这是在C ++中实现有界缓冲区的正确方法吗?

时间:2013-04-03 21:12:19

标签: c++ multithreading vector thread-safety buffer

我正在开发一个程序,处理多个线程访问,存入和从有界缓冲容器中撤出。我注意到线程的一些主要问题,并怀疑我的缓冲区在某处部分或根本不正确。

为了确保我知道我在做什么,我希望看看我的缓冲区代码。这个类使用了我在其他地方实现的信号量,我现在假设它可以工作(如果没有,我会尽快弄清楚!)我添加了试图解释我的推理的评论。

首先是.h文件:

#ifndef BOUNDED_BUFFER_H
#define BOUNDED_BUFFER_H

#include "Semaphore.H"
#include <string> 
#include <vector>  

using namespace std; 

class Item{ //supposed to eventually be more extensible...seems silly with just a string for now

  public:
    Item(string _content){content = _content;} 
    string GetContent() {return content;}     

  private:  
};   

    class BoundedBuffer{

      public:
        BoundedBuffer(); 

        void Deposit(Item* _item); 
        Item* Retrieve();        
        int GetNumItems() {return count;} 
        vector<Item*> GetBuffer() {return buffer;} 
        void SetSize(int _size){
          capacity = _size;
          buffer.reserve(_size);  //or do I need to call "resize" 
        }  

      private:
        int capacity; 
        vector<Item*> buffer; //I originally wanted to use an array but had trouble with  
                              //initilization/sizing, etc. 
        int nextin; 
            int nextout; 
            int count; 

            Semaphore notfull;   //wait on this one before depositing an item
            Semaphore notempty;  //wait on this one before retrieving an item
        };

    #endif

接下来是.cpp:

#include "BoundedBuffer.H"
#include <iostream>

using namespace std; 

BoundedBuffer::BoundedBuffer(){

  notfull.SetValue(0); 
  notempty.SetValue(0); 
  nextin = 0; 
  nextout = 0; 
  count = 0; 
}

void BoundedBuffer::Deposit(Item* _item){
  if(count == capacity){ 
    notfull.P(); //Cannot deposit into full buffer; wait
  }

  buffer[nextin] = _item; 
  nextin = (nextin + 1) % capacity;  //wrap-around
  count += 1;
  notempty.V();  //Signal that retrieval is safe 
}

Item* BoundedBuffer::Retrieve(){
  if(count == 0){
    notempty.P(); //cannot take from empty buffer; wait 
  }

  Item* x = buffer[nextout]; 
  nextout = (nextout + 1) % capacity;
  buffer.pop_back();  //or a different erase methodology? 
  count -= 1; 
  notfull.V(); //Signal that deposit is safe 
  return x; 
}

我认为问题可能来自我选择的矢量作为底层容器(或者更确切地说,错误地使用它),或者可能需要更多的安全阻塞机制(互斥锁等等)?事物的外观,任何人都可以提供一些反馈吗?

2 个答案:

答案 0 :(得分:4)

这是一个非常常见的问题(关于如何正确执行多线程队列)。我之前见过的最佳答案是this stack overflow questionthis web site。这些答案是针对无限队列的,所以我将在这里扩展并显示一个有界队列的答案。

您需要使用互斥锁保护存款和检索功能,并使用条件变量进行唤醒。

#include <mutex>
#include <condition_variable>

std::mutex the_mutex;
std::condition_variable the_notfull_cvar;
std::condition_variable the_notempty_cvar;

...

void BoundedBuffer::Deposit(Item* _item){
  std::unique_lock<std::mutex> lock(the_mutex);
  while ( /* buffer is full */ ){
    /* simultaneously wait and release the mutex */
    the_notfull_cvar.wait(lock);
    /* the mutex is reaquired at this point */
  }

  /* buffer has space and we own the mutex: insert the item */
  ...
  /* tell anyone waiting on an empty buffer that they can wake up. */
  the_notempty_cvar.notify_all();
}

Item* BoundedBuffer::Retrieve(){
  std::unique_lock<std::mutex> lock(the_mutex);
  while ( /* buffer is empty */ ){
    /* simultaneously wait and release the mutex */
    the_notempty_cvar.wait(lock);
    /* the mutex is reaquired at this point */
  }

  /* buffer has something in it and we own the mutex: get the item */
  ...
  /* tell anyone waiting on a full buffer that they can wake up. */
  the_notfull_cvar.notify_all();

  return x;
}

您的GetNumItems(),GetBuffer()和SetSize()函数也需要使用unique_locks进行保护。

答案 1 :(得分:1)

你看过Boost:Circular_buffer了吗?这是一个固定大小的存储领域,具有标准库接口。这可能会做你想要的,或者给你一些指示。如果在满的时候写一些东西,循环缓冲区会覆盖开头。也许你不希望这样,尽管有一个full()测试,所以你可以在代码中避免这种情况。