Spring声明式事务管理源码解读之事务提交

/**
*作者：张荣华(ahuaxuan)
*2007-06-11
*转载请注明出处及作者
*/

简介：上次说到spring声明式事务管理的事务开始部分，按流程来讲，下面应该提交事务了， spring的声明式事务管理其实是比较复杂的，事实上这种复杂性正是由于事务本身的复杂性导致的，如果能用两三句话就把这部分内容说清楚是不现实的，也是不成熟的，而我对这部分的理解也可能是不全面的，还是那句话，希望大家和我一起把本贴的质量提交起来。
在下面的文章中，我讲会多次提到第一篇文章，第一篇文章的地址是：http://www.iteye.com/topic/87426
如果要理解事务提交的话，理解事务开始是一个前提条件,所以请先看第一篇文章，再来看这篇
如果你仔细看下去，我想肯定是有很多收获，因为我们确实能从spring的代码和思想中学到很多东西。

正文：

其实俺的感觉就是事务提交要比事务开始复杂，看事务是否提交我们还是要回到TransactionInterceptor类的invoke方法
Java代码
public Object invoke(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;  
          
        // Create transaction if necessary.  
        TransactionInfo txInfo = createTransactionIfNecessary(invocation.getMethod(), targetClass);  
 
        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  
            doCloseTransactionAfterThrowing(txInfo, ex);  
            throw ex;  
        }  
        finally {  
            doFinally(txInfo);//业务方法出栈后必须先执行的一个方法  
        }  
        doCommitTransactionAfterReturning(txInfo);  
        return retVal;  
    } 

public Object invoke(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;

// Create transaction if necessary.
TransactionInfo txInfo = createTransactionIfNecessary(invocation.getMethod(), targetClass);

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
doCloseTransactionAfterThrowing(txInfo, ex);
throw ex;
}
finally {
doFinally(txInfo);//业务方法出栈后必须先执行的一个方法
}
doCommitTransactionAfterReturning(txInfo);
return retVal;
}
其中的doFinally(txInfo)那一行很重要，也就是说不管如何，这个doFinally方法都是要被调用的，为什么它这么重要呢，举个例子:
我们还是以propregation_required来举例子吧，假设情况是这样的，AService中有一个方法调用了BService中的，这两个方法都处在事务体之中，他们的传播途径都是required。那么调用开始了，AService的方法首先入方法栈，并创建了TransactionInfo的实例，接着BService的方法入栈，又创建了一个TransactionInfo的实例，而重点要说明的是TransactionInfo是一个自身关联的内部类，第二个方法入栈时，会给新创建的TransactionInfo的实例设置一个属性，就是TransactionInfo对象中的private TransactionInfo oldTransactionInfo;属性，这个属性表明BService方法的创建的TransactionInfo对象是有一个old的transactionInfo对象的，这个oldTransactionInfo对象就是AService方法入栈时创建的TransactionInfo对象，我们还记得在createTransactionIfNecessary方法里有这样一个方法吧：
Java代码
protected TransactionInfo createTransactionIfNecessary(Method method, Class targetClass) {  
                // We always bind the TransactionInfo to the thread, even if we didn't create  
        // a new transaction here. This guarantees that the TransactionInfo stack  
        // will be managed correctly even if no transaction was created by this aspect.  
        txInfo.bindToThread();  
        return txInfo;  
    }  
 
就是这个bindToThread()方法在作怪：  
private void bindToThread() {  
            // Expose current TransactionStatus, preserving any existing transactionStatus for  
            // restoration after this transaction is complete.  
            oldTransactionInfo = (TransactionInfo) currentTransactionInfo.get();  
            currentTransactionInfo.set(this);  
        } 

protected TransactionInfo createTransactionIfNecessary(Method method, Class targetClass) {
// We always bind the TransactionInfo to the thread, even if we didn't create
// a new transaction here. This guarantees that the TransactionInfo stack
// will be managed correctly even if no transaction was created by this aspect.
txInfo.bindToThread();
return txInfo;
}

就是这个bindToThread()方法在作怪：
private void bindToThread() {
// Expose current TransactionStatus, preserving any existing transactionStatus for
// restoration after this transaction is complete.
oldTransactionInfo = (TransactionInfo) currentTransactionInfo.get();
currentTransactionInfo.set(this);
}
如果当前线程中已经有了一个TransactionInfo，则拿出来放到新建的transactionInfo对象的oldTransactionInfo属性中，然后再把新建的TransactionInfo设置到当前线程中。

这里有一个概念要搞清楚，就是TransactionInfo对象并不是表明事务状态的对象，表明事务状态的对象是TransactionStatus对象，这个对象同样是TransactionInfo的一个属性（这一点，我在前面一篇文章中并没有讲清楚）。

接下来BService中的那个方法返回，那么该它退栈了，它退栈后要做的就是doFinally方法，即把它的oldTransactionInfo设置到当前线程中（这个TransactionInfo对象显然就是AService方法入栈时创建的，怎么现在又要设置到线程中去呢，原因就是BService的方法出栈时并不提交事务，因为BService的传播途径是required，所以要把栈顶的方法所创建transactioninfo给设置到当前线程中），即调用AService的方法时所创建的TransactionInfo对象。那么在AServie的方法出栈时同样会设置TransactionInfo对象的oldTransactionInfo到当前线程，这时候显然oldTransactionInfo是空的，但AService中的方法会提交事务，所以它的oldTransactionInfo也应该是空了。

在这个小插曲之后，么接下来就应该是到提交事务了，之前在AService的方法出栈时，我们拿到了它入栈时创建的TransactionInfo对象，这个对象中包含了AService的方法事务状态。即TransactionStatus对象，很显然，太显然了，事务提交中的任何属性都和事务开始时的创建的对象息息相关，这个TransactionStatus对象哪里来的，我们再回头看看createTransactionIfNessary方法吧：
Java代码
protected TransactionInfo createTransactionIfNecessary(Method method, Class targetClass) {  
            txInfo.newTransactionStatus(this.transactionManager.getTransaction(txAttr));  
        } 

protected TransactionInfo createTransactionIfNecessary(Method method, Class targetClass) {
txInfo.newTransactionStatus(this.transactionManager.getTransaction(txAttr));
}
再看看transactionManager.getTransaction(txAttr)方法吧：
Java代码
public final TransactionStatus getTransaction(TransactionDefinition definition) throws TransactionException {  
          
        else if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_REQUIRED ||  
                definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_REQUIRES_NEW ||  
            definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_NESTED) {  
            if (debugEnabled) {  
                logger.debug("Creating new transaction with name [" + definition.getName() + "]");  
            }  
            doBegin(transaction, definition);  
            boolean newSynchronization = (this.transactionSynchronization != SYNCHRONIZATION_NEVER);  
            return newTransactionStatus(definition, transaction, true, newSynchronization, debugEnabled, null);//注意这里的返回值，返回的就是一个TransactionStatus对象,这个对象表明了一个事务的状态，比如说是否是一个新的事务，事务是否已经结束，等等，这个对象是非常重要的，在事务提交的时候还是会用到它的。        }  
            }  
    } 

public final TransactionStatus getTransaction(TransactionDefinition definition) throws TransactionException {

else if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_REQUIRED ||
definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_REQUIRES_NEW ||
    definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_NESTED) {
if (debugEnabled) {
logger.debug("Creating new transaction with name [" + definition.getName() + "]");
}
doBegin(transaction, definition);
boolean newSynchronization = (this.transactionSynchronization != SYNCHRONIZATION_NEVER);
return newTransactionStatus(definition, transaction, true, newSynchronization, debugEnabled, null);//注意这里的返回值，返回的就是一个TransactionStatus对象,这个对象表明了一个事务的状态，比如说是否是一个新的事务，事务是否已经结束，等等，这个对象是非常重要的，在事务提交的时候还是会用到它的。 }
}
}
还有一点需要说明的是，AService的方法在执行之前创建的transactionstatus确实是通过这个方法创建的，但是，BService的方法在执行之前创建transactionstatus的方法就与这个不一样了，下面会有详解。

回顾了事务开始时所调用的方法之后，是不是觉得现在对spring如何处理事务越来越清晰了呢。由于这么几个方法的调用，每个方法入栈之前它的事务状态就已经被设置好了。这个事务状态就是为了在方法出栈时被调用而准备的。

让我们再次回到BService中的方法出栈的那个时间段，看看spring都做了些什么，我们知道，后入栈的肯定是先出栈，BService中的方法后入栈，拿它肯定要先出栈了，它出栈的时候是要判断是否要提交事务，释放资源的，让我们来看看TransactionInterceptor的invoke的最后那个方法doCommitTransactionAfterReturning：

Java代码
protected void doCommitTransactionAfterReturning(TransactionInfo txInfo) {  
        if (txInfo != null && txInfo.hasTransaction()) {  
            if (logger.isDebugEnabled()) {  
                logger.debug("Invoking commit for transaction on " + txInfo.joinpointIdentification());  
            }  
            this.transactionManager.commit(txInfo.getTransactionStatus());  
//瞧：提交事务时用到了表明事务状态的那个TransactionStatus对象了。  
        }  
    } 

protected void doCommitTransactionAfterReturning(TransactionInfo txInfo) {
if (txInfo != null && txInfo.hasTransaction()) {
if (logger.isDebugEnabled()) {
logger.debug("Invoking commit for transaction on " + txInfo.joinpointIdentification());
}
this.transactionManager.commit(txInfo.getTransactionStatus());
//瞧：提交事务时用到了表明事务状态的那个TransactionStatus对象了。
}
}
看这个方法的名字就知道spring是要在业务方法出栈时提交事务，貌似很简单，但是事实是这样的吗？ 我们接着往下看。
Java代码
public final void commit(TransactionStatus status) throws TransactionException {  
        DefaultTransactionStatus defStatus = (DefaultTransactionStatus) status;  
 
        if (defStatus.isCompleted()) {  
            throw new IllegalTransactionStateException(  
                    "Transaction is already completed - do not call commit or rollback more than once per transaction");  
        }  
        if (defStatus.isLocalRollbackOnly()) {  
            if (defStatus.isDebug()) {  
                logger.debug("Transactional code has requested rollback");  
            }  
            processRollback(defStatus);  
            return;  
        }  
        if (!shouldCommitOnGlobalRollbackOnly() && defStatus.isGlobalRollbackOnly()) {  
            if (defStatus.isDebug()) {  
                logger.debug("Global transaction is marked as rollback-only but transactional code requested commit");  
            }  
            processRollback(defStatus);  
            throw new UnexpectedRollbackException(  
                    "Transaction has been rolled back because it has been marked as rollback-only");  
        }  
 
        processCommit(defStatus);  
    } 

public final void commit(TransactionStatus status) throws TransactionException {
DefaultTransactionStatus defStatus = (DefaultTransactionStatus) status;

if (defStatus.isCompleted()) {
throw new IllegalTransactionStateException(
"Transaction is already completed - do not call commit or rollback more than once per transaction");
}
if (defStatus.isLocalRollbackOnly()) {
if (defStatus.isDebug()) {
logger.debug("Transactional code has requested rollback");
}
processRollback(defStatus);
return;
}
if (!shouldCommitOnGlobalRollbackOnly() && defStatus.isGlobalRollbackOnly()) {
if (defStatus.isDebug()) {
logger.debug("Global transaction is marked as rollback-only but transactional code requested commit");
}
processRollback(defStatus);
throw new UnexpectedRollbackException(
"Transaction has been rolled back because it has been marked as rollback-only");
}

processCommit(defStatus);
}
上面这段代码就是transactionmanager中的commit，但是看上去，它又把自己的职责分配给别人了，从代码里我们看到，如果事务已经结束了就抛异常，如果事务是rollbackonly的，那么就rollback吧，但是按照正常流程，我们还是想来看一下，事务的提交，就是processCommit(status)这个方法吧。
Java代码
private void processCommit(DefaultTransactionStatus status) throws TransactionException {  
        try {  
            boolean beforeCompletionInvoked = false;  
            try {  
                triggerBeforeCommit(status);  
                triggerBeforeCompletion(status);  
                beforeCompletionInvoked = true;  
                if (status.hasSavepoint()) {  
                    if (status.isDebug()) {  
                        logger.debug("Releasing transaction savepoint");  
                    }  
                    status.releaseHeldSavepoint();  
                }  
                else if (status.isNewTransaction()) {//这个判断非常重要，下面会详细讲解这个判断的作用  
                    if (status.isDebug()) {  
                        logger.debug("Initiating transaction commit");  
                    }  
                    boolean globalRollbackOnly = status.isGlobalRollbackOnly();  
                    doCommit(status);  
                    // Throw UnexpectedRollbackException if we have a global rollback-only  
                    // marker but still didn't get a corresponding exception from commit.  
                    `````````````````````  
    } 

private void processCommit(DefaultTransactionStatus status) throws TransactionException {
try {
boolean beforeCompletionInvoked = false;
try {
triggerBeforeCommit(status);
triggerBeforeCompletion(status);
beforeCompletionInvoked = true;
if (status.hasSavepoint()) {
if (status.isDebug()) {
logger.debug("Releasing transaction savepoint");
}
status.releaseHeldSavepoint();
}
else if (status.isNewTransaction()) {//这个判断非常重要，下面会详细讲解这个判断的作用
if (status.isDebug()) {
logger.debug("Initiating transaction commit");
}
boolean globalRollbackOnly = status.isGlobalRollbackOnly();
doCommit(status);
// Throw UnexpectedRollbackException if we have a global rollback-only
// marker but still didn't get a corresponding exception from commit.
`````````````````````
}
我们注意到，在判断一个事务是否是新事务之前还有一个status.hasSavepoint()的判断，我认为这个判断事实上就是嵌套事务的判断，即判断这个事务是否是嵌套事务，如果不是嵌套事务，则再判断它是否是一个新事务，下面这段话就非常重要了，BService的中的方法是先出栈的，也就是说在调用BService之前的创建的那个事务状态对象在这里要先被判断，但是由于在调用BService的方法之前已经创建了一个Transaction和Session（假设我们使用的是hibernate3），这时候在创建第二个TransactionInfo（再强调一下吧，TransactionInfo并不是Transaction，Transaction是真正的事务对象，TransactionInfo只不过是一个辅助类而已，用来记录一系列状态的辅助类）的TransactionStatus的时候就会进入下面这个方法（当然在这之前会判断一下当前线程中是否已经有了一个SessionHolder对象，不清楚SessionHolder作用的同学情况第一篇文章），这个方法其实应该放到第一篇文章中讲的，但是想到如果不讲事务提交就讲这个方法好像没有这么贴切，废话少说，我们来看一下吧：
Java代码
private TransactionStatus handleExistingTransaction(  
            TransactionDefinition definition, Object transaction, boolean debugEnabled)  
            throws TransactionException {  
 
        if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_NEVER) {  
            throw new IllegalTransactionStateException(  
                    "Transaction propagation 'never' but existing transaction found");  
        }  
 
        if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_NOT_SUPPORTED) {  
            if (debugEnabled) {  
                logger.debug("Suspending current transaction");  
            }  
            Object suspendedResources = suspend(transaction);  
            boolean newSynchronization = (this.transactionSynchronization == SYNCHRONIZATION_ALWAYS);  
            return newTransactionStatus(  
                    definition, null, false, newSynchronization, debugEnabled, suspendedResources);  
        }  
 
        if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_REQUIRES_NEW) {  
            if (debugEnabled) {  
                logger.debug("Suspending current transaction, creating new transaction with name [" +  
                        definition.getName() + "]");  
            }  
            Object suspendedResources = suspend(transaction);  
            doBegin(transaction, definition);  
            boolean newSynchronization = (this.transactionSynchronization != SYNCHRONIZATION_NEVER);  
            return newTransactionStatus(  
                    definition, transaction, true, newSynchronization, debugEnabled, suspendedResources);  
        }  
 
        if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_NESTED) {  
            if (!isNestedTransactionAllowed()) {  
                throw new NestedTransactionNotSupportedException(  
                        "Transaction manager does not allow nested transactions by default - " +  
                        "specify 'nestedTransactionAllowed' property with value 'true'");  
            }  
            if (debugEnabled) {  
                logger.debug("Creating nested transaction with name [" + definition.getName() + "]");  
            }  
            if (useSavepointForNestedTransaction()) {  
                // Create savepoint within existing Spring-managed transaction,  
                // through the SavepointManager API implemented by TransactionStatus.  
                // Usually uses JDBC 3.0 savepoints. Never activates Spring synchronization.  
                DefaultTransactionStatus status =  
                        newTransactionStatus(definition, transaction, false, false, debugEnabled, null);  
                status.createAndHoldSavepoint();  
                return status;  
            }  
            else {  
                // Nested transaction through nested begin and commit/rollback calls.  
                // Usually only for JTA: Spring synchronization might get activated here  
                // in case of a pre-existing JTA transaction.  
                doBegin(transaction, definition);  
                boolean newSynchronization = (this.transactionSynchronization != SYNCHRONIZATION_NEVER);  
                return newTransactionStatus(definition, transaction, true, newSynchronization, debugEnabled, null);  
            }  
        }  
 
        // Assumably PROPAGATION_SUPPORTS.  
        if (debugEnabled) {  
            logger.debug("Participating in existing transaction");  
        }  
        boolean newSynchronization = (this.transactionSynchronization != SYNCHRONIZATION_NEVER);  
        return newTransactionStatus(definition, transaction, false, newSynchronization, debugEnabled, null);  
    } 

private TransactionStatus handleExistingTransaction(
TransactionDefinition definition, Object transaction, boolean debugEnabled)
throws TransactionException {

if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_NEVER) {
throw new IllegalTransactionStateException(
"Transaction propagation 'never' but existing transaction found");
}

if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_NOT_SUPPORTED) {
if (debugEnabled) {
logger.debug("Suspending current transaction");
}
Object suspendedResources = suspend(transaction);
boolean newSynchronization = (this.transactionSynchronization == SYNCHRONIZATION_ALWAYS);
return newTransactionStatus(
definition, null, false, newSynchronization, debugEnabled, suspendedResources);
}

if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_REQUIRES_NEW) {
if (debugEnabled) {
logger.debug("Suspending current transaction, creating new transaction with name [" +
definition.getName() + "]");
}
Object suspendedResources = suspend(transaction);
doBegin(transaction, definition);
boolean newSynchronization = (this.transactionSynchronization != SYNCHRONIZATION_NEVER);
return newTransactionStatus(
definition, transaction, true, newSynchronization, debugEnabled, suspendedResources);
}

if (definition.getPropagationBehavior() == TransactionDefinition.PROPAGATION_NESTED) {
if (!isNestedTransactionAllowed()) {
throw new NestedTransactionNotSupportedException(
"Transaction manager does not allow nested transactions by default - " +
"specify 'nestedTransactionAllowed' property with value 'true'");
}
if (debugEnabled) {
logger.debug("Creating nested transaction with name [" + definition.getName() + "]");
}
if (useSavepointForNestedTransaction()) {
// Create savepoint within existing Spring-managed transaction,
// through the SavepointManager API implemented by TransactionStatus.
// Usually uses JDBC 3.0 savepoints. Never activates Spring synchronization.
DefaultTransactionStatus status =
newTransactionStatus(definition, transaction, false, false, debugEnabled, null);
status.createAndHoldSavepoint();
return status;
}
else {
// Nested transaction through nested begin and commit/rollback calls.
// Usually only for JTA: Spring synchronization might get activated here
// in case of a pre-existing JTA transaction.
doBegin(transaction, definition);
boolean newSynchronization = (this.transactionSynchronization != SYNCHRONIZATION_NEVER);
return newTransactionStatus(definition, transaction, true, newSynchronization, debugEnabled, null);
}
}

// Assumably PROPAGATION_SUPPORTS.
if (debugEnabled) {
logger.debug("Participating in existing transaction");
}
boolean newSynchronization = (this.transactionSynchronization != SYNCHRONIZATION_NEVER);
return newTransactionStatus(definition, transaction, false, newSynchronization, debugEnabled, null);
}
我们看到这个方法其实很明了，就是什么样的传播途径就创建什么样的transactionstatus，这个方法是在事务开始时被调用的，拿到我们之前举的例子中来看下，我们就恍然大悟了，原来，如果之前已经创建过事务，那个这个新建的transactionstauts就不应该是属于一个newTransaction了，所以第3个参数就是false了。

也就是说，在BService的方法出栈要要执行processcommit，但是由于BService的那个TransactionStatus不是一个newTransaction，所以它根本不会触发这个动作：
Java代码
else if (status.isNewTransaction()) {//这个判断非常重要，下面会详细讲解这个判断的作用  
                    if (status.isDebug()) {  
                        logger.debug("Initiating transaction commit");  
                    }  
boolean globalRollbackOnly = status.isGlobalRollbackOnly();  
                    doCommit(status);  
} 

else if (status.isNewTransaction()) {//这个判断非常重要，下面会详细讲解这个判断的作用
if (status.isDebug()) {
logger.debug("Initiating transaction commit");
}
boolean globalRollbackOnly = status.isGlobalRollbackOnly();
doCommit(status);
}
也就是说在BService的方法出栈后，事务是不会提交的。这完全符合propragation_required的模型。
而在AService的方法出栈后，AService的方法所对应的那个TransactionStatus对象的newTransaction属性是为true的，即它会触发上面这段代码，进行真正的事务提交。让我们回想一下AService方法入栈之前创建TransactionStatus对象的情形吧：
newTransactionStatus(definition, transaction, true, newSynchronization, debugEnabled, null);看到第3个参数为true没有。

那么事务该提交了吧，事务的提交我想使用过hibernate的人都知道怎么提交了：
txObject.getSessionHolder().getTransaction().commit();
从当前线程中拿到SessionHolder，再拿到开始事务的那个Transaction对象，然后再commit事务。在没有用spring之前，我们经常这么做。呵呵。

好吧，我已经说到了spring声明式事务管理的70%到80%的内容了，这70%到80%的内容看上去还是非常容易理解的，如果把这两篇文章认真看过，我相信会有所收获的，剩下的内容需要靠大家自己去挖掘了，因为另剩下的内容可是需要花费很多时间的，因为牵扯的东西实在是太多了，呵呵。最后祝大家阅读愉快，因为我的文笔实在是让大家的眼睛受罪了。

作者：张荣华，未经作者同意不得随意转载！