spring 事务管理实现方式

注意：这里只是分析spring事务的实现方式。也就是spring的事务管理是怎么发生作用的，而不分析具体的实现细节

Spring是通过NameSpaceHandler来解析配置文件中的标签的。下面就已事务的配置为例，讲述一下

事务配置的标签的解析过程，从来理解事物是如何通过aop产生作用的。

 

 以上的配置相信很多人已经很熟悉了，在此不赘述。而是具体分析一下原理。

 先来分析<tx:advice>...</tx:advice>。

tx是TransactionNameSpace可从/spring-tx/src/main/resources/META-INF/spring.handlers中得到对应的是handler是TxNamespaceHandler.

http\://www.springframework.org/schema/tx=org.springframework.transaction.config.TxNamespaceHandler

这个类一个init方法：

	public void init() {
		registerBeanDefinitionParser("advice", new TxAdviceBeanDefinitionParser());
		registerBeanDefinitionParser("annotation-driven", new AnnotationDrivenBeanDefinitionParser());
		registerBeanDefinitionParser("jta-transaction-manager", new JtaTransactionManagerBeanDefinitionParser());
	}

这个方法是在DefaultNamespaceHandlerResolver的resolve中调用的。在为对应的标签寻找namespacehandler的时候，调用这个resolve方法。resolve方法先寻找namespaceUri对应的namespacehandler,如果找到了就先调用Init方法。

 

    OK.我们的<tx:advice>对应的解析器也注册了，那就是上面代码里面的

<pre name="code" class="java">new TxAdviceBeanDefinitionParser()

    那么，这个解析器是什么时候调用的哪？

    上一篇提到了，对应标签解析时会先选择namespacehandler,然后调用其parse方法。

    TxNamespaceHandler的parse方法在其父类NamespaceHandlerSupport中，代码如下:

 

public BeanDefinition parse(Element element, ParserContext parserContext) {
		return findParserForElement(element, parserContext).parse(element, parserContext);
	}

   这下明白了吧？<tx:advice>在解析出来的Document里面是一个Element,而这个Element的parse就是上面注册了的

TxAdviceBeanDefinitionParser

  现在这个parser的parse方法在NamespaceHandlerSupport的parse方法中被调用了，下面我们来看看这个

  TxAdviceBeanDefinitionParser的parse方法吧，这个方法在TxAdviceBeanDefinitionParser的祖父类AbstractBeanDefinitionParser中：

 

public final BeanDefinition parse(Element element, ParserContext parserContext) {
		AbstractBeanDefinition definition = parseInternal(element, parserContext);
		if (definition != null && !parserContext.isNested()) {
			try {
				String id = resolveId(element, definition, parserContext);
				if (!StringUtils.hasText(id)) {
					parserContext.getReaderContext().error(
							"Id is required for element '" + parserContext.getDelegate().getLocalName(element)
									+ "' when used as a top-level tag", element);
				}
				String[] aliases = new String[0];
				String name = element.getAttribute(NAME_ATTRIBUTE);
				if (StringUtils.hasLength(name)) {
					aliases = StringUtils.trimArrayElements(StringUtils.commaDelimitedListToStringArray(name));
				}
				BeanDefinitionHolder holder = new BeanDefinitionHolder(definition, id, aliases);
				registerBeanDefinition(holder, parserContext.getRegistry());
				if (shouldFireEvents()) {
					BeanComponentDefinition componentDefinition = new BeanComponentDefinition(holder);
					postProcessComponentDefinition(componentDefinition);
					parserContext.registerComponent(componentDefinition);
				}
			}
			catch (BeanDefinitionStoreException ex) {
				parserContext.getReaderContext().error(ex.getMessage(), element);
				return null;
			}
		}
		return definition;
	}

   注意其中这样一行：

 

<pre name="code" class="java">AbstractBeanDefinition definition = parseInternal(element, parserContext);

    这个parseInternal是在TxAdviceBeanDefinitionParser的父类AbstractSingleBeanDefinitionParser中实现的，代码如下：

 protected final AbstractBeanDefinition parseInternal(Element element, ParserContext parserContext) {
		BeanDefinitionBuilder builder = BeanDefinitionBuilder.genericBeanDefinition();
		String parentName = getParentName(element);
		if (parentName != null) {
			builder.getRawBeanDefinition().setParentName(parentName);
		}
		Class beanClass = getBeanClass(element);
		if (beanClass != null) {
			builder.getRawBeanDefinition().setBeanClass(beanClass);
		}
		else {
			String beanClassName = getBeanClassName(element);
			if (beanClassName != null) {
				builder.getRawBeanDefinition().setBeanClassName(beanClassName);
			}
		}
		builder.getRawBeanDefinition().setSource(parserContext.extractSource(element));
		if (parserContext.isNested()) {
			// Inner bean definition must receive same scope as containing bean.
			builder.setScope(parserContext.getContainingBeanDefinition().getScope());
		}
		if (parserContext.isDefaultLazyInit()) {
			// Default-lazy-init applies to custom bean definitions as well.
			builder.setLazyInit(true);
		}
		doParse(element, parserContext, builder);
		return builder.getBeanDefinition();
	}

    其中有一行：

 

Class beanClass = getBeanClass(element);

   getBeanClass是在TxAdviceBeanDefinitionParser中实现的，很简单:

<pre name="code" class="java">	protected Class getBeanClass(Element element) {
		return TransactionInterceptor.class;
	}

  至此，这个标签解析的流程已经基本清晰了。那就是：解析除了一个以TransactionInerceptor为classname的beandefinition并且注册这个bean。剩下来要看的，就是这个TranscationInterceptor到底是什么？

 看看这个类的接口定义，就明白了：

 

<pre name="code" class="java">public class TransactionInterceptor extends TransactionAspectSupport implements MethodInterceptor, Serializable

  这根本就是一个spring AOP的advice嘛！现在明白为什么事务的配置能通过aop产生作用了吧？

  下面具体看看这个advice的advice:

 

<pre name="code" class="java">public Object invoke(final MethodInvocation invocation) throws Throwable {
		// Work out the target class: may be <code>null</code>.
		// The TransactionAttributeSource should be passed the target class
		// as well as the method, which may be from an interface.
		Class targetClass = (invocation.getThis() != null ? invocation.getThis().getClass() : null);

		// If the transaction attribute is null, the method is non-transactional.
		final TransactionAttribute txAttr =
				getTransactionAttributeSource().getTransactionAttribute(invocation.getMethod(), targetClass);
		final PlatformTransactionManager tm = determineTransactionManager(txAttr);
		final String joinpointIdentification = methodIdentification(invocation.getMethod());

		if (txAttr == null || !(tm instanceof CallbackPreferringPlatformTransactionManager)) {
			// Standard transaction demarcation with getTransaction and commit/rollback calls.
			TransactionInfo txInfo = createTransactionIfNecessary(tm, txAttr, joinpointIdentification);
			Object retVal = null;
			try {
				// This is an around advice: Invoke the next interceptor in the chain.
				// This will normally result in a target object being invoked.
				retVal = invocation.proceed();
			}
			catch (Throwable ex) {
				// target invocation exception
				completeTransactionAfterThrowing(txInfo, ex);
				throw ex;
			}
			finally {
				cleanupTransactionInfo(txInfo);
			}
			commitTransactionAfterReturning(txInfo);
			return retVal;
		}

		else {
			// It's a CallbackPreferringPlatformTransactionManager: pass a TransactionCallback in.
			try {
				Object result = ((CallbackPreferringPlatformTransactionManager) tm).execute(txAttr,
						new TransactionCallback<Object>() {
							public Object doInTransaction(TransactionStatus status) {
								TransactionInfo txInfo = prepareTransactionInfo(tm, txAttr, joinpointIdentification, status);
								try {
									return invocation.proceed();
								}
								catch (Throwable ex) {
									if (txAttr.rollbackOn(ex)) {
										// A RuntimeException: will lead to a rollback.
										if (ex instanceof RuntimeException) {
											throw (RuntimeException) ex;
										}
										else {
											throw new ThrowableHolderException(ex);
										}
									}
									else {
										// A normal return value: will lead to a commit.
										return new ThrowableHolder(ex);
									}
								}
								finally {
									cleanupTransactionInfo(txInfo);
								}
							}
						});

				// Check result: It might indicate a Throwable to rethrow.
				if (result instanceof ThrowableHolder) {
					throw ((ThrowableHolder) result).getThrowable();
				}
				else {
					return result;
				}
			}
			catch (ThrowableHolderException ex) {
				throw ex.getCause();
			}
		}
	}

 

   哦，原来就是在这里控制了method invocation(spring aop是基于method的)!根据我们的配置，来决定

  是不是对method使用事务。

 

   至此，spring的事务管理已经基本清晰了。就是解析出一个advice bean(of class : TransactionInterceptor)来，

   然后在aop中配置pointcut和这个advice,就能产生作用了！

原文: http://michael-softtech.iteye.com/blog/813835 稍作修改