reference to the bean defining this JTA PlatformTransactionManager instance as the value of the jtaTransactionManager property for LocalSessionFactoryBean.Spring then makes the object available to Hibernate.

• More likely you do not already have the JTA PlatformTransactionManager instance, because Spring’s JtaTransactionManagercan find it itself. Thus you need to configure Hibernate to look up JTA PlatformTransactionManager directly. You do this by configuring an application server- specific TransactionManagerLookup class in the Hibernate configuration, as described in the Hibernate manual.

The remainder of this section describes the sequence of events that occur with and without Hibernate’s awareness of the JTA PlatformTransactionManager.

When Hibernate is not configured with any awareness of the JTA PlatformTransactionManager, the following events occur when a JTA transaction commits:

• The JTA transaction commits.

• Spring’s JtaTransactionManager is synchronized to the JTA transaction, so it is called back through an afterCompletioncallback by the JTA transaction manager.

• Among other activities, this synchronization can trigger a callback by Spring to Hibernate, through Hibernate’s afterTransactionCompletion callback (used to clear the Hibernate cache), followed by an explicit close() call on the Hibernate Session, which causes Hibernate to attempt to close() the JDBC Connection.

• In some environments, this Connection.close() call then triggers the warning or error, as the application server no longer considers the Connection usable at all, because the transaction has already been committed.

When Hibernate is configured with awareness of the JTA PlatformTransactionManager, the following events occur when a JTA transaction commits:

• the JTA transaction is ready to commit.

• Spring’s JtaTransactionManager is synchronized to the JTA transaction, so the transaction is called back through a beforeCompletion callback by the JTA transaction manager.

• Spring is aware that Hibernate itself is synchronized to the JTA transaction, and behaves differently than in the previous scenario. Assuming the Hibernate Session needs to be closed at all, Spring will close it now.

• The JTA transaction commits.

• Hibernate is synchronized to the JTA transaction, so the transaction is called back through an afterCompletion callback by the JTA transaction manager, and can properly clear its cache.

4.4. JPA

The Spring JPA, available under the org.springframework.orm.jpa package, offers comprehensive support for the Java Persistence API in a similar manner to the integration with Hibernate, while being aware of the underlying implementation in order to provide additional features.

4.4.1. Three options for JPA setup in a Spring environment

The Spring JPA support offers three ways of setting up the JPA EntityManagerFactory that will be used by the application to obtain an entity manager.

LocalEntityManagerFactoryBean

Only use this option in simple deployment environments such as stand-alone applications and integration tests.

The LocalEntityManagerFactoryBean creates an EntityManagerFactory suitable for simple deployment environments where the application uses only JPA for data access. The factory bean uses the JPA PersistenceProvider autodetection mechanism (according to JPA’s Java SE bootstrapping) and, in most cases, requires you to specify only the persistence unit name:

<beans>        <bean id="myEmf" class="org.springframework.orm.jpa.LocalEntityManagerFactoryBean">                <property name="persistenceUnitName" value="myPersistenceUnit"/>        </bean></beans>

This form of JPA deployment is the simplest and the most limited. You cannot refer to an existing JDBC DataSource bean definition and no support for global transactions exists. Furthermore, weaving (byte-code transformation) of persistent classes is provider-specific, often requiring a specific JVM agent to specified on startup. This option is sufficient only for stand-alone applications and test environments, for which the JPA specification is designed.

Obtaining an EntityManagerFactory from JNDI

Use this option when deploying to a Java EE server. Check your server’s documentation on how to deploy a custom JPA provider into your server, allowing for a different provider than the server’s default.

Obtaining an EntityManagerFactory from JNDI (for example in a Java EE environment), is simply a matter of changing the XML configuration:

<beans>        <jee:jndi-lookup id="myEmf" jndi-name="persistence/myPersistenceUnit"/></beans>

This action assumes standard Java EE bootstrapping: the Java EE server autodetects persistence units (in effect, META-INF/persistence.xml files in application jars) and persistence-unit-ref entries in the Java EE deployment descriptor (for example, web.xml) and defines environment naming context locations for those persistence units.

In such a scenario, the entire persistence unit deployment, including the weaving (byte-code transformation) of persistent classes, is up to the Java EE server. The JDBC DataSource is defined through a JNDI location in the META-INF/persistence.xmlfile; EntityManager transactions are integrated with the server’s JTA subsystem. Spring merely uses the obtained EntityManagerFactory, passing it on to application objects through dependency injection, and managing transactions for the persistence unit, typically through JtaTransactionManager.

If multiple persistence units are used in the same application, the bean names of such JNDI-retrieved persistence units should match the persistence unit names that the application uses to refer to them, for example, in @PersistenceUnit and@PersistenceContext annotations.

LocalContainerEntityManagerFactoryBean

Use this option for full JPA capabilities in a Spring-based application environment. This includes web containers such as Tomcat as well as stand-alone applications and integration tests with sophisticated persistence requirements.

The LocalContainerEntityManagerFactoryBean gives full control over EntityManagerFactory configuration and is appropriate for environments where fine-grained customization is required. The LocalContainerEntityManagerFactoryBean creates a PersistenceUnitInfo instance based on the persistence.xml file, the supplied dataSourceLookup strategy, and the specified loadTimeWeaver. It is thus possible to work with custom data sources outside of JNDI and to control the weaving process. The following example shows a typical bean definition for a LocalContainerEntityManagerFactoryBean:

<beans>        <bean id="myEmf" class="org.springframework.orm.jpa.LocalContainerEntityManagerFactoryBean">                <property name="dataSource" ref="someDataSource"/>                <property name="loadTimeWeaver">                        <bean class="org.springframework.instrument.classloading.InstrumentationLoadTimeWeaver"/>                </property>        </bean></beans>

The following example shows a typical persistence.xml file:

<persistence xmlns="http://java.sun.com/xml/ns/persistence" version="1.0">        <persistence-unit name="myUnit" transaction-type="RESOURCE_LOCAL">                <mapping-file>META-INF/orm.xml</mapping-file>                <exclude-unlisted-classes/>        </persistence-unit></persistence>

The <exclude-unlisted-classes/> shortcut indicates that no scanning for annotated entity classes is supposed to occur. An explicit 'true' value specified - <exclude-unlisted-classes>true</exclude-unlisted-classes/> - also means no scan. <exclude-unlisted-classes>false</exclude-unlisted-classes/> does trigger a scan; however, it is recommended to simply omit the exclude-unlisted-classes element if you want entity class scanning to occur.

Using the LocalContainerEntityManagerFactoryBean is the most powerful JPA setup option, allowing for flexible local configuration within the application. It supports links to an existing JDBC DataSource, supports both local and global transactions, and so on. However, it also imposes requirements on the runtime environment, such as the availability of a weaving-capable class loader if the persistence provider demands byte-code transformation.

This option may conflict with the built-in JPA capabilities of a Java EE server. In a full Java EE environment, consider obtaining your EntityManagerFactory from JNDI. Alternatively, specify a custom persistenceXmlLocation on yourLocalContainerEntityManagerFactoryBean definition, for example, META-INF/my-persistence.xml, and only include a descriptor with that name in your application jar files. Because the Java EE server only looks for default META-INF/persistence.xml files, it ignores such custom persistence units and hence avoid conflicts with a Spring-driven JPA setup upfront. (This applies to Resin 3.1, for example.)

When is load-time weaving required?

Not all JPA providers require a JVM agent. Hibernate is an example of one that does not. If your provider does not require an agent or you have other alternatives, such as applying enhancements at build time through a custom compiler or an ant task, the load-time weaver should not be used.

The LoadTimeWeaver interface is a Spring-provided class that allows JPA ClassTransformer instances to be plugged in a specific manner, depending whether the environment is a web container or application server. Hooking ClassTransformers through anagent typically is not efficient. The agents work against the entire virtual machine and inspect every class that is loaded, which is usually undesirable in a production server environment.

Spring provides a number of LoadTimeWeaver implementations for various environments, allowing ClassTransformer instances to be applied only per class loader and not per VM.

Refer to Spring configuration in the AOP chapter for more insight regarding the LoadTimeWeaver implementations and their setup, either generic or customized to various platforms (such as Tomcat, WebLogic, GlassFish, Resin and JBoss).

As described in the aforementioned section, you can configure a context-wide LoadTimeWeaver using the @EnableLoadTimeWeaving annotation of context:load-time-weaver XML element. Such a global weaver is picked up by all JPALocalContainerEntityManagerFactoryBeans automatically. This is the preferred way of setting up a load-time weaver, delivering autodetection of the platform (WebLogic, GlassFish, Tomcat, Resin, JBoss or VM agent) and automatic propagation of the weaver to all weaver-aware beans:

<context:load-time-weaver/><bean id="emf" class="org.springframework.orm.jpa.LocalContainerEntityManagerFactoryBean">        ...</bean>

However, if needed, one can manually specify a dedicated weaver through the loadTimeWeaver property:

<bean id="emf" class="org.springframework.orm.jpa.LocalContainerEntityManagerFactoryBean">        <property name="loadTimeWeaver">                <bean class="org.springframework.instrument.classloading.ReflectiveLoadTimeWeaver"/>        </property></bean>

No matter how the LTW is configured, using this technique, JPA applications relying on instrumentation can run in the target platform (ex: Tomcat) without needing an agent. This is important especially when the hosting applications rely on different JPA implementations because the JPA transformers are applied only at class loader level and thus are isolated from each other.

Dealing with multiple persistence units

For applications that rely on multiple persistence units locations, stored in various JARS in the classpath, for example, Spring offers the PersistenceUnitManager to act as a central repository and to avoid the persistence units discovery process, which can be expensive. The default implementation allows multiple locations to be specified that are parsed and later retrieved through the persistence unit name. (By default, the classpath is searched for META-INF/persistence.xml files.)

<bean id="pum" class="org.springframework.orm.jpa.persistenceunit.DefaultPersistenceUnitManager">        <property name="persistenceXmlLocations">                <list>                        <value>org/springframework/orm/jpa/domain/persistence-multi.xml</value>                        <value>classpath:/my/package/**/custom-persistence.xml</value>                        <value>classpath*:META-INF/persistence.xml</value>                </list>        </property>        <property name="dataSources">                <map>                        <entry key="localDataSource" value-ref="local-db"/>                        <entry key="remoteDataSource" value-ref="remote-db"/>                </map>        </property>        <!-- if no datasource is specified, use this one -->        <property name="defaultDataSource" ref="remoteDataSource"/></bean><bean id="emf" class="org.springframework.orm.jpa.LocalContainerEntityManagerFactoryBean">        <property name="persistenceUnitManager" ref="pum"/>        <property name="persistenceUnitName" value="myCustomUnit"/></bean>

The default implementation allows customization of the PersistenceUnitInfo instances, before they are fed to the JPA provider, declaratively through its properties, which affect all hosted units, or programmatically, through thePersistenceUnitPostProcessor, which allows persistence unit selection. If no PersistenceUnitManager is specified, one is created and used internally by LocalContainerEntityManagerFactoryBean.

4.4.2. Implementing DAOs based on JPA: EntityManagerFactory and EntityManager

Although EntityManagerFactory instances are thread-safe, EntityManager instances are not. The injected JPA EntityManager behaves like an EntityManager fetched from an application server’s JNDI environment, as defined by the JPA specification. It delegates all calls to the current transactional EntityManager, if any; otherwise, it falls back to a newly created EntityManager per operation, in effect making its usage thread-safe.

It is possible to write code against the plain JPA without any Spring dependencies, by using an injected EntityManagerFactoryor EntityManager. Spring can understand @PersistenceUnit and @PersistenceContext annotations both at field and method level if a PersistenceAnnotationBeanPostProcessor is enabled. A plain JPA DAO implementation using the @PersistenceUnitannotation might look like this:

public class ProductDaoImpl implements ProductDao {        private EntityManagerFactory emf;        @PersistenceUnit        public void setEntityManagerFactory(EntityManagerFactory emf) {                this.emf = emf;        }        public Collection loadProductsByCategory(String category) {                EntityManager em = this.emf.createEntityManager();                try {                        Query query = em.createQuery("from Product as p where p.category = ?1");                        query.setParameter(1, category);                        return query.getResultList();                }                finally {                        if (em != null) {                                em.close();                        }                }        }}

The DAO above has no dependency on Spring and still fits nicely into a Spring application context. Moreover, the DAO takes advantage of annotations to require the injection of the default EntityManagerFactory:

<beans>        <!-- bean post-processor for JPA annotations -->        <bean class="org.springframework.orm.jpa.support.PersistenceAnnotationBeanPostProcessor"/>        <bean id="myProductDao" class="product.ProductDaoImpl"/></beans>

As an alternative to defining a PersistenceAnnotationBeanPostProcessor explicitly, consider using the Spring context:annotation-config XML element in your application context configuration. Doing so automatically registers all Spring standard post-processors for annotation-based configuration, including CommonAnnotationBeanPostProcessor and so on.

<beans>        <!-- post-processors for all standard config annotations -->        <context:annotation-config/>        <bean id="myProductDao" class="product.ProductDaoImpl"/></beans>

The main problem with such a DAO is that it always creates a new EntityManager through the factory. You can avoid this by requesting a transactional EntityManager (also called "shared EntityManager" because it is a shared, thread-safe proxy for the actual transactional EntityManager) to be injected instead of the factory:

public class ProductDaoImpl implements ProductDao {        @PersistenceContext        private EntityManager em;        public Collection loadProductsByCategory(String category) {                Query query = em.createQuery("from Product as p where p.category = :category");                query.setParameter("category", category);                return query.getResultList();        }}

The @PersistenceContext annotation has an optional attribute type, which defaults to PersistenceContextType.TRANSACTION. This default is what you need to receive a shared EntityManager proxy. The alternative, PersistenceContextType.EXTENDED, is a completely different affair: This results in a so-called extended EntityManager, which is not thread-safe and hence must not be used in a concurrently accessed component such as a Spring-managed singleton bean. Extended EntityManagers are only supposed to be used in stateful components that, for example, reside in a session, with the lifecycle of the EntityManager not tied to a current transaction but rather being completely up to the application.

Method- and field-level Injection

Annotations that indicate dependency injections (such as @PersistenceUnit and @PersistenceContext) can be applied on field or methods inside a class, hence the expressions method-level injection and field-level injection. Field-level annotations are concise and easier to use while method-level allows for further processing of the injected dependency. In both cases the member visibility (public, protected, private) does not matter.

What about class-level annotations?

On the Java EE platform, they are used for dependency declaration and not for resource injection.

The injected EntityManager is Spring-managed (aware of the ongoing transaction). It is important to note that even though the new DAO implementation uses method level injection of an EntityManager instead of an EntityManagerFactory, no change is required in the application context XML due to annotation usage.

The main advantage of this DAO style is that it only depends on Java Persistence API; no import of any Spring class is required. Moreover, as the JPA annotations are understood, the injections are applied automatically by the Spring container. This is appealing from a non-invasiveness perspective, and might feel more natural to JPA developers.

4.4.3. Spring-driven JPA transactions

You are strongly encouraged to read Declarative transaction management if you have not done so, to get a more detailed coverage of Spring’s declarative transaction support.

The recommended strategy for JPA is local transactions via JPA’s native transaction support. Spring’s JpaTransactionManagerprovides many capabilities known from local JDBC transactions, such as transaction-specific isolation levels and resource-level read-only optimizations, against any regular JDBC connection pool (no XA requirement).

Spring JPA also allows a configured JpaTransactionManager to expose a JPA transaction to JDBC access code that accesses the same DataSource, provided that the registered JpaDialect supports retrieval of the underlying JDBC Connection. Out of the box, Spring provides dialects for the EclipseLink and Hibernate JPA implementations. See the next section for details on the JpaDialect mechanism.

4.4.4. JpaDialect and JpaVendorAdapter

As an advanced feature JpaTransactionManager and subclasses of AbstractEntityManagerFactoryBean support a custom JpaDialect, to be passed into the jpaDialect bean property. A JpaDialect implementation can enable some advanced features supported by Spring, usually in a vendor-specific manner:

• Applying specific transaction semantics such as custom isolation level or transaction timeout)

• Retrieving the transactional JDBC Connection for exposure to JDBC-based DAOs)

• Advanced translation of PersistenceExceptions to Spring DataAccessExceptions

This is particularly valuable for special transaction semantics and for advanced translation of exception. The default implementation used (DefaultJpaDialect) does not provide any special capabilities and if the above features are required, you have to specify the appropriate dialect.

As an even broader provider adaptation facility primarily for Spring’s full-featuredLocalContainerEntityManagerFactoryBean setup, JpaVendorAdapter combines the capabilities of JpaDialectwith other provider-specific defaults. Specifying a HibernateJpaVendorAdapter or EclipseLinkJpaVendorAdapter is the most convenient way of auto-configuring an EntityManagerFactory setup for Hibernate or EclipseLink, respectively. Note that those provider adapters are primarily designed for use with Spring-driven transaction management, i.e. for use with JpaTransactionManager.

See the JpaDialect and JpaVendorAdapter javadocs for more details of its operations and how they are used within Spring’s JPA support.

4.4.5. Setting up JPA with JTA transaction management

As an alternative to JpaTransactionManager, Spring also allows for multi-resource transaction coordination via JTA, either in a Java EE environment or with a standalone transaction coordinator such as Atomikos. Aside from choosing Spring’sJtaTransactionManager instead of JpaTransactionManager, there are a few further steps to take:

• The underlying JDBC connection pools need to be XA-capable and integrated with your transaction coordinator. This is usually straightforward in a Java EE environment, simply exposing a different kind of DataSource via JNDI. Check your application server documentation for details. Analogously, a standalone transaction coordinator usually comes with special XA-integrated DataSource implementations; again, check its docs.

• The JPA EntityManagerFactory setup needs to be configured for JTA. This is provider-specific, typically via special properties to be specified as "jpaProperties" on LocalContainerEntityManagerFactoryBean. In the case of Hibernate, these properties are even version-specific; please check your Hibernate documentation for details.

• Spring’s HibernateJpaVendorAdapter enforces certain Spring-oriented defaults such as the connection release mode "on-close" which matches Hibernate’s own default in Hibernate 5.0 but not anymore in 5.1/5.2. For a JTA setup, either do not declare HibernateJpaVendorAdapter to begin with, or turn off its prepareConnection flag. Alternatively, set Hibernate 5.2’s "hibernate.connection.handling_mode" property to "DELAYED_ACQUISITION_AND_RELEASE_AFTER_STATEMENT" to restore Hibernate’s own default. See Spurious application server warnings with Hibernate for a related note about WebLogic.

• Alternatively, consider obtaining the EntityManagerFactory from your application server itself, i.e. via a JNDI lookup instead of a locally declared LocalContainerEntityManagerFactoryBean. A server-provided EntityManagerFactory might require special definitions in your server configuration, making the deployment less portable, but will be set up for the server’s JTA environment out of the box.