Java并发包中的同步队列SynchronousQueue实现原理

转自: http://ifeve.com/java-synchronousqueue/

介绍

Java 6的并发编程包中的SynchronousQueue是一个没有数据缓冲的BlockingQueue，生产者线程对其的插入操作put必须等待消费者的移除操作take，反过来也一样。

不像ArrayBlockingQueue或LinkedListBlockingQueue，SynchronousQueue内部并没有数据缓存空间，你不能调用peek()方法来看队列中是否有数据元素，因为数据元素只有当你试着取走的时候才可能存在，不取走而只想偷窥一下是不行的，当然遍历这个队列的操作也是不允许的。队列头元素是第一个排队要插入数据的线程，而不是要交换的数据。数据是在配对的生产者和消费者线程之间直接传递的，并不会将数据缓冲数据到队列中。可以这样来理解：生产者和消费者互相等待对方，握手，然后一起离开。

SynchronousQueue的一个使用场景是在线程池里。Executors.newCachedThreadPool()就使用了SynchronousQueue，这个线程池根据需要（新任务到来时）创建新的线程，如果有空闲线程则会重复使用，线程空闲了60秒后会被回收。

实现原理

阻塞队列的实现方法有许多：

阻塞算法实现

阻塞算法实现通常在内部采用一个锁来保证多个线程中的put()和take()方法是串行执行的。采用锁的开销是比较大的，还会存在一种情况是线程A持有线程B需要的锁，B必须一直等待A释放锁，即使A可能一段时间内因为B的优先级比较高而得不到时间片运行。所以在高性能的应用中我们常常希望规避锁的使用。

01
public class NativeSynchronousQueue<E> {

02
    boolean putting = false;

03
    E item = null;

04
 
05
    public synchronized E take() throws InterruptedException {

06
        while (item == null)

07
            wait();

08
        E e = item;

09
        item = null;

10
        notifyAll();

11
        return e;

12
    }

13
 
14
    public synchronized void put(E e) throws InterruptedException {

15
        if (e==null) return;

16
        while (putting)

17
            wait();

18
        putting = true;

19
        item = e;

20
        notifyAll();

21
        while (item!=null)

22
            wait();

23
        putting = false;

24
        notifyAll();

25
    }

26
}

信号量实现

经典同步队列实现采用了三个信号量，代码很简单，比较容易理解：

01
public class SemaphoreSynchronousQueue<E> {

02
    E item = null;

03
    Semaphore sync = new Semaphore(0);

04
    Semaphore send = new Semaphore(1);

05
    Semaphore recv = new Semaphore(0);

06
 
07
    public E take() throws InterruptedException {

08
        recv.acquire();

09
        E x = item;

10
        sync.release();

11
        send.release();

12
        return x;

13
    }

14
 
15
    public void put (E x) throws InterruptedException{

16
        send.acquire();

17
        item = x;

18
        recv.release();

19
        sync.acquire();

20
    }

21
}

在多核机器上，上面方法的同步代价仍然较高，操作系统调度器需要上千个时间片来阻塞或唤醒线程，而上面的实现即使在生产者put()时已经有一个消费者在等待的情况下，阻塞和唤醒的调用仍然需要。

Java 5实现

01
public class Java5SynchronousQueue<E> {

02
    ReentrantLock qlock = new ReentrantLock();

03
    Queue waitingProducers = new Queue();

04
    Queue waitingConsumers = new Queue();

05
 
06
    static class Node extends AbstractQueuedSynchronizer {

07
        E item;

08
        Node next;

09
 
10
        Node(Object x) { item = x; }

11
        void waitForTake() { /* (uses AQS) */ }

12
           E waitForPut() { /* (uses AQS) */ }

13
    }

14
 
15
    public E take() {

16
        Node node;

17
        boolean mustWait;

18
        qlock.lock();

19
        node = waitingProducers.pop();

20
        if(mustWait = (node == null))

21
           node = waitingConsumers.push(null);

22
         qlock.unlock();

23
 
24
        if (mustWait)

25
           return node.waitForPut();

26
        else

27
            return node.item;

28
    }

29
 
30
    public void put(E e) {

31
         Node node;

32
         boolean mustWait;

33
         qlock.lock();

34
         node = waitingConsumers.pop();

35
         if (mustWait = (node == null))

36
             node = waitingProducers.push(e);

37
         qlock.unlock();

38
 
39
         if (mustWait)

40
             node.waitForTake();

41
         else

42
            node.item = e;

43
    }

44
}

Java 5的实现相对来说做了一些优化，只使用了一个锁，使用队列代替信号量也可以允许发布者直接发布数据，而不是要首先从阻塞在信号量处被唤醒。

Java6实现

Java 6的SynchronousQueue的实现采用了一种性能更好的无锁算法 — 扩展的“Dual stack and Dual queue”算法。性能比Java5的实现有较大提升。竞争机制支持公平和非公平两种：非公平竞争模式使用的数据结构是后进先出栈(Lifo Stack)；公平竞争模式则使用先进先出队列（Fifo Queue），性能上两者是相当的，一般情况下，Fifo通常可以支持更大的吞吐量，但Lifo可以更大程度的保持线程的本地化。

代码实现里的Dual Queue或Stack内部是用链表(LinkedList)来实现的，其节点状态为以下三种情况：

持有数据 – put()方法的元素
持有请求 – take()方法
空

这个算法的特点就是任何操作都可以根据节点的状态判断执行，而不需要用到锁。

其核心接口是Transfer，生产者的put或消费者的take都使用这个接口，根据第一个参数来区别是入列（栈）还是出列（栈）。

01
/**

02
    * Shared internal API for dual stacks and queues.

03
    */

04
   static abstract class Transferer {

05
       /**

06
        * Performs a put or take.

07
        *

08
        * @param e if non-null, the item to be handed to a consumer;

09
        *          if null, requests that transfer return an item

10
        *          offered by producer.

11
        * @param timed if this operation should timeout

12
        * @param nanos the timeout, in nanoseconds

13
        * @return if non-null, the item provided or received; if null,

14
        *         the operation failed due to timeout or interrupt --

15
        *         the caller can distinguish which of these occurred

16
        *         by checking Thread.interrupted.

17
        */

18
       abstract Object transfer(Object e, boolean timed, long nanos);

19
   }

TransferQueue实现如下(摘自Java 6源代码)，入列和出列都基于Spin和CAS方法：

01
/**

02
    * Puts or takes an item.

03
    */

04
   Object transfer(Object e, boolean timed, long nanos) {

05
       /* Basic algorithm is to loop trying to take either of

06
        * two actions:

07
        *

08
        * 1. If queue apparently empty or holding same-mode nodes,

09
        *    try to add node to queue of waiters, wait to be

10
        *    fulfilled (or cancelled) and return matching item.

11
        *

12
        * 2. If queue apparently contains waiting items, and this

13
        *    call is of complementary mode, try to fulfill by CAS'ing

14
        *    item field of waiting node and dequeuing it, and then

15
        *    returning matching item.

16
        *

17
        * In each case, along the way, check for and try to help

18
        * advance head and tail on behalf of other stalled/slow

19
        * threads.

20
        *

21
        * The loop starts off with a null check guarding against

22
        * seeing uninitialized head or tail values. This never

23
        * happens in current SynchronousQueue, but could if

24
        * callers held non-volatile/final ref to the

25
        * transferer. The check is here anyway because it places

26
        * null checks at top of loop, which is usually faster

27
        * than having them implicitly interspersed.

28
        */

29
 
30
       QNode s = null; // constructed/reused as needed

31
       boolean isData = (e != null);

32
 
33
       for (;;) {

34
           QNode t = tail;

35
           QNode h = head;

36
           if (t == null || h == null)         // saw uninitialized value

37
               continue;                       // spin

38
 
39
           if (h == t || t.isData == isData) { // empty or same-mode

40
               QNode tn = t.next;

41
               if (t != tail)                  // inconsistent read

42
                   continue;

43
               if (tn != null) {               // lagging tail

44
                   advanceTail(t, tn);

45
                   continue;

46
               }

47
               if (timed &amp;&amp; nanos &lt;= 0)        // can't wait

48
                   return null;

49
               if (s == null)

50
                   s = new QNode(e, isData);

51
               if (!t.casNext(null, s))        // failed to link in

52
                   continue;

53
 
54
               advanceTail(t, s);              // swing tail and wait

55
               Object x = awaitFulfill(s, e, timed, nanos);

56
               if (x == s) {                   // wait was cancelled

57
                   clean(t, s);

58
                   return null;

59
               }

60
 
61
               if (!s.isOffList()) {           // not already unlinked

62
                   advanceHead(t, s);          // unlink if head

63
                   if (x != null)              // and forget fields

64
                       s.item = s;

65
                   s.waiter = null;

66
               }

67
               return (x != null)? x : e;

68
 
69
           } else {                            // complementary-mode

70
               QNode m = h.next;               // node to fulfill

71
               if (t != tail || m == null || h != head)

72
                   continue;                   // inconsistent read

73
 
74
               Object x = m.item;

75
               if (isData == (x != null) ||    // m already fulfilled

76
                   x == m ||                   // m cancelled

77
                   !m.casItem(x, e)) {         // lost CAS

78
                   advanceHead(h, m);          // dequeue and retry

79
                   continue;

80
               }

81
 
82
               advanceHead(h, m);              // successfully fulfilled

83
               LockSupport.unpark(m.waiter);

84
               return (x != null)? x : e;

85
           }

86
       }

87
   }