热门标签 | HotTags
当前位置:  开发笔记 > 编程语言 > 正文

Threadsynchronization:WaitandPulsedemystified

转载至 https:www.codeproject.comArticles28785Thread-synchronization-Wait-and-Pulse-demystifie

转载至 https://www.codeproject.com/Articles/28785/Thread-synchronization-Wait-and-Pulse-demystified

Table of Contents



  • Introduction

  • Basics

  • Locks

  • Queues

  • Recommended pattern

  • Uses

  • Tips

  • Example

  • Conclusion


Introduction

This short article is about the three less well understood methods of the Monitor class: WaitPulse, and PulseAll. The MSDN documentation for these methods explains what they do, but not why or how to use them. I will try to fill this gap and demystify them.

These methods provide low-level synchronization between threads. They are the most versatile constructs, but they are harder to use correctly than other synchronization primitives like AutoResetEvent. However, if you follow the recommended pattern I will describe, they can be used with confidence.

Basics


public partial static class Monitor
{
public static bool Wait( object o ) { ... } // and overloads
public static void Pulse( object o ) { ... }
public static void PulseAll( object o ) { ... }
}

At the most basic level, a thread calls Wait when it wants to enter a wait state, and another thread calls Pulse when it wants to wake up the waiting thread. Note that if a thread calls Pulse when no other threads are waiting, the Pulse is lost. This is in contrast to primitives like AutoResetEvent, where a call to Set is remembered until a subsequent call to WaitOne occurs.

There is a requirement that all these methods must be called from within a lock statement. If they are called while not holding the lock, then they will throw a SynchronizationLockException. We will see why this is actually useful, later on.

For example:

readonly object key = new object();
// thread A
lock ( key ) Monitor.Wait( key );
// thread B
lock ( key ) Monitor.Pulse( key );

If thread A runs first, it acquires the lock and executes the Wait method. Then, thread B runs, releasing thread A to allow it to continue.

Locks

You may have noticed a little problem with the code above. If thread A holds the lock on the key object, why does thread B not block when it tries to acquire the lock? This is, of course, handled properly. The call to Wait in thread A releases the lock before it waits. This allows thread B to acquire the lock and call Pulse. Thread A then resumes, but it has to wait until thread B releases the lock, so it can reacquire it and complete the Wait call. Note that Pulse never blocks.

技术分享图片

Queues

The ready queue is the collection of threads that are waiting for a particular lock. The Monitor.Wait methods introduce another queue: the waiting queue. This is required as waiting for a Pulse is distinct from waiting to acquire a lock. Like the ready queue, the waiting queue is FIFO.

技术分享图片

Recommended pattern

These queues can lead to unexpected behaviour. When a Pulse occurs, the head of the waiting queue is released and is added to the ready queue. However, if there are other threads in the ready queue, they will acquire the lock before the thread that was released. This is a problem, because the thread that acquires the lock can alter the state that the pulsed thread relies on.

The solution is to use a while condition inside the lock statement:

readonly object key = new object();
bool block = true;
// thread A
lock ( key )
{
while ( block )
Monitor.Wait( key );
block = true;
}
// thread B
lock ( key )
{
block = false;
Monitor.Pulse( key );
}

This pattern shows the reason for the rule that locks must be used: they protect the condition variable from concurrent access. Locks are also memory barriers, so you do not have to declare the condition variables as volatile.

The while loop also solves a couple of other problems. Firstly, if thread B executes before thread A calls Wait, the Pulse is lost. However, when thread A eventually runs and acquires the lock, the first thing it does is check the condition. Thread B has already set block to false, so Thread A continues without ever calling Wait.

Secondly, it solves the problem of the queues. If thread A is pulsed, it leaves the waiting queue and is added to the ready queue. If, however, a different thread acquires the lock and this thread sets block back to true, it now doesn‘t matter. This is because thread A will go round the while loop, find the condition to block is true, and execute Wait again.

Uses

The code above is actually an implementation of an AutoResetEvent. If you omitted the block = true statement in thread A, it would be a ManualResetEvent. If you use an int instead of bool for the condition, it would be a Semaphore. This shows how versatile this pattern is. You have complete control over what condition you use in the while loop.

So, the rule of thumb is: use higher level primitives if they fit. If you need finer control, use the Monitor methods with this pattern to ensure correctness.

Tips

It is usually okay to use PulseAll instead of Pulse. This causes all the waiting threads to re-evaluate their conditions and either continue or go back to waiting. As long as the condition evaluation is not too expensive, this is normally acceptable. PulseAll is also useful where you have multiple threads waiting on the same synchronization object, but with different conditions.

There are overloads of Wait that take a timeout parameter. Similar to above, this usually doesn‘t hurt. It can help prevent stuck Waits or missed Pulses by re-evaluating the conditions on a regular basis.

Example

Here is an example with full source code that demonstrates the versatility of this pattern. It implements a blocking queue that can be stopped. A blocking queue is a fixed-size queue. If the queue is full, attempts to add an item will block. If the queue is empty, attempts to remove an item will block. When Quit() is called, the queue is stopped. This means that you can‘t add any more items, but you can remove existing items until the queue is empty. At that point, the queue is finished.

This is a fairly complex set of conditions. You could implement this using a combination of higher-level constructs, but it would be harder. The pattern makes this implementation relatively trivial.

class BlockingQueue
{
readonly int _Size = 0;
readonly Queue _Queue = new Queue();
readonly object _Key = new object();
bool _Quit = false;
public BlockingQueue( int size )
{
_Size = size;
}
public void Quit()
{
lock ( _Key )
{
_Quit = true;
Monitor.PulseAll( _Key );
}
}
public bool Enqueue( T t )
{
lock ( _Key )
{
while ( !_Quit && _Queue.Count >= _Size ) Monitor.Wait( _Key );
if ( _Quit ) return false;
_Queue.Enqueue( t );
Monitor.PulseAll( _Key );
}
return true;
}
public bool Dequeue( out T t )
{
t = default( T );
lock ( _Key )
{
while ( !_Quit && _Queue.Count == 0 ) Monitor.Wait( _Key );
if ( _Queue.Count == 0 ) return false;
t = _Queue.Dequeue();
Monitor.PulseAll( _Key );
}
return true;
}
}

This implementation is safe for concurrent access by an arbitrary number of producers and consumers. Here is an example with one fast producer and two slow consumers:

internal static void Test()
{
var q = new BlockingQueue( 4 );
// Producer
new Thread( () =>
{
for ( int x = 0 ; ; x++ )
{
if ( !q.Enqueue( x ) ) break;
Trace.WriteLine( x.ToString( "0000" ) + " >" );
}
Trace.WriteLine( "Producer quitting" );
} ).Start();
// Consumers
for ( int i = 0 ; i <2 ; i++ )
{
new Thread( () =>
{
for ( ; ; )
{
Thread.Sleep( 100 );
int x = 0;
if ( !q.Dequeue( out x ) ) break;
Trace.WriteLine( " <" + x.ToString( "0000" ) );
}
Trace.WriteLine( "Consumer quitting" );
} ).Start();
}
Thread.Sleep( 1000 );
Trace.WriteLine( "Quitting" );
q.Quit();
}

And, here is the output of one run:

0.00000000 0000 >
0.00006564 0001 >
0.00009096 0002 >
0.00011540 0003 >
0.09100076 <0000
0.09105981 <0001
0.09118936 0004 >
0.09121715 0005 >
0.19127709 <0002
0.19138214 0006 >
0.19141905 0007 >
0.19156006 <0003
0.29184034 <0004
0.29195839 <0005
0.29209006 0008 >
0.29211268 0009 >
0.39240077 <0006
0.39249521 <0007
0.39265713 0010 >
0.39268187 0011 >
0.49300483 <0008
0.49308145 0012 >
0.49310759 0013 >
0.49324051 <0009
0.59353358 <0010
0.59361452 <0011
0.59378797 0014 >
0.59381104 0015 >
0.69410956 <0012
0.69421405 0016 >
0.69423932 0017 >
0.69443953 <0013
0.79467082 <0014
0.79478532 <0015
0.79493624 0018 >
0.79496473 0019 >
0.89524573 <0016
0.89536309 <0017
0.89549100 0020 >
0.89552164 0021 >
0.98704302 Quitting
0.98829663 Producer quitting
0.99580252 <0018
0.99590403 <0019
1.09638131 <0020
1.09647286 <0021
1.19700873 Consumer quitting
1.19717586 Consumer quitting


推荐阅读
  • Html5-Canvas实现简易的抽奖转盘效果
    本文介绍了如何使用Html5和Canvas标签来实现简易的抽奖转盘效果,同时使用了jQueryRotate.js旋转插件。文章中给出了主要的html和css代码,并展示了实现的基本效果。 ... [详细]
  • 基于layUI的图片上传前预览功能的2种实现方式
    本文介绍了基于layUI的图片上传前预览功能的两种实现方式:一种是使用blob+FileReader,另一种是使用layUI自带的参数。通过选择文件后点击文件名,在页面中间弹窗内预览图片。其中,layUI自带的参数实现了图片预览功能。该功能依赖于layUI的上传模块,并使用了blob和FileReader来读取本地文件并获取图像的base64编码。点击文件名时会执行See()函数。摘要长度为169字。 ... [详细]
  • HDU 2372 El Dorado(DP)的最长上升子序列长度求解方法
    本文介绍了解决HDU 2372 El Dorado问题的一种动态规划方法,通过循环k的方式求解最长上升子序列的长度。具体实现过程包括初始化dp数组、读取数列、计算最长上升子序列长度等步骤。 ... [详细]
  • 本文讨论了Alink回归预测的不完善问题,指出目前主要针对Python做案例,对其他语言支持不足。同时介绍了pom.xml文件的基本结构和使用方法,以及Maven的相关知识。最后,对Alink回归预测的未来发展提出了期待。 ... [详细]
  • 本文讨论了如何优化解决hdu 1003 java题目的动态规划方法,通过分析加法规则和最大和的性质,提出了一种优化的思路。具体方法是,当从1加到n为负时,即sum(1,n)sum(n,s),可以继续加法计算。同时,还考虑了两种特殊情况:都是负数的情况和有0的情况。最后,通过使用Scanner类来获取输入数据。 ... [详细]
  • 本文介绍了OC学习笔记中的@property和@synthesize,包括属性的定义和合成的使用方法。通过示例代码详细讲解了@property和@synthesize的作用和用法。 ... [详细]
  • 本文讲述了作者通过点火测试男友的性格和承受能力,以考验婚姻问题。作者故意不安慰男友并再次点火,观察他的反应。这个行为是善意的玩人,旨在了解男友的性格和避免婚姻问题。 ... [详细]
  • Voicewo在线语音识别转换jQuery插件的特点和示例
    本文介绍了一款名为Voicewo的在线语音识别转换jQuery插件,该插件具有快速、架构、风格、扩展和兼容等特点,适合在互联网应用中使用。同时还提供了一个快速示例供开发人员参考。 ... [详细]
  • 1,关于死锁的理解死锁,我们可以简单的理解为是两个线程同时使用同一资源,两个线程又得不到相应的资源而造成永无相互等待的情况。 2,模拟死锁背景介绍:我们创建一个朋友 ... [详细]
  • 《数据结构》学习笔记3——串匹配算法性能评估
    本文主要讨论串匹配算法的性能评估,包括模式匹配、字符种类数量、算法复杂度等内容。通过借助C++中的头文件和库,可以实现对串的匹配操作。其中蛮力算法的复杂度为O(m*n),通过随机取出长度为m的子串作为模式P,在文本T中进行匹配,统计平均复杂度。对于成功和失败的匹配分别进行测试,分析其平均复杂度。详情请参考相关学习资源。 ... [详细]
  • 动态规划算法的基本步骤及最长递增子序列问题详解
    本文详细介绍了动态规划算法的基本步骤,包括划分阶段、选择状态、决策和状态转移方程,并以最长递增子序列问题为例进行了详细解析。动态规划算法的有效性依赖于问题本身所具有的最优子结构性质和子问题重叠性质。通过将子问题的解保存在一个表中,在以后尽可能多地利用这些子问题的解,从而提高算法的效率。 ... [详细]
  • CF:3D City Model(小思维)问题解析和代码实现
    本文通过解析CF:3D City Model问题,介绍了问题的背景和要求,并给出了相应的代码实现。该问题涉及到在一个矩形的网格上建造城市的情景,每个网格单元可以作为建筑的基础,建筑由多个立方体叠加而成。文章详细讲解了问题的解决思路,并给出了相应的代码实现供读者参考。 ... [详细]
  • Java验证码——kaptcha的使用配置及样式
    本文介绍了如何使用kaptcha库来实现Java验证码的配置和样式设置,包括pom.xml的依赖配置和web.xml中servlet的配置。 ... [详细]
  • 高质量SQL书写的30条建议
    本文提供了30条关于优化SQL的建议,包括避免使用select *,使用具体字段,以及使用limit 1等。这些建议是基于实际开发经验总结出来的,旨在帮助读者优化SQL查询。 ... [详细]
  • 在project.properties添加#Projecttarget.targetandroid-19android.library.reference.1..Sliding ... [详细]
author-avatar
liu
这个家伙很懒,什么也没留下!
PHP1.CN | 中国最专业的PHP中文社区 | DevBox开发工具箱 | json解析格式化 |PHP资讯 | PHP教程 | 数据库技术 | 服务器技术 | 前端开发技术 | PHP框架 | 开发工具 | 在线工具
Copyright © 1998 - 2020 PHP1.CN. All Rights Reserved | 京公网安备 11010802041100号 | 京ICP备19059560号-4 | PHP1.CN 第一PHP社区 版权所有