Java中常見的并發(fā)控制手段淺析

更新時(shí)間：2021年08月13日 09:53:25 作者：一灰灰

所謂并發(fā)控制就是幫助我們程序員更容易的讓線程之間合作,讓線程之間相互配合來滿足業(yè)務(wù)邏輯,這篇文章主要給大家介紹了關(guān)于Java中常見的并發(fā)控制手段的相關(guān)資料,需要的朋友可以參考下

前言

單實(shí)例的并發(fā)控制，主要是針對JVM內(nèi)，我們常規(guī)的手段即可滿足需求，常見的手段大概有下面這些

同步代碼塊
CAS自旋
鎖
阻塞隊(duì)列，令牌桶等

1.1 同步代碼塊

通過同步代碼塊，來確保同一時(shí)刻只會(huì)有一個(gè)線程執(zhí)行對應(yīng)的業(yè)務(wù)邏輯，常見的使用姿勢如下

public synchronized doProcess() {
    // 同步代碼塊，只會(huì)有一個(gè)線程執(zhí)行
}

一般推薦使用最小區(qū)間使用原則，盡量不要直接在方法上加synchronized，比如經(jīng)典的雙重判定單例模式

public class Single {
  private static volatile Single instance;
  private Single() {}
  public static Single getInstance() {
      if (instance == null) {
          synchronized(Single.class) {
              if (instance == null) instance = new Single();
          }
      }
      return instance;
  }
}

1.2 CAS自旋方式

比如AtomicXXX原子類中的很多實(shí)現(xiàn)，就是借助unsafe的CAS來實(shí)現(xiàn)的，如下

public final int getAndIncrement() {
    return unsafe.getAndAddInt(this, valueOffset, 1);
}


// unsafe 實(shí)現(xiàn)
// cas + 自選，不斷的嘗試更新設(shè)置，直到成功為止
public final int getAndAddInt(Object var1, long var2, int var4) {
    int var5;
    do {
        var5 = this.getIntVolatile(var1, var2);
    } while(!this.compareAndSwapInt(var1, var2, var5, var5 + var4));

    return var5;
}

1.3 鎖

jdk本身提供了不少的鎖，為了實(shí)現(xiàn)單實(shí)例的并發(fā)控制，我們需要選擇寫鎖；如果支持多讀，單實(shí)例寫，則可以考慮讀寫鎖；一般使用姿勢也比較簡單

private void doSome(ReentrantReadWriteLock.WriteLock writeLock) {
    try {
        writeLock.lock();
        System.out.println("持有鎖成功 " + Thread.currentThread().getName());
        Thread.sleep(1000);
        System.out.println("執(zhí)行完畢! " + Thread.currentThread().getName());
        writeLock.unlock();
    } catch (Exception e) {
        e.printStackTrace();
    }
}

@Test
public void lock() throws InterruptedException {
    ReentrantReadWriteLock reentrantReadWriteLock = new ReentrantReadWriteLock();

    new Thread(()->doSome(reentrantReadWriteLock.writeLock())).start();
    new Thread(()->doSome(reentrantReadWriteLock.writeLock())).start();
    new Thread(()->doSome(reentrantReadWriteLock.writeLock())).start();

    Thread.sleep(20000);
}

1.4 阻塞隊(duì)列

借助同步阻塞隊(duì)列，也可以實(shí)現(xiàn)并發(fā)控制的效果，比如隊(duì)列中初始化n個(gè)元素，每次消費(fèi)從隊(duì)列中獲取一個(gè)元素，如果拿不到則阻塞；執(zhí)行完畢之后，重新塞入一個(gè)元素，這樣就可以實(shí)現(xiàn)一個(gè)簡單版的并發(fā)控制

demo版演示，下面指定隊(duì)列長度為2，表示最大并發(fā)數(shù)控制為2；設(shè)置為1時(shí)，可以實(shí)現(xiàn)單線程的訪問控制

AtomicInteger cnt = new AtomicInteger();

private void consumer(LinkedBlockingQueue<Integer> queue) {
    try {
        // 同步阻塞拿去數(shù)據(jù)
        int val = queue.take();
        Thread.sleep(2000);
        System.out.println("成功拿到: " + val + " Thread: " + Thread.currentThread());
    } catch (InterruptedException e) {
        e.printStackTrace();
    } finally {
        // 添加數(shù)據(jù)
        System.out.println("結(jié)束 " + Thread.currentThread());
        queue.offer(cnt.getAndAdd(1));
    }
}

@Test
public void blockQueue() throws InterruptedException {
    LinkedBlockingQueue<Integer> queue = new LinkedBlockingQueue<>(2);
    queue.add(cnt.getAndAdd(1));
    queue.add(cnt.getAndAdd(1));


    new Thread(() -> consumer(queue)).start();
    new Thread(() -> consumer(queue)).start();
    new Thread(() -> consumer(queue)).start();
    new Thread(() -> consumer(queue)).start();

    Thread.sleep(10000);
}

1.5 信號(hào)量Semaphore

上面隊(duì)列的實(shí)現(xiàn)方式，可以使用信號(hào)量Semaphore來完成，通過設(shè)置信號(hào)量，來控制并發(fā)數(shù)

private void semConsumer(Semaphore semaphore) {
    try {
        //同步阻塞，嘗試獲取信號(hào)
        semaphore.acquire(1);
        System.out.println("成功拿到信號(hào)，執(zhí)行: " + Thread.currentThread());
        Thread.sleep(2000);
        System.out.println("執(zhí)行完畢，釋放信號(hào): " + Thread.currentThread());
        semaphore.release(1);
    } catch (Exception e) {
        e.printStackTrace();
    }
}

@Test
public void semaphore() throws InterruptedException {
    Semaphore semaphore = new Semaphore(2);

    new Thread(() -> semConsumer(semaphore)).start();
    new Thread(() -> semConsumer(semaphore)).start();
    new Thread(() -> semConsumer(semaphore)).start();
    new Thread(() -> semConsumer(semaphore)).start();
    new Thread(() -> semConsumer(semaphore)).start();

    Thread.sleep(20_000);
}

1.6 計(jì)數(shù)器CountDownLatch

計(jì)數(shù)，應(yīng)用場景更偏向于多線程的協(xié)同，比如多個(gè)線程執(zhí)行完畢之后，再處理某些事情；不同于上面的并發(fā)數(shù)的控制，它和柵欄一樣，更多的是行為結(jié)果的統(tǒng)一

這種場景下的使用姿勢一般如下

重點(diǎn)：countDownLatch 計(jì)數(shù)為0時(shí)放行

@Test
public void countDown() throws InterruptedException {
    CountDownLatch countDownLatch = new CountDownLatch(2);

    new Thread(() -> {
        try {
            System.out.println("do something in " + Thread.currentThread());
            Thread.sleep(2000);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } finally {
            countDownLatch.countDown();
        }
    }).start();

    new Thread(() -> {
        try {
            System.out.println("do something in t2: " + Thread.currentThread());
            Thread.sleep(1000);
        } catch (InterruptedException e) {
            e.printStackTrace();
        } finally {
            countDownLatch.countDown();
        }
    }).start();

    countDownLatch.await();
    System.out.printf("結(jié)束");
}

1.7 柵欄 CyclicBarrier

CyclicBarrier的作用與上面的CountDownLatch相似，區(qū)別在于正向計(jì)數(shù)+1, 只有達(dá)到條件才放行; 且支持通過調(diào)用reset()重置計(jì)數(shù)，而CountDownLatch則不行

一個(gè)簡單的demo

private void cyclicBarrierLogic(CyclicBarrier barrier, long sleep) {
    // 等待達(dá)到條件才放行
    try {
        System.out.println("準(zhǔn)備執(zhí)行: " + Thread.currentThread() + " at: " + LocalDateTime.now());
        Thread.sleep(sleep);
        int index = barrier.await();
        System.out.println("開始執(zhí)行: " + index + " thread: " + Thread.currentThread() + " at: " + LocalDateTime.now());
    } catch (Exception e) {
        e.printStackTrace();
    }
}

@Test
public void testCyclicBarrier() throws InterruptedException {
    // 到達(dá)兩個(gè)工作線程才能繼續(xù)往后面執(zhí)行
    CyclicBarrier barrier = new CyclicBarrier(2);
    // 三秒之后，下面兩個(gè)線程的才會(huì)輸出 開始執(zhí)行
    new Thread(() -> cyclicBarrierLogic(barrier, 1000)).start();
    new Thread(() -> cyclicBarrierLogic(barrier, 3000)).start();

    Thread.sleep(4000);
    // 重置，可以再次使用
    barrier.reset();
    new Thread(() -> cyclicBarrierLogic(barrier, 1)).start();
    new Thread(() -> cyclicBarrierLogic(barrier, 1)).start();
    Thread.sleep(10000);
}

1.8 guava令牌桶

guava封裝了非常簡單的并發(fā)控制工具類RateLimiter，作為單機(jī)的并發(fā)控制首選

一個(gè)控制qps為2的簡單demo如下:

private void guavaProcess(RateLimiter rateLimiter) {
    try {
        // 同步阻塞方式獲取
        System.out.println("準(zhǔn)備執(zhí)行: " + Thread.currentThread() + " > " + LocalDateTime.now());
        rateLimiter.acquire();
        System.out.println("執(zhí)行中: " + Thread.currentThread() + " > " + LocalDateTime.now());
    } catch (Exception e) {
        e.printStackTrace();
    }
}

@Test
public void testGuavaRate() throws InterruptedException {
    // 1s 中放行兩個(gè)請求
    RateLimiter rateLimiter = RateLimiter.create(2.0d);
    new Thread(() -> guavaProcess(rateLimiter)).start();
    new Thread(() -> guavaProcess(rateLimiter)).start();
    new Thread(() -> guavaProcess(rateLimiter)).start();
    new Thread(() -> guavaProcess(rateLimiter)).start();
    new Thread(() -> guavaProcess(rateLimiter)).start();
    new Thread(() -> guavaProcess(rateLimiter)).start();
    new Thread(() -> guavaProcess(rateLimiter)).start();

    Thread.sleep(20_000);
}

輸出:

準(zhǔn)備執(zhí)行: Thread[Thread-2,5,main] > 2021-04-13T10:18:05.263
準(zhǔn)備執(zhí)行: Thread[Thread-1,5,main] > 2021-04-13T10:18:05.263
準(zhǔn)備執(zhí)行: Thread[Thread-5,5,main] > 2021-04-13T10:18:05.264
準(zhǔn)備執(zhí)行: Thread[Thread-7,5,main] > 2021-04-13T10:18:05.264
準(zhǔn)備執(zhí)行: Thread[Thread-3,5,main] > 2021-04-13T10:18:05.263
準(zhǔn)備執(zhí)行: Thread[Thread-4,5,main] > 2021-04-13T10:18:05.264
準(zhǔn)備執(zhí)行: Thread[Thread-6,5,main] > 2021-04-13T10:18:05.263
執(zhí)行中: Thread[Thread-2,5,main] > 2021-04-13T10:18:05.267
執(zhí)行中: Thread[Thread-6,5,main] > 2021-04-13T10:18:05.722
執(zhí)行中: Thread[Thread-4,5,main] > 2021-04-13T10:18:06.225
執(zhí)行中: Thread[Thread-3,5,main] > 2021-04-13T10:18:06.721
執(zhí)行中: Thread[Thread-7,5,main] > 2021-04-13T10:18:07.221
執(zhí)行中: Thread[Thread-5,5,main] > 2021-04-13T10:18:07.720
執(zhí)行中: Thread[Thread-1,5,main] > 2021-04-13T10:18:08.219