CountDownLatch和CyclicBarrier：如何让多线程步调一致?

CountDownLatch和CyclicBarrier主要解决线程的调度问题。例如如下场景:订单的对账，订单系统完成的订单和已经派发的订单的对账，对于有异常的订单要修复。这里是3个操作，见下面伪代码

Order order=getOrder();

POrder porder=getPOder();

checkAndFix(order,porder);

如果每天的订单量很大这就会很耗时，我们可以考虑串行的方法，可以用线程池的方法同时去查询俩个订单，然后在check这时候我们要解决线程调度的问题，CountDownLatch就排上了用场,代码如下

CountDownLatch countDownLatch = new CountDownLatch(2);
ExecutorService pool = Executors.newFixedThreadPool(2);
AtomicReference<Object> order = new AtomicReference<>();
pool.execute(() -> {
    order.set(getOrder());
    countDownLatch.countDown();
});

AtomicReference<Object> porder = new AtomicReference<>();
pool.execute(() -> {
    porder.set(getPOrder());
    countDownLatch.countDown();
});

try {
    countDownLatch.await();
} catch (InterruptedException e) {
    e.printStackTrace();
}
checkAndFix(order.get(),porder.get());

不过这里checkAndFix还是主线程串行的能否在进一步优化呢，答案是能这里用到CyclicBarrier

ExecutorService callBack = Executors.newFixedThreadPool(1);
        Vector<Object> orders=new Vector<>();
        Vector<Object> porders=new Vector<>();
        CyclicBarrier cyclicBarrier = new CyclicBarrier(2,()->{
            callBack.execute(()->{
                checkAndFix( orders.get(0),porders.get(0));
            });
        });

        while(true){
            ExecutorService pool = Executors.newFixedThreadPool(2);
            pool.execute(() -> {
                orders.add(getOrder());
                try {
                    cyclicBarrier.await();
                } catch (InterruptedException | BrokenBarrierException e) {
                    e.printStackTrace();
                }
            });
            pool.execute(() -> {
                porders.add(getPOrder());
                try {
                    cyclicBarrier.await();
                } catch (InterruptedException | BrokenBarrierException e) {
                    e.printStackTrace();
                }
            });
        }

CyclicBarrier和CountDownLatch的区别是可重入的，当见到0还会从2开始。另外CyclicBarrier可以设置Callback