最近接受到接口优化的任务,查看代码逻辑后发现在批量处理数据耗时长,想到使用多线程处理批量数据,又要保持原来的事务一致性。
@Component
@Slf4j
public class MultiThreadingTransactionManager {
/**
* 数据源事务管理器
*/
@Autowired
private DataSourceTransactionManager dataSourceTransactionManager;
@Autowired
private ThreadPoolTaskExecutor executorService;
private long timeout = 120;
/**
* 用于判断子线程业务是否处理完成
* 处理完成时threadCountDownLatch的值为0
*/
private CountDownLatch threadCountDownLatch;
/**
* 用于等待子线程全部完成后,子线程统一进行提交和回滚
* 进行提交和回滚时mainCountDownLatch的值为0
*/
private final CountDownLatch mainCountDownLatch = new CountDownLatch(1);
/**
* 是否提交事务,默认是true,当子线程有异常发生时,设置为false,回滚事务
*/
private final AtomicBoolean isSubmit = new AtomicBoolean(true);
public boolean execute(List<Runnable> runnableList,String factorySchema) {
isSubmit.set(true);
setThreadCountDownLatch(runnableList.size());
runnableList.forEach(runnable -> executorService.execute(() -> executeThread(factorySchema,runnable, threadCountDownLatch, mainCountDownLatch, isSubmit)));
// 等待子线程全部执行完毕
try {
// 若计数器变为零了,则返回 true
boolean isFinish = threadCountDownLatch.await(timeout, TimeUnit.SECONDS);
if (!isFinish) {
// 如果还有为执行完成的就回滚
isSubmit.set(false);
log.info("存在子线程在预期时间内未执行完毕,任务将全部回滚");
}
} catch (Exception exception) {
log.info("主线程发生异常,异常为: " + exception.getMessage());
} finally {
// 计数器减1,代表该主线程执行完毕
mainCountDownLatch.countDown();
}
// 返回结果,是否执行成功,事务提交即为执行成功,事务回滚即为执行失败
return isSubmit.get();
}
private void executeThread(String factorySchema,Runnable runnable, CountDownLatch threadCountDownLatch, CountDownLatch mainCountDownLatch, AtomicBoolean isSubmit) {
log.info("子线程: [" + Thread.currentThread().getName() + "]");
// 判断别的子线程是否已经出现错误,错误别的线程已经出现错误,那么所有的都要回滚,这个子线程就没有必要执行了
if (!isSubmit.get()) {
log.info("整个事务中有子线程执行失败需要回滚, 子线程: [" + Thread.currentThread().getName() + "] 终止执行");
// 计数器减1,代表该子线程执行完毕
threadCountDownLatch.countDown();
return;
}
//动态数据源切换
SchemaContextHolder.setSchema(factorySchema);
// 开启事务
DefaultTransactionDefinition defaultTransactionDefinition = new DefaultTransactionDefinition();
TransactionStatus transactionStatus = dataSourceTransactionManager.getTransaction(defaultTransactionDefinition);
try {
// 执行业务逻辑
runnable.run();
} catch (Exception exception) {
// 发生异常需要进行回滚,设置isSubmit为false
isSubmit.set(false);
log.info("子线程: [" + Thread.currentThread().getName() + "]执行业务发生异常,异常为: " + exception.getMessage());
} finally {
// 计数器减1,代表该子线程执行完毕
threadCountDownLatch.countDown();
}
try {
// 等待主线程执行
mainCountDownLatch.await();
} catch (Exception exception) {
log.info("子线程: [" + Thread.currentThread().getName() + "]等待提交或回滚异常,异常为: " + exception.getMessage());
}
try {
// 提交
if (isSubmit.get()) {
dataSourceTransactionManager.commit(transactionStatus);
log.info("子线程: [" + Thread.currentThread().getName() + "]进行事务提交");
} else {
dataSourceTransactionManager.rollback(transactionStatus);
log.info("子线程: [" + Thread.currentThread().getName() + "]进行事务回滚");
}
} catch (Exception exception) {
log.info("子线程: [" + Thread.currentThread().getName() + "]进行事务提交或回滚出现异常,异常为:" + exception.getMessage());
}
}
private void setThreadCountDownLatch(int num) {
this.threadCountDownLatch = new CountDownLatch(num);
}
}
@RestController
@RequestMapping("test")
public class TestController {
@Autowired
TestService testService;
@Autowired
MultiThreadingTransactionManager multiThreadingTransactionManager;
@RequestMapping("test")
public String test(){
List<TestBean> list = new ArrayList<>();
list.add(new TestBean("2",1));
list.add(new TestBean("3",2));
List<Runnable> runnableList = new ArrayList<>();
list.forEach(testBean -> runnableList.add(() -> {
testService.insert(testBean);
}));
boolean isSuccess = multiThreadingTransactionManager.execute(runnableList,"db9771");
System.out.println(isSuccess);
return "ok";
};
}
大体思路,就是所有子线程在各自线程内开启事务,执行业务逻辑后,判断是否抛错,一旦抛错,会把全局AtomicBoolean置为false,因为其具有原子性所以不会有线程不安全问题。所有子线程完业务代码会等待主线程,全部子线程执行业务结束后,主线程等待结束,判断AtomicBoolean是什么状态,一旦false,所有子线程回滚,否则提交。