浅谈.NET下的多线程和并行计算(六)线程池基础下

1/6/2010来源:ASP技巧人气:3330

这节我们按照线程池的核心思想来自定义一个简单的线程池:

1) 池中使用的线程不少于一定数量,不多于一定数量

2) 池中线程不够的时候创建,富裕的时候收回

3) 任务排队,没有可用线程时,任务等待

我们的目的只是实现这些“需求”,不去考虑性能(比如等待一段时间再去创建新的线程等策略)以及特殊的处理(异常),在实现这个需求的过程中我们也回顾了线程以及线程同步的基本概念。

首先,把任务委托和任务需要的状态数据封装一个对象:

public class WorkItem
{
    public WaitCallback Action { get; set; }
    public object State { get; set; }

    public WorkItem(WaitCallback action, object state)
    {
        this.Action = action;
        this.State = state;
    }
}然后来创建一个对象作为线程池中的一个线程:

public class SimpleThreadPoolThread
{
    PRivate object locker = new object();
    private AutoResetEvent are = new AutoResetEvent(false);
    private WorkItem wi;
    private Thread t;
    private bool b = true;
    private bool isWorking;

    public bool IsWorking
    {
        get
        {
            lock (locker)
            {
                return isWorking;
            }
        }
    }
    public event Action<SimpleThreadPoolThread> WorkComplete;

    public SimpleThreadPoolThread()
    {
        lock (locker)
        {
            // 当前没有实际任务
            isWorking = false;
        }
        t = new Thread(Work) { IsBackground = true };
        t.Start();
    }

    public void SetWork(WorkItem wi)
    {
        this.wi = wi;
    }

    public void StartWork()
    {
        // 发出信号
        are.Set();
    }

    public void StopWork()
    {
        // 空任务
        wi = null;
        // 停止线程循环
        b = false;
        // 发出信号结束线程
        are.Set();
    }

    private void Work()
    {
        while (b)
        {
            // 没任务,等待信号
            are.WaitOne();
            if (wi != null)
            {
                lock (locker)
                {
                    // 开始
                    isWorking = true;
                }
                // 执行任务
                wi.Action(wi.State);
                lock (locker)
                {
                    // 结束
                    isWorking = false;
                }
                // 结束事件
                WorkComplete(this);
            }
        }
    }代码的细节可以看注释,对这段代码的整体结构作一个说明:

1) 由于这个线程是被线程池中任务复用的,所以线程的任务处于循环中,除非线程池打算回收这个线程,否则不会退出循环结束任务

2) 使用自动信号量让线程没任务的时候等待,由线程池在外部设置任务后发出信号来执行实际的任务,执行完毕后继续等待

3) 线程公开一个完成的事件,线程池可以挂接处理方法,在任务完成后更新线程池状态

4) 线程池中的所有线程都是后台线程

下面再来实现线程池:

public class SimpleThreadPool : IDisposable
{
    private object locker = new object();
    private bool b = true;
    private int minThreads;
    private int maxThreads;
    private int currentActiveThreadCount;
    private List<SimpleThreadPoolThread> simpleThreadPoolThreadList = new List<SimpleThreadPoolThread>();
    private Queue<WorkItem> workItemQueue = new Queue<WorkItem>();

    public int CurrentActiveThreadCount
    {
        get
        {
            lock (locker)
            {
                return currentActiveThreadCount;
            }
        }

    }

    public int CurrentThreadCount
    {
        get
        {
            lock (locker)
            {
                return simpleThreadPoolThreadList.Count;
            }
        }
    }

    public int CurrentQueuedWorkCount
    {
        get
        {
            lock (locker)
            {
                return workItemQueue.Count;
            }
        }
    }

    public SimpleThreadPool()
    {
        minThreads = 4;
        maxThreads = 25;
        Init();
    }

    public SimpleThreadPool(int minThreads, int maxThreads)
    {
        if (minThreads > maxThreads)
            throw new ArgumentException("minThreads > maxThreads", "minThreads,maxThreads");
        this.minThreads = minThreads;
        this.maxThreads = maxThreads;
        Init();
    }

    public void QueueUserWorkItem(WorkItem wi)
    {
        lock (locker)
        {
            // 任务入列
            workItemQueue.Enqueue(wi);
        }
    }

    private void Init()
    {
        lock (locker)
        {
            // 一开始创建最小线程
            for (int i = 0; i < minThreads; i++)
            {
                CreateThread();
            }
            currentActiveThreadCount = 0;
        }
        new Thread(Work) { IsBackground = true }.Start();
    }

    private SimpleThreadPoolThread CreateThread()
    {
        SimpleThreadPoolThread t = new SimpleThreadPoolThread();
        // 挂接任务结束事件
        t.WorkComplete += new Action<SimpleThreadPoolThread>(t_WorkComplete);
        // 线程入列
        simpleThreadPoolThreadList.Add(t);
        return t;
    }

    private void Work()
    {
        // 线程池主循环
        while (b)
        {
            Thread.Sleep(100);
            lock (locker)
            {
                // 如果队列中有任务并且当前线程小于最大线程
                if (workItemQueue.Count > 0 && CurrentActiveThreadCount < maxThreads)
                {
                    WorkItem wi = workItemQueue.Dequeue();
                    // 寻找闲置线程
                    SimpleThreadPoolThread availableThread = simpleThreadPoolThreadList.FirstOrDefault(t => t.IsWorking == false);
                    // 无则创建
                    if (availableThread == null)
                        availableThread = CreateThread();
                    // 设置任务
                    availableThread.SetWork(wi);
                    // 开始任务
                    availableThread.StartWork();
                    // 增加个活动线程
                    currentActiveThreadCount++;
                }
            }
        }
    }

    private void t_WorkComplete(SimpleThreadPoolThread t)
    {
        lock (locker)
        {
            // 减少个活动线程
            currentActiveThreadCount--;
            // 如果当前线程数有所富裕并且比最小线程多
            if ((workItemQueue.Count + currentActiveThreadCount) < minThreads && CurrentThreadCount > minThreads)
            {
                // 停止已完成的线程
                t.StopWork();
                // 从线程池删除线程
                simpleThreadPoolThreadList.Remove(t);
            }
        }
    }

    public void Dispose()
    {
        // 所有线程停止
        foreach (var t in simpleThreadPoolThreadList)
        {
            t.StopWork();
        }
        // 线程池主循环停止
        b = false;
    }
}线程池的结构如下:

1) 在构造方法中可以设置线程池最小和最大线程

2) 维护一个任务队列和一个线程池中线程的列表

3) 初始化线程池的时候就创建最小线程数量定义的线程

4) 线程池主循环每20毫秒就去处理一次,如果有任务并且线程池还可以处理任务的话,先是找闲置线程,找不到则创建一个

5) 通过设置任务委托以及发出信号量来开始任务

6) 线程池提供了三个属性来查看当前活动线程数,当前总线程数和当前队列中的任务数

7) 任务完成的回调事件中我们判断如果当前线程有富裕并且比最小线程多则回收线程

8) 线程池是IDispose对象,在Dispose()方法中停止所有线程后停止线程池主循环

写一段代码来测试线程池:

using (SimpleThreadPool t = new SimpleThreadPool(2, 4))
{
    Stopwatch sw2 = Stopwatch.StartNew();
    for (int i = 0; i < 10; i++)
    {
        t.QueueUserWorkItem(new WorkItem((index =>
        {
            Console.WriteLine(string.Format("#{0} : {1} / {2}", Thread.CurrentThread.ManagedThreadId, DateTime.Now.ToString("mm:ss"), index));
            Console.WriteLine(string.Format("CurrentActiveThread: {0} / CurrentThread: {1} / CurrentQueuedWork: {2}", t.CurrentActiveThreadCount, t.CurrentThreadCount, t.CurrentQueuedWorkCount));
            Thread.Sleep(1000);
        }), i));
    }
    while (t.CurrentQueuedWorkCount > 0 || t.CurrentActiveThreadCount > 0)
    {
        Thread.Sleep(10);
    }
    Console.WriteLine("All work completed");
    Console.WriteLine(string.Format("CurrentActiveThread: {0} / CurrentThread: {1} / CurrentQueuedWork: {2}", t.CurrentActiveThreadCount, t.CurrentThreadCount, t.CurrentQueuedWorkCount));
    Console.WriteLine(sw2.ElapsedMilliseconds);
} 代码中我们向线程池推入10个任务,每个任务需要1秒执行,任务执行前输出当前任务的所属线程的Id,当前时间以及状态值。然后再输出线程池的几个状态属性。主线程循环等待所有任务完成后再次输出线程池状态属性以及所有任务完成耗费的时间: