Java Thread 那些事

这篇文章被压在草稿箱许久，最近公司内部的技术社区有同学贴出了几篇分享 Java线程的文章，发觉有很多概念没有讲清楚，所以花点时间继续撰写，便有了这篇博文。 本文只聚焦 JVM 层面的线程模型，不考虑和真实的操作系统 Thread 模型挂钩（由于篇幅有限，本文不会介绍Thread dump结构，也不会介绍调优过程中对工具的综合使用，如ps,top,iostat,jstack,TDA plugin,Thread inspector.如果有问题，欢迎大家留言交流）。后面会考虑对 xUnix 和 Windows 平台的线程 / 进程模型进行深入分析，也希望大家能够喜欢。

ok，言归正传。上图：

Java的线程状态一共有NEW，RUNNABLE，BLOCKED，WAITING，TIMED_WAITING，TERMINATED 6种状态。这里重点关注一下BLOCKED和TIMED_WAITING状态。

BLOCKED状态：线程进入此状态的前提一般有两个：waiting for monitor（intrinsic or external） entry 或者 reenter 同步代码块。讲到这我们先了解一下Java线程模型中的两个队列。如图所示：

每个 Monitor在某个时刻，只能被一个线程拥有，该线程就是 “Active Thread”，而其它线程都是 “Waiting Thread”，分别在两个队列 “Entry Set”和 “Wait Set”里面等候。在 “Entry Set”中等待的线程状态是 “Waiting for monitor entry”，而在 “Wait Set”中等待的线程状态是 “in Object.wait()”。如果你不恰当的使用了ReentrantLock或者ReentrantReadWriteLock类，就有可能陷入BLOCKED状态，这个也是我们调优中经常会遇到的情况，解决方案也很简单，找到等待上锁的地址，分析是否发生了Thread starvation。
         至于TIME_WAITING状态，官方文档也讲解的比较好，即你在调用下面方法时，线程会进入该状态。

          Thread.sleep
	  Object.wait with timeout
	  Thread.join with timeout
	  LockSupport.parkNanos
	  LockSupport.parkUntil

这里重点关注一下LockSupport，该类是用来创建锁和其他同步类的基本线程阻塞原语，是一个针对Thread.suspend和Thread.resume()的优化，也是针对忙等，防止过度自旋的一种优化（关于这一点，感兴趣的同学可以参阅一下文献5）。

ok，在简单介绍完几个重点的线程状态后，我们通过几个具体的case来看了解下Thread stack：

Case 1：NIO 中的Acceptor

"qtp589745448-36 Acceptor0 SelectChannelConnector@0.0.0.0:8161" prio=10 tid=0x00007f02f8eea800 nid=0x18ee runnable [0x00007f02e70b3000]
   java.lang.Thread.State: RUNNABLE
	at sun.nio.ch.ServerSocketChannelImpl.accept0(Native Method)
	at sun.nio.ch.ServerSocketChannelImpl.accept(ServerSocketChannelImpl.java:241)
	- locked <0x00000000ec8ffde8> (a java.lang.Object)
	at org.eclipse.jetty.server.nio.SelectChannelConnector.accept(SelectChannelConnector.java:109)
	at org.eclipse.jetty.server.AbstractConnector$Acceptor.run(AbstractConnector.java:938)
	at org.eclipse.jetty.util.thread.QueuedThreadPool.runJob(QueuedThreadPool.java:608)
	at org.eclipse.jetty.util.thread.QueuedThreadPool$3.run(QueuedThreadPool.java:543)
	at java.lang.Thread.run(Thread.java:724)

   Locked ownable synchronizers:
	- None

瞅瞅源代码中是怎么实现的，如下：

public void accept(int acceptorID) throws IOException
100    {
101        ServerSocketChannel server;
102        synchronized(this)
103        {
104            server = _acceptChannel;
105        }
106
107        if (server!=null && server.isOpen() && _manager.isStarted())
108        {
109            SocketChannel channel = server.accept();
110            channel.configureBlocking(false);
111            Socket socket = channel.socket();
112            configure(socket);
113            _manager.register(channel);
114        }
115    }

关于Thread stack，这里强调一点：nid，native lwp id，即本地轻量级进程（即线程）ID。

Case 2: NIO从的Selector

"qtp589745448-35 Selector0" prio=10 tid=0x00007f02f8ee9800 nid=0x18ed runnable [0x00007f02e71b4000]
   java.lang.Thread.State: RUNNABLE
	at sun.nio.ch.EPollArrayWrapper.epollWait(Native Method)
	at sun.nio.ch.EPollArrayWrapper.poll(EPollArrayWrapper.java:228)
	at sun.nio.ch.EPollSelectorImpl.doSelect(EPollSelectorImpl.java:81)
	at sun.nio.ch.SelectorImpl.lockAndDoSelect(SelectorImpl.java:87)
	- locked <0x00000000ec9006f0> (a sun.nio.ch.Util$2)
	- locked <0x00000000ec9006e0> (a java.util.Collections$UnmodifiableSet)
	- locked <0x00000000ec9004c0> (a sun.nio.ch.EPollSelectorImpl)
	at sun.nio.ch.SelectorImpl.select(SelectorImpl.java:98)
	at org.eclipse.jetty.io.nio.SelectorManager$SelectSet.doSelect(SelectorManager.java:569)
	at org.eclipse.jetty.io.nio.SelectorManager$1.run(SelectorManager.java:290)
	at org.eclipse.jetty.util.thread.QueuedThreadPool.runJob(QueuedThreadPool.java:608)
	at org.eclipse.jetty.util.thread.QueuedThreadPool$3.run(QueuedThreadPool.java:543)
	at java.lang.Thread.run(Thread.java:724)

   Locked ownable synchronizers:
	- None

代码片段如下：

       // If we should wait with a select
566                     if (wait>0)
567                     {
568                         long before=now;
569                         selector.select(wait);
570                         now = System.currentTimeMillis();
571                         _timeout.setNow(now);
572
573                         // If we are monitoring for busy selector
574                         // and this select did not wait more than 1ms
575                         if (__MONITOR_PERIOD>0 && now-before <=1)
576                         {
577                             // count this as a busy select and if there have been too many this monitor cycle
578                             if (++_busySelects>__MAX_SELECTS)
579                             {
580                                 // Start injecting pauses
581                                 _pausing=true;
582
583                                 // if this is the first pause
584                                 if (!_paused)
585                                 {
586                                     // Log and dump some status
587                                     _paused=true;
588                                     LOG.warn("Selector {} is too busy, pausing!",this);
589                                 }
590                             }
591                         }
592                     }

Case 3: ActveMQ中针对MQTT协议的Handler

"ActiveMQ Transport Server Thread Handler: mqtt://0.0.0.0:1883?maximumConnections=1000&wireFormat.maxFrameSize=104857600" daemon prio=10 tid=0x00007f02f8ba6000 nid=0x18dc waiting on condition [0x00007f02ec824000]
   java.lang.Thread.State: TIMED_WAITING (parking)
	at sun.misc.Unsafe.park(Native Method)
	- parking to wait for  <0x00000000faad0458> (a java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject)
	at java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:226)
	at java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject.awaitNanos(AbstractQueuedSynchronizer.java:2082)
	at java.util.concurrent.LinkedBlockingQueue.poll(LinkedBlockingQueue.java:467)
	at org.apache.activemq.transport.tcp.TcpTransportServer$1.run(TcpTransportServer.java:373)
	at java.lang.Thread.run(Thread.java:724)

   Locked ownable synchronizers:
	- None

代码片段：

  @Override
    protected void doStart() throws Exception {
        if (useQueueForAccept) {
            Runnable run = new Runnable() {
                @Override
                public void run() {
                    try {
                        while (!isStopped() && !isStopping()) {
                            Socket sock = socketQueue.poll(1, TimeUnit.SECONDS);
                            if (sock != null) {
                                handleSocket(sock);
                            }
                        }

                    } catch (InterruptedException e) {
                        LOG.info("socketQueue interuppted - stopping");
                        if (!isStopping()) {
                            onAcceptError(e);
                        }
                    }
                }
            };
            socketHandlerThread = new Thread(null, run, "ActiveMQ Transport Server Thread Handler: " + toString(), getStackSize());
            socketHandlerThread.setDaemon(true);
            socketHandlerThread.setPriority(ThreadPriorities.BROKER_MANAGEMENT - 1);
            socketHandlerThread.start();
        }
        super.doStart();
    }

Case 5: 模拟银行转帐存款

"withdraw" prio=10 tid=0x00007f3428110800 nid=0x2b6b waiting for monitor entry [0x00007f34155bb000]
   java.lang.Thread.State: BLOCKED (on object monitor)
	at com.von.thread.research.DeadThread.depositMoney(DeadThread.java:13)
	- waiting to lock <0x00000000d7fae540> (a java.lang.Object)
	- locked <0x00000000d7fae530> (a java.lang.Object)
	at com.von.thread.research.DeadThread.run(DeadThread.java:28)
	at java.lang.Thread.run(Thread.java:724)

   Locked ownable synchronizers:
	- None

"deposit" prio=10 tid=0x00007f342810f000 nid=0x2b6a waiting for monitor entry [0x00007f34156bc000]
   java.lang.Thread.State: BLOCKED (on object monitor)
	at com.von.thread.research.DeadThread.withdrawMoney(DeadThread.java:21)
	- waiting to lock <0x00000000d7fae530> (a java.lang.Object)
	- locked <0x00000000d7fae540> (a java.lang.Object)
	at com.von.thread.research.DeadThread.run(DeadThread.java:29)
	at java.lang.Thread.run(Thread.java:724)

   Locked ownable synchronizers:
	- None

Found one Java-level deadlock:
=============================
"withdraw":
  waiting to lock monitor 0x00007f3400003620 (object 0x00000000d7fae540, a java.lang.Object),
  which is held by "deposit"
"deposit":
  waiting to lock monitor 0x00007f3400004b20 (object 0x00000000d7fae530, a java.lang.Object),
  which is held by "withdraw"

Java stack information for the threads listed above:
===================================================
"withdraw":
	at com.von.thread.research.DeadThread.depositMoney(DeadThread.java:13)
	- waiting to lock <0x00000000d7fae540> (a java.lang.Object)
	- locked <0x00000000d7fae530> (a java.lang.Object)
	at com.von.thread.research.DeadThread.run(DeadThread.java:28)
	at java.lang.Thread.run(Thread.java:724)
"deposit":
	at com.von.thread.research.DeadThread.withdrawMoney(DeadThread.java:21)
	- waiting to lock <0x00000000d7fae530> (a java.lang.Object)
	- locked <0x00000000d7fae540> (a java.lang.Object)
	at com.von.thread.research.DeadThread.run(DeadThread.java:29)
	at java.lang.Thread.run(Thread.java:724)

Found 1 deadlock.

这里是一个非顺序加锁诱发的一个死锁场景。

好了，差不多了。 总结一下，在调优过程中，重点关注以下三类情况：

1. waiting for monitor entry – thread state blocked。可能发生的问题: deadlock（sequential deadlock，starvation deadlock...）
       2. waiting on condition – sleeping or timed_waiting。可能发生的问题: IO bottleneck
       3. Object.wait – TIMED_WAITING。问题：wait  & notifyAll使用上需要明确其性能及其局限性问题。External锁，更多的考虑使用await和singal方法。

参考文献：

1. http://architects.dzone.com/articles/how-analyze-java-thread-dumps

2. http://stackoverflow.com/questions/7698861/simple-java-example-runs-with-14-threads-why

3. http://www.longene.org/forum/viewtopic.php?f=5&t=94&p=399#p399

4. http://www.slideshare.net/Byungwook/analysis-bottleneck-in-j2ee-application

5. http://docs.oracle.com/javase/1.5.0/docs/guide/misc/threadPrimitiveDeprecation.html

6. http://www.artima.com/insidejvm/ed2/threadsynch.html

7. JavaConcurrency in practice