Spring多线程事务处理

一、背景

本文主要介绍了spring多线程事务的解决方案，心急的可以跳过上面的理论介绍分析部分直接看最终解决方案。

在我们日常的业务活动中，经常会出现大规模的修改插入操作，比如在3.0的活动赛事创建，涉及到十几张表的插入（一张表可能插入一行或者多行数据），由于单线程模型的关系，所有的sql都是串行，即后面的sql必须都要等到前面的sql执行完成才能继续。但是在很多场景下，sql的执行顺序并不影响业务的结果，面对这样的场景，我们很自然的想到了使用异步的方式去处理，可是我们同时又希望整个创建操作是事务性的，即要全部成功，要么全部失败，但是单纯的使用异步线程并不能达到我们理想的效果。

这个时候，我们需要一种多线程下保证事务的解决方案。

代码片段，大量的同步保存操作

    public void much(){
        //业务操作1
        doBusiness1();
        //业务操作2
        doBusiness2();
        //业务操作3
        doBusiness3();
        //业务操作4
        doBusiness4();

    }

    private void doBusiness1() {
        //执行sql1
        //执行sql2
        //执行sql3
        //执行sql4
    }

每个业务操作可以是相关联的，也有可能是完全无关的，但如果做成异步的话我们就无法保证事务，怎么去解决这个问题呢？

二、理论先行

1.事务介绍

我们先确定spring事务的本质是什么，spring本身不支持事务，spring实现事务只是对我们原有的业务逻辑做了一层包装，他替我们决定了什么时候开启事务，什么情况下应该向数据库提交，什么时候回滚，及实现我们设置的一些事务参数，包括回滚的条件，传播类型等。

我们所熟知的spring事务有两种主流的解决方式，一种是声明式事务，一种是编程式事务。

先来讲我们最常用的声明式事务。

1.1声明式事务

声明式事务就是我们最常用的@Transactional注解，通常我们只需要在我们想交由spring控制的事务方法上加上注解即可，这个注解有一些重要的参数，由于不是本文重点，就不在此展开。这是一个经典的spring的aop实现，为了弄清楚在加上@Transactional注解后spring到底为我们做了什么，我们可以从两方面入手，一是spring如何给我们生成相应的代理对象，二是这个代理对象为我们做了什么。

事务的开始是由@EnableTransactionManagement 注解产生，这个注解在运行时会导入TransactionManagementConfigurationSelector这个类，这个类本质上是一个ImportSelector，他根据adviceMode将特定的配置类导入进去，分别为AutoProxyRegistrar 后置处理器和ProxyTransactionManagementConfiguration Advisor。

AutoProxyRegistrar 实现了ImportBeanDefinitionRegistrar 重写了registerBeanDefinitions 方法

public void registerBeanDefinitions(AnnotationMetadata importingClassMetadata, BeanDefinitionRegistry registry) {
    boolean candidateFound = false;
    Set<String> annTypes = importingClassMetadata.getAnnotationTypes();
    for (String annType : annTypes) {
        // ...

        AopConfigUtils.registerAutoProxyCreatorIfNecessary(registry);
    }
    // ...
}

@Nullable
public static BeanDefinition registerAutoProxyCreatorIfNecessary(
    BeanDefinitionRegistry registry, @Nullable Object source) {

    return registerOrEscalateApcAsRequired(InfrastructureAdvisorAutoProxyCreator.class, registry, source);
}

。该方法最终注入了InfrastructureAdvisorAutoProxyCreator。InfrastructureAdvisorAutoProxyCreator这个类就是一个bean的后置处理器，最终的作用就是处理需要的代理对象。

public Object postProcessAfterInitialization(@Nullable Object bean, String beanName) {
if (bean != null) {
    Object cacheKey = getCacheKey(bean.getClass(), beanName);
    if (this.earlyProxyReferences.remove(cacheKey) != bean) {
        return wrapIfNecessary(bean, beanName, cacheKey);
    }
}
return bean;
}

protected Object wrapIfNecessary(Object bean, String beanName, Object cacheKey) {
    // ...
    // 拿当前bean去匹配容器中的 Advisors，如果找到符合的就生成代理对象
    // Create proxy if we have advice.
    Object[] specificInterceptors = getAdvicesAndAdvisorsForBean(bean.getClass(), beanName, null);
    if (specificInterceptors != DO_NOT_PROXY) {
        this.advisedBeans.put(cacheKey, Boolean.TRUE);
        Object proxy = createProxy(
            bean.getClass(), beanName, specificInterceptors, new SingletonTargetSource(bean));
        this.proxyTypes.put(cacheKey, proxy.getClass());
        return proxy;
    }

    this.advisedBeans.put(cacheKey, Boolean.FALSE);
    return bean;
}

ProxyTransactionManagementConfiguration的作用就是来生成具体的Advisor，他注册了三个bean，

1. 该类主要完成以下几个任务：
2. 创建TransactionAttributeSource对象：用于解析@Transactional注解并生成事务属性。
3. 创建TransactionInterceptor对象：用于创建事务通知，将事务属性应用到目标方法，这其实就是一个事务模板，如下所示

protected Object invokeWithinTransaction(Method method, @Nullable Class<?> targetClass,
                                         final InvocationCallback invocation) throws Throwable {
    //TransactionAttributeSource内部保存着当前类某个方法对应的TransactionAttribute---事务属性源
    //可以看做是一个存放TransactionAttribute与method方法映射的池子
    TransactionAttributeSource tas = getTransactionAttributeSource();
    //获取当前事务方法对应的TransactionAttribute
    final TransactionAttribute txAttr = (tas != null ? tas.getTransactionAttribute(method, targetClass) : null);
    //定位TransactionManager
    final TransactionManager tm = determineTransactionManager(txAttr);
    .....
    //类型转换为局部事务管理器
    PlatformTransactionManager ptm = asPlatformTransactionManager(tm);
    final String joinpointIdentification = methodIdentification(method, targetClass, txAttr);

    if (txAttr == null || !(ptm instanceof CallbackPreferringPlatformTransactionManager)) {
        //TransactionManager根据TransactionAttribute创建事务后返回
        //TransactionInfo封装了当前事务的信息--包括TransactionStatus
        TransactionInfo txInfo = createTransactionIfNecessary(ptm, txAttr, joinpointIdentification);

        Object retVal;
        try {
            //继续执行过滤器链---过滤链最终会调用目标方法
            //因此可以理解为这里是调用目标方法
            retVal = invocation.proceedWithInvocation();
        }
        catch (Throwable ex) {
            //目标方法抛出异常则进行判断是否需要回滚
            completeTransactionAfterThrowing(txInfo, ex);
            throw ex;
        }
        finally {
            //清除当前事务信息
            cleanupTransactionInfo(txInfo);
        }
        ...
        //正常返回,那么就正常提交事务呗(当然还是需要判断TransactionStatus状态先)
        commitTransactionAfterReturning(txInfo);
        return retVal;
    }
        ...

1. 创建TransactionAdvisor对象：将事务通知和切点（Pointcut）组合成Advisor。
2. 创建TransactionAttributeSourceAdvisor对象：将事务属性和切点组合成Advisor

@Configuration(proxyBeanMethods = false)
@Role(BeanDefinition.ROLE_INFRASTRUCTURE)
public class ProxyTransactionManagementConfiguration extends AbstractTransactionManagementConfiguration {

    @Bean(name = TransactionManagementConfigUtils.TRANSACTION_ADVISOR_BEAN_NAME)
    @Role(BeanDefinition.ROLE_INFRASTRUCTURE)
    public BeanFactoryTransactionAttributeSourceAdvisor transactionAdvisor(
        TransactionAttributeSource transactionAttributeSource, TransactionInterceptor transactionInterceptor) {

        BeanFactoryTransactionAttributeSourceAdvisor advisor = new BeanFactoryTransactionAttributeSourceAdvisor();
        advisor.setTransactionAttributeSource(transactionAttributeSource);
        advisor.setAdvice(transactionInterceptor);
        if (this.enableTx != null) {
            advisor.setOrder(this.enableTx.<Integer>getNumber("order"));
        }
        return advisor;
    }

    @Bean
    @Role(BeanDefinition.ROLE_INFRASTRUCTURE)
    public TransactionAttributeSource transactionAttributeSource() {
        // TransactionAttributeSource 是一个接口，具体注入的是 Annotationxxxx
        return new AnnotationTransactionAttributeSource();
    }

    @Bean
    @Role(BeanDefinition.ROLE_INFRASTRUCTURE)
    public TransactionInterceptor transactionInterceptor(TransactionAttributeSource transactionAttributeSource) {
        TransactionInterceptor interceptor = new TransactionInterceptor();
        interceptor.setTransactionAttributeSource(transactionAttributeSource);
        if (this.txManager != null) {
            interceptor.setTransactionManager(this.txManager);
        }
        return interceptor;
    }
}

观察TransactionAdvisor代码实现@SuppressWarnings("serial")
public class BeanFactoryTransactionAttributeSourceAdvisor extends AbstractBeanFactoryPointcutAdvisor {

@Nullable
private TransactionAttributeSource transactionAttributeSource;

private final TransactionAttributeSourcePointcut pointcut = new TransactionAttributeSourcePointcut() {
    @Override
    @Nullable
    protected TransactionAttributeSource getTransactionAttributeSource() {
        return transactionAttributeSource;
    }
};

/**
 * Set the transaction attribute source which is used to find transaction
 * attributes. This should usually be identical to the source reference
 * set on the transaction interceptor itself.
 * @see TransactionInterceptor#setTransactionAttributeSource
 */
public void setTransactionAttributeSource(TransactionAttributeSource transactionAttributeSource) {
this.transactionAttributeSource = transactionAttributeSource;
}

/**
 * Set the {@link ClassFilter} to use for this pointcut.
 * Default is {@link ClassFilter#TRUE}.
 */
public void setClassFilter(ClassFilter classFilter) {
    this.pointcut.setClassFilter(classFilter);
}

@Override
public Pointcut getPointcut() {
    return this.pointcut;
}

可以见到里面已经包含了pointcut，这就能将我们需要被增加的事务方法找出。

ProxyTransactionManagementConfiguration负责将需要包装的bean和方法找出并包装成advisor，InfrastructureAdvisorAutoProxyCreator根据advisor生成相应的代理对象。

小结：InfrastructureAdvisorAutoProxyCreator遍历容器中的bean，尝试去自动代理，匹配的工作就交由advisor中的point，如果匹配成功就为其创建代理对象，这个代理对象中放入了TransactionInterceptor拦截器

，等到相关方法调用时，调用的是代理对象的方法，然后通过责任链模式通过TransactionInterceptor处理，以此来进行事务的操作。

声明式事务的介绍先到这里，接下来我们来介绍下编程式事务。

1.2编程式事务

编程式事务的核心就是将spring为我们做好的那些步骤拆出来，交由开发者去控制事务何时开启、提交、回滚，他的运行本质和声明式事务并没有两样。

模板如下

public class TransactionMain {
    public static void main(String[] args) throws ClassNotFoundException, SQLException {
        test();
    }

    private static void test() {
        DataSource dataSource = getDS();
        JdbcTransactionManager jtm = new JdbcTransactionManager(dataSource);
        //JdbcTransactionManager根据TransactionDefinition信息来进行一些连接属性的设置
        //包括隔离级别和传播行为等
        DefaultTransactionDefinition transactionDef = new DefaultTransactionDefinition();
        //开启一个新事务---此时autocommit已经被设置为了false,并且当前没有事务,这里创建的是一个新事务
        TransactionStatus ts = jtm.getTransaction(transactionDef);
        //进行业务逻辑操作
        try {
            update(dataSource);
            jtm.commit(ts);
        }catch (Exception e){
            jtm.rollback(ts);
            System.out.println("发生异常,我已回滚");
        }
    }

    private static void update(DataSource dataSource) throws Exception {
        JdbcTemplate jt = new JdbcTemplate();
        jt.setDataSource(dataSource);
        jt.update("UPDATE Department SET Dname=\"大忽悠\" WHERE id=6");
        throw new Exception("我是来捣乱的");
    }

}

三、方案探索

1.直接使用多线程

我们在开启代码中事务，并在业务逻辑中直接使用多线程，是否能保证事务？

  @Transactional
    public void testDirect() {
        new Thread(()->{
            Per per = new Per();
            per.setName("t1");
            perService.save(per);
        }).start();
        new Thread(()->{
            Per per1 = new Per();
            per1.setName("t2");
            perService.save(per1);
            throw new RuntimeException("Exception test");
        }).start();
    }

显然，这种方式并不能保证事务，哪怕加上了事务注解，因为子线程抛出的异常并不能在主线程中捕获，也不能被其他线程感知到。

2.事务模板中使用多线程

package com.user.util;

import lombok.RequiredArgsConstructor;
import org.springframework.jdbc.datasource.DataSourceTransactionManager;
import org.springframework.stereotype.Component;
import org.springframework.transaction.TransactionStatus;
import org.springframework.transaction.support.DefaultTransactionDefinition;

import javax.sql.DataSource;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.Executor;
import java.util.concurrent.atomic.AtomicBoolean;

   @Component
   @RequiredArgsConstructor
   public class MultiplyThreadTransactionManager {
   /**

    * 如果是多数据源的情况下,需要指定具体是哪一个数据源
      */
      private final DataSource dataSource;

      public void runAsyncButWaitUntilAllDown(List<Runnable> tasks, Executor executor) {
      if(executor==null){
          throw new IllegalArgumentException("线程池不能为空");
      }
      DataSourceTransactionManager transactionManager = getTransactionManager();
      //是否发生了异常
      AtomicBoolean ex=new AtomicBoolean();

      List<CompletableFuture> taskFutureList=new ArrayList<>(tasks.size());
      List<TransactionStatus> transactionStatusList=new ArrayList<>(tasks.size());

      tasks.forEach(task->{
          taskFutureList.add(CompletableFuture.runAsync(
                  () -> {
                      try{
                          //1.开启新事务
                          transactionStatusList.add(openNewTransaction(transactionManager));
                          //2.异步任务执行
                          task.run();
                      }catch (Throwable throwable){
                          //打印异常
                          throwable.printStackTrace();
                          //其中某个异步任务执行出现了异常,进行标记
                          ex.set(Boolean.TRUE);
                          //其他任务还没执行的不需要执行了
                          taskFutureList.forEach(completableFuture -> completableFuture.cancel(true));
                      }
                  }
                  , executor)
          );
      });

      try {
          //阻塞直到所有任务全部执行结束---如果有任务被取消,这里会抛出异常滴,需要捕获
          CompletableFuture.allOf(taskFutureList.toArray(new CompletableFuture[]{})).get();
      } catch (InterruptedException | ExecutionException e) {
          e.printStackTrace();
      }

      //发生了异常则进行回滚操作,否则提交
      if(ex.get()){
          System.out.println("发生异常,全部事务回滚");
          transactionStatusList.forEach(transactionManager::rollback);
      }else {
          System.out.println("全部事务正常提交");
          transactionStatusList.forEach(transactionManager::commit);
      }
      }

   private TransactionStatus openNewTransaction(DataSourceTransactionManager transactionManager) {
       //JdbcTransactionManager根据TransactionDefinition信息来进行一些连接属性的设置
       //包括隔离级别和传播行为等
       DefaultTransactionDefinition transactionDef = new DefaultTransactionDefinition();
       //开启一个新事务---此时autocommit已经被设置为了false,并且当前没有事务,这里创建的是一个新事务
       return transactionManager.getTransaction(transactionDef);
   }

   private DataSourceTransactionManager getTransactionManager() {
       return new DataSourceTransactionManager(dataSource);
   }
   }

测试

public void test(){
    List<Runnable> tasks=new ArrayList<>();

    tasks.add(()->{
            Per per = new Per();
            per.setName("t1");
            perService.save(per);
    });

    tasks.add(()->{
            Per per = new Per();
            per.setName("t2");
            perService.save(per);
    });

    multiplyThreadTransactionManager.runAsyncButWaitUntilAllDown(tasks, Executors.newCachedThreadPool());

}

执行结果

java.lang.IllegalStateException: No value for key [HikariDataSource (HikariPool-1)] bound to thread
	at org.springframework.transaction.support.TransactionSynchronizationManager.unbindResource(TransactionSynchronizationManager.java:198) ~[spring-tx-5.3.10.jar:5.3.10]
	at org.springframework.jdbc.datasource.DataSourceTransactionManager.doCleanupAfterCompletion(DataSourceTransactionManager.java:371) ~[spring-jdbc-5.3.10.jar:5.3.10]
	at org.springframework.transaction.support.AbstractPlatformTransactionManager.cleanupAfterCompletion(AbstractPlatformTransactionManager.java:992) ~[spring-tx-5.3.10.jar:5.3.10]
	at org.springframework.transaction.suppoAbstractPlatformTransactionrt.AbstractPlatformTransactionManager.processCommit(AbstractPlatformTransactionManager

结果报了这个错，这个错误信息室找不到绑定在线程上的key为HikariDataSource的资源，因为事务资源都是绑定在线程上的，当事务提交或者回滚时，他需要寻找绑定在当前线程上的资源，如果找不到，就会报错。

原理剖析：

首先我们找到绑定线程资源的关键方法org.springframework.transaction.support.TransactionSynchronizationManager#bindResource

	/**
	 * Bind the given resource for the given key to the current thread.
	 * @param key the key to bind the value to (usually the resource factory)
	 * @param value the value to bind (usually the active resource object)
	 * @throws IllegalStateException if there is already a value bound to the thread
	 * @see ResourceTransactionManager#getResourceFactory()
	 */
	public static void bindResource(Object key, Object value) throws IllegalStateException {
		Object actualKey = TransactionSynchronizationUtils.unwrapResourceIfNecessary(key);
		Assert.notNull(value, "Value must not be null");
		Map<Object, Object> map = resources.get();
		// set ThreadLocal Map if none found
		if (map == null) {
			map = new HashMap<>();
			resources.set(map);
		}
		Object oldValue = map.put(actualKey, value);
		// Transparently suppress a ResourceHolder that was marked as void...
		if (oldValue instanceof ResourceHolder && ((ResourceHolder) oldValue).isVoid()) {
			oldValue = null;
		}
		if (oldValue != null) {
			throw new IllegalStateException(
					"Already value [" + oldValue + "] for key [" + actualKey + "] bound to thread");
		}
	}

根据debug会发现，spring在开启事务时会自动为我们调用这个方法，绑定key为HikariDataSource，value为ConnectionHolder到threadlocal中。第二次sql执行时会绑定key为DefaultSqlSessionFactory，value为DefaultSqlSessionFactory。

既然讲到了事务资源的绑定时机，下面就顺便讲一下这两种资源在何时释放。我们再回顾一下事务的执行流程及机制。spring处理事务的原理就是基于aop，每个需要实现事务的方法都要通过TransactionInterceptor这个拦截器，通过这个拦截器去实现事务增强。

	@Override
	@Nullable
	public Object invoke(MethodInvocation invocation) throws Throwable {
		// Work out the target class: may be {@code null}.
		// The TransactionAttributeSource should be passed the target class
		// as well as the method, which may be from an interface.
		Class<?> targetClass = (invocation.getThis() != null ? AopUtils.getTargetClass(invocation.getThis()) : null);

		// Adapt to TransactionAspectSupport's invokeWithinTransaction...
		//重点，这里注册了一个回调，最后会调回下面
       //父类实现
        return invokeWithinTransaction(invocation.getMethod(), targetClass, new CoroutinesInvocationCallback() {
			@Override
			@Nullable
			public Object proceedWithInvocation() throws Throwable {
                //原始方法
				return invocation.proceed();
			}
			@Override
			public Object getTarget() {
				return invocation.getThis();
			}
			@Override
			public Object[] getArguments() {
				return invocation.getArguments();
			}
		});
	}

他最终会交给他的父类模板类org.springframework.transaction.interceptor.TransactionAspectSupport#invokeWithinTransaction实现，在来看一下这个类为我们处理了什么，直接看重点

	/**
	 * General delegate for around-advice-based subclasses, delegating to several other template
	 * methods on this class. Able to handle {@link CallbackPreferringPlatformTransactionManager}
	 * as well as regular {@link PlatformTransactionManager} implementations and
	 * {@link ReactiveTransactionManager} implementations for reactive return types.
	 * @param method the Method being invoked
	 * @param targetClass the target class that we're invoking the method on
	 * @param invocation the callback to use for proceeding with the target invocation
	 * @return the return value of the method, if any
	 * @throws Throwable propagated from the target invocation
	 */
	@Nullable
	protected Object invokeWithinTransaction(Method method, @Nullable Class<?> targetClass,
			final InvocationCallback invocation) throws Throwable {

		// If the transaction attribute is null, the method is non-transactional.
        //说人话就是获取事务资源，装配事务管理器
		TransactionAttributeSource tas = getTransactionAttributeSource();
		final TransactionAttribute txAttr = (tas != null ? tas.getTransactionAttribute(method, targetClass) : null);
		final TransactionManager tm = determineTransactionManager(txAttr);

		PlatformTransactionManager ptm = asPlatformTransactionManager(tm);
		final String joinpointIdentification = methodIdentification(method, targetClass, txAttr);

		if (txAttr == null || !(ptm instanceof CallbackPreferringPlatformTransactionManager)) {
			// Standard transaction demarcation with getTransaction and commit/rollback calls.
            //事务相关信息，包括传播级别，什么异常下回滚等
			TransactionInfo txInfo = createTransactionIfNecessary(ptm, txAttr, joinpointIdentification);

			Object retVal;
			try {
				// This is an around advice: Invoke the next interceptor in the chain.
				// This will normally result in a target object being invoked.
                //就是他的子类注册的回调，真正的业务逻辑
				retVal = invocation.proceedWithInvocation();
			}
			catch (Throwable ex) {
				// target invocation exception
                //回滚
				completeTransactionAfterThrowing(txInfo, ex);
				throw ex;
			}
			finally {
                //清理事务信息
				cleanupTransactionInfo(txInfo);
			}

			if (retVal != null && vavrPresent && VavrDelegate.isVavrTry(retVal)) {
				// Set rollback-only in case of Vavr failure matching our rollback rules...
				TransactionStatus status = txInfo.getTransactionStatus();
				if (status != null && txAttr != null) {
					retVal = VavrDelegate.evaluateTryFailure(retVal, txAttr, status);
				}
			}
            //提交
			commitTransactionAfterReturning(txInfo);
			return retVal;
		}

        ...

	}

继续往下，观察commitTransactionAfterReturning(txInfo)的实现，发现这一步是由事务管理器完成的。

	protected void commitTransactionAfterReturning(@Nullable TransactionInfo txInfo) {
		if (txInfo != null && txInfo.getTransactionStatus() != null) {
			if (logger.isTraceEnabled()) {
				logger.trace("Completing transaction for [" + txInfo.getJoinpointIdentification() + "]");
			}
			txInfo.getTransactionManager().commit(txInfo.getTransactionStatus());
		}
	}

进而可以推断，rollback操作也是同样的道理，有兴趣的小伙伴可以自己debug一下，继续走下去，观察事务管理器为我们做了什么。

@Override
	public final void commit(TransactionStatus status) throws TransactionException {
		if (status.isCompleted()) {
			throw new IllegalTransactionStateException(
					"Transaction is already completed - do not call commit or rollback more than once per transaction");
		}

		DefaultTransactionStatus defStatus = (DefaultTransactionStatus) status;
		if (defStatus.isLocalRollbackOnly()) {
			if (defStatus.isDebug()) {
				logger.debug("Transactional code has requested rollback");
			}
			processRollback(defStatus, false);
			return;
		}

		if (!shouldCommitOnGlobalRollbackOnly() && defStatus.isGlobalRollbackOnly()) {
			if (defStatus.isDebug()) {
				logger.debug("Global transaction is marked as rollback-only but transactional code requested commit");
			}
			processRollback(defStatus, true);
			return;
		}

		processCommit(defStatus);
	}

到此，spring的整个事务流程就已经非常清晰了，我们想要实现多事务管理的方法也找到了，难就是去控制事务的资源。只要我们拿到了相应的事务资源，然后在创建自己的事务管理器控制事务何时提交或者回滚，这样我们就可以实现一个多线程同时提交回滚，类似于二阶段提交的操作，来达到多线程事务的统一。

3.多线程事务管理器

不多说，直接上代码看最终版本

package com.controller;

import lombok.Builder;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import org.slf4j.MDC;
import org.springframework.jdbc.datasource.DataSourceTransactionManager;
import org.springframework.stereotype.Component;
import org.springframework.transaction.TransactionStatus;
import org.springframework.transaction.support.DefaultTransactionDefinition;
import org.springframework.transaction.support.TransactionSynchronization;
import org.springframework.transaction.support.TransactionSynchronizationManager;
import org.springframework.util.CollectionUtils;

import javax.sql.DataSource;
import java.util.*;
import java.util.concurrent.*;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicInteger;

/**
 * 多线程事务管理
 */
@Component
@Slf4j
@RequiredArgsConstructor
public class MultiplyThreadTransactionManager {
    /**
     * 如果是多数据源的情况下,需要指定具体是哪一个数据源
     */
    private final DataSource dataSource;

    private final static ThreadLocal<Boolean> immediatelyCommitFlag = new ThreadLocal<>();

    private final static ThreadLocal<List<TransactionStatus>> transactionStatusListThreadLocal = new ThreadLocal<>();

    private final static ThreadLocal<List<TransactionResource>> transactionResourcesthreadLocal = new ThreadLocal<>();

    private final static ThreadLocal<Map<Object, Object>> mainNativeResourceThreadLocal = new ThreadLocal<>();

    /**
     * 多线程下事务执行
     *
     * @param tasks             任务列表
     * @param immediatelyCommit 是否需要立即提交
     */
    public List<CompletableFuture> runAsyncButWaitUntilAllDown(List<Runnable> tasks,  Boolean immediatelyCommit) {

        Executor executor = Executors.newCachedThreadPool();

        DataSourceTransactionManager transactionManager = getTransactionManager();
        //是否发生了异常
        AtomicBoolean ex = new AtomicBoolean();

        List<CompletableFuture> taskFutureList = new CopyOnWriteArrayList<>();
        List<TransactionStatus> transactionStatusList = new CopyOnWriteArrayList<>();
        List<TransactionResource> transactionResources = new CopyOnWriteArrayList<>();

        //记录原生主事务资源
        //这一步可能在原生sql执行前，也可能在原生sql执行后，所以这个资源可能不够充分，需要在下面继续处理
        //如果返回的是原资源集合的引用，下面一步可以不用
        Map<Object, Object> resourceMap = TransactionSynchronizationManager.getResourceMap();
        if (!CollectionUtils.isEmpty(resourceMap)) {
            mainNativeResourceThreadLocal.set(new HashMap<>(resourceMap));
        }
        Map<String, String> copyOfContextMap = MDC.getCopyOfContextMap();
        Executor finalExecutor = executor;
        AtomicInteger atomicInteger = new AtomicInteger(0);
        tasks.forEach(task -> {
            taskFutureList.add(CompletableFuture.runAsync(
                    () -> {
                        log.info("任务开始");
                        try {
                            //1.开启新事务
                            TransactionStatus transactionStatus = openNewTransaction(transactionManager);
                            log.info("开启新事务 successfully");
                            transactionStatusList.add(transactionStatus);
                            atomicInteger.incrementAndGet();
                            System.out.println("atomicInteger.get()"+atomicInteger.incrementAndGet());
                            System.out.println(transactionStatus);
                            //2.异步任务执行
                            task.run();
                            log.info("异步任务执行 successfully");
                            //3.继续事务资源复制,因为在sql执行是会产生新的资源对象
                            transactionResources.add(TransactionResource.copyTransactionResource());

                        } catch (Throwable throwable) {
                            log.info("任务执行异常"+throwable.getMessage());
                            log.error("任务执行异常",throwable);
                            //其中某个异步任务执行出现了异常,进行标记
                            ex.set(Boolean.TRUE);
                            //其他任务还没执行的不需要执行了
                            taskFutureList.forEach(completableFuture -> completableFuture.cancel(true));
                        }
                    }
                    , finalExecutor)
            );
        });

        try {
            //阻塞直到所有任务全部执行结束---如果有任务被取消,这里会抛出异常滴,需要捕获
            CompletableFuture.allOf(taskFutureList.toArray(new CompletableFuture[]{})).get();
        } catch (InterruptedException | ExecutionException e) {
            log.info("任务被取消");
            log.error("任务被取消",e);

        }

        //发生了异常则进行回滚操作,否则提交
        if (ex.get()) {
            log.info("发生异常,全部事务回滚");
            for (int i = 0; i < transactionStatusList.size(); i++) {
                transactionResources.get(i).autoWiredTransactionResource();
                Map<Object, Object> rollBackResourceMap = TransactionSynchronizationManager.getResourceMap();
                log.info("回滚前事务资源size{}，本身{}",rollBackResourceMap.size(),rollBackResourceMap);
                transactionManager.rollback(transactionStatusList.get(i));
                transactionResources.get(i).removeTransactionResource();
            }
        } else {
            if (immediatelyCommit) {
                log.info("全部事务正常提交");
                for (int i = 0; i < transactionStatusList.size(); i++) {
                    //transactionResources.get(i).autoWiredTransactionResource();
                    Map<Object, Object> commitResourceMap = TransactionSynchronizationManager.getResourceMap();
                    log.info("提交前事务资源size{}，本身{}",commitResourceMap.size(),commitResourceMap);
                    transactionManager.commit(transactionStatusList.get(i));
                    transactionResources.get(i).removeTransactionResource();
                }
            } else {
                //缓存全部待提交数据
                immediatelyCommitFlag.set(immediatelyCommit);
                transactionResourcesthreadLocal.set(transactionResources);
                transactionStatusListThreadLocal.set(transactionStatusList);
            }
        }

        //交还给主事务
        if (immediatelyCommit) {
            mainTransactionResourceBack(!ex.get());
        }

        return taskFutureList;
    }

    public void multiplyThreadTransactionCommit() {
        try {
            Boolean immediatelyCommit = immediatelyCommitFlag.get();
            if (immediatelyCommit) {
                throw new IllegalStateException("immediatelyCommit cant call multiplyThreadTransactionCommit");
            }
            //提交
            //获取存储的事务资源和状态
            List<TransactionResource> transactionResources = transactionResourcesthreadLocal.get();
            List<TransactionStatus> transactionStatusList = transactionStatusListThreadLocal.get();
            if (CollectionUtils.isEmpty(transactionResources) || CollectionUtils.isEmpty(transactionStatusList)) {
                throw new IllegalStateException("transactionResources or transactionStatusList is empty");
            }
            //重新提交
            DataSourceTransactionManager transactionManager = getTransactionManager();
            log.info("全部事务正常提交");
            for (int i = 0; i < transactionStatusList.size(); i++) {
                transactionResources.get(i).autoWiredTransactionResource();
                Map<Object, Object> commitResourceMap = TransactionSynchronizationManager.getResourceMap();
                log.info("提交前事务资源size{}，本身{}",commitResourceMap.size(),commitResourceMap);
                transactionManager.commit(transactionStatusList.get(i));
                transactionResources.get(i).removeTransactionResource();
            }
        } catch (Exception e) {
            mainTransactionResourceBack(false);
            log.error("multiplyThreadTransactionCommit fail", e);
        } finally {
            transactionResourcesthreadLocal.remove();
            transactionStatusListThreadLocal.remove();
            immediatelyCommitFlag.remove();
        }

        //交还给主事务
        mainTransactionResourceBack(true);
    }

    //主线程事务资源返还
    public void mainTransactionResourceBack(Boolean subTransactionSuccess) {

        if (CollectionUtils.isEmpty(mainNativeResourceThreadLocal.get())) {
            //清除数据
            mainNativeResourceThreadLocal.remove();
            return;
        }
        Map<Object, Object> nativeResource = new HashMap<>(mainNativeResourceThreadLocal.get());
        Map<Object, Object> resourceMap = TransactionSynchronizationManager.getResourceMap();
        log.info("当前线程资事务源size{}--------------------------------{}",resourceMap.size(), resourceMap);
        log.info("原生线程事务资源size{}--------------------------------{}",nativeResource.size(), nativeResource);
        //已经被绑定的资源不能重复绑定
        if (!CollectionUtils.isEmpty(resourceMap)) {
            for (Object o : resourceMap.keySet()) {
                if (nativeResource.containsKey(o)) {
                    nativeResource.remove(o);
                }
            }
        }

        nativeResource.forEach((k,v)->{
            if (!(k instanceof DataSource)){
                log.info("nativeResource 没有 DataSource");
            }
        });
        //交还不能绑定factory
        nativeResource.forEach((k,v)->{
            if (k instanceof DataSource){
                TransactionSynchronizationManager.bindResource(k,v);
            }
        });
        Map<Object, Object> finResource = TransactionSynchronizationManager.getResourceMap();
        log.info("主线程最终事务源size{}--------------------------------{}",finResource.size(), finResource);
        //防止未激活事务
        if (!TransactionSynchronizationManager.isSynchronizationActive()) {
            TransactionSynchronizationManager.initSynchronization();
        }
        //清除数据
        mainNativeResourceThreadLocal.remove();
        if (!subTransactionSuccess) {
            throw new RuntimeException("子事务失败，需要回滚");
        }
    }

    private TransactionStatus openNewTransaction(DataSourceTransactionManager transactionManager) {
        //JdbcTransactionManager根据TransactionDefinition信息来进行一些连接属性的设置
        //包括隔离级别和传播行为等
        DefaultTransactionDefinition transactionDef = new DefaultTransactionDefinition();
        //开启一个新事务---此时autocommit已经被设置为了false,并且当前没有事务,这里创建的是一个新事务
        return transactionManager.getTransaction(transactionDef);
    }

    private DataSourceTransactionManager getTransactionManager() {
        return new DataSourceTransactionManager(dataSource);
    }

    /**
     * 保存当前事务资源,用于线程间的事务资源COPY操作
     */
    @Builder
    private static class TransactionResource {
        //事务结束后默认会移除集合中的DataSource作为key关联的资源记录
        private Map<Object, Object> resources = new HashMap<>();

        //下面五个属性会在事务结束后被自动清理,无需我们手动清理
        private Set<TransactionSynchronization> synchronizations = new HashSet<>();

        private String currentTransactionName;

        private Boolean currentTransactionReadOnly;

        private Integer currentTransactionIsolationLevel;

        private Boolean actualTransactionActive;

        public static TransactionResource copyTransactionResource() {
            return TransactionResource.builder()
                    //返回的是不可变集合，这里为了更加灵活，copy出一个集合过来
                    .resources(new HashMap<>(TransactionSynchronizationManager.getResourceMap()))
                    //如果需要注册事务监听者,这里记得修改--我们这里不需要,就采用默认负责--spring事务内部默认也是这个值
                    .synchronizations(new LinkedHashSet<>())
                    .currentTransactionName(TransactionSynchronizationManager.getCurrentTransactionName())
                    .currentTransactionReadOnly(TransactionSynchronizationManager.isCurrentTransactionReadOnly())
                    .currentTransactionIsolationLevel(TransactionSynchronizationManager.getCurrentTransactionIsolationLevel())
                    .actualTransactionActive(TransactionSynchronizationManager.isActualTransactionActive())
                    .build();
        }

        //装配事务资源，为提交/回滚做储备
        public void autoWiredTransactionResource() {
            //获取当前线程事务资源
            Map<Object, Object> resourceMap = TransactionSynchronizationManager.getResourceMap();
            for (Object o : resourceMap.keySet()) {
                if (resourceMap.containsKey(o)) {
                    //移除重复事务资源key，避免绑定报错
                    resources.remove(o);
                }

            }
            boolean synchronizationActive = TransactionSynchronizationManager.isSynchronizationActive();
            //绑定事务资源，注意 绑定是绑定到当前主线程上，记得最后释放交换主线程，再由主线程收回原有事务自选
            resources.forEach(TransactionSynchronizationManager::bindResource);
            //如果需要注册事务监听者,这里记得修改--我们这里不需要,就采用默认负责--spring事务内部默认也是这个值
            //避免重复激活或者事务未激活
            if (!synchronizationActive) {
                TransactionSynchronizationManager.initSynchronization();
            }

            TransactionSynchronizationManager.setActualTransactionActive(actualTransactionActive);
            TransactionSynchronizationManager.setCurrentTransactionName(currentTransactionName);
            TransactionSynchronizationManager.setCurrentTransactionIsolationLevel(currentTransactionIsolationLevel);
            TransactionSynchronizationManager.setCurrentTransactionReadOnly(currentTransactionReadOnly);
        }

        public void removeTransactionResource() {
            Map<Object, Object> resourceMap = new HashMap<>(TransactionSynchronizationManager.getResourceMap());

            //事务结束后默认会移除集合中的DataSource作为key关联的资源记录
            //DataSource如果重复移除,unbindResource时会因为不存在此key关联的事务资源而报错
            resources.keySet().forEach(key -> {
                if (resourceMap.containsKey(key)) {
                    TransactionSynchronizationManager.unbindResource(key);
                }
            });
        }
    }
}

验证

    @Transactional
    public String test(Integer par) {
        log.info("get(" + par + ")");
        if (par == 3 || par == 5 || par == 6) {
            Per per2 = new Per();
            per2.setName("t3");
            per2.setGrou(Thread.currentThread().getName());
            perService.save(per2);
        }

        List<Runnable> list = new ArrayList<>();
        list.add(() -> {
            Per per = new Per();
            per.setName("t1");
            per.setGrou(Thread.currentThread().getName());
            log.info("任务开始save");
            perService.save(per);
            log.info("任务完成save");
            if (par == 1) {
                throw new RuntimeException();
            }
        });
        list.add(() -> {
            Per per1 = new Per();
            per1.setName("t2");
            per1.setGrou(Thread.currentThread().getName());
            log.info("任务开始save");
            perService.save(per1);
            log.info("任务完成save");
            if (par == 2) {
                throw new RuntimeException();
            }

        });

        log.info("runAsyncButWaitUntilAllDown start");
        multiplyThreadTransactionManager.runAsyncButWaitUntilAllDown(list, false);

        if (par == 4 || par == 5 || par == 6) {
            Per per3 = new Per();
            per3.setName("t4");
            per3.setGrou(Thread.currentThread().getName());
            perService.save(per3);
            if (par == 6) {
                throw new RuntimeException();
            }
        }
        log.info("multiplyThreadTransactionCommit start");
        multiplyThreadTransactionManager.multiplyThreadTransactionCommit();

        return "ss";
    }

2024-03-11 16:00:22.048  INFO 44900 --- [       Thread-1] c.c.MultiplyThreadTransactionManager     : 提交前事务资源size2，本身{org.apache.ibatis.session.defaults.DefaultSqlSessionFactory@3c1f842f=org.mybatis.spring.SqlSessionHolder@f0c0875, HikariDataSource (HikariPool-1)=org.springframework.jdbc.datasource.ConnectionHolder@3b3508e5}
2024-03-11 16:00:22.141  INFO 44900 --- [       Thread-1] c.c.MultiplyThreadTransactionManager     : 当前线程资事务源size0--------------------------------{}
2024-03-11 16:00:22.141  INFO 44900 --- [       Thread-1] c.c.MultiplyThreadTransactionManager     : 原生线程事务资源size2--------------------------------{org.apache.ibatis.session.defaults.DefaultSqlSessionFactory@3c1f842f=org.mybatis.spring.SqlSessionHolder@493bb8e6, HikariDataSource (HikariPool-1)=org.springframework.jdbc.datasource.ConnectionHolder@3950dc5b}
2024-03-11 16:00:22.141  INFO 44900 --- [       Thread-1] c.c.MultiplyThreadTransactionManager     : nativeResource 没有 DataSource
2024-03-11 16:00:22.141  INFO 44900 --- [       Thread-1] c.c.MultiplyThreadTransactionManager     : 主线程最终事务源size1--------------------------------{HikariDataSource (HikariPool-1)=org.springframework.jdbc.datasource.ConnectionHolder@3950dc5b}

有兴趣的小伙伴可以做进一步测试。

在本公司项目组中利用这个机制优化现有的业务，性能提升了约70%。

比起网上常见的多线程事务管理器，主要做了如下增强

1.支持在已存在事务下运行。

在很多场景下，我们可能会遇到多线程事务外还存在其他事物的场景下，我们需要支持兼容多种事务环境。

2.支持自定义提交时机。

有时候我们不希望事务立马提交，希望他能够和外围事务保持一致，这时候可以将runAsyncButWaitUntilAllDown的immediatelyCommit参数写成false，并手动调用multiplyThreadTransactionCommit方法去主动提交。

我们需要注意的地方，任何事都是有舍有得的，耗时的显著降低是因为利用了更多的资源，比如线程资源和数据库连接资源，尤其是数据库连接资源，更是额外宝贵，我们一定要合理评估我们的每一项决策是否有意义，风险和回报是否成正比。

还有一点需要注意，在极高并发的情况下，多线程事务容易造成死锁，因为当主事务开启的情况下，他要为他下面的子线程事务开启连接，当连接不够时就容易造成循环等待。一个比较好的做法是提前获得所有连接，并设置一个合理的超时时间

如果有小伙伴遇到了其他相关疑问，或者使用此代码发现了问题，欢迎留言讨论，共同进步。

站在巨人的肩膀上，在解决这一问题时，我也参考了很多网上其他的文章。我不否认我借鉴了许多，但是终究是在此基础上有所突破，如果有感兴趣的小伙伴可以去看一下，在此附上地址。

https://www.cnblogs.com/hefeng2014/p/17759000.html

https://d9bp4nr5ye.feishu.cn/wiki/OJdiwdYeXirkdBk3NV8c5evrnmh