阻塞队列实现：争用条件在哪里？

Blocking Queue implementation: Where's the race condition?

本文关键字：争用条件 在哪里 实现 队列      更新时间：2023-10-16

又是我和我的BlockingQueue…我根据这篇文章和这个问题重写了它。它发送一些项，然后由于访问冲突而崩溃。下面是代码:

template <typename T>
bool DRA::CommonCpp::CTBlockingQueue<T>::Push( T pNewValue ){
    volatile long oldSize;
    ::InterlockedExchange( &oldSize, m_Size );
    CTNode* pNewNode = new CTNode();
    pNewNode->m_pValue = pNewValue;
    {//RAII block
        CGuard g( m_TailCriticalSection );
        m_pTailNode->m_pNext = pNewNode;
        m_pTailNode = pNewNode;
        ::InterlockedIncrement( &m_Size );
    }
    if( oldSize == 0 )
        m_eAtLeastOneElement.set();
    return true;
}
template <typename T>
bool DRA::CommonCpp::CTBlockingQueue<T>::Pop( T& pValue ){
    CTNode* pCurrentNode;
    {//RAII block
        CGuard g( m_HeadCriticalSection );
        pCurrentNode = m_pHeadNode;
        CTNode* pNewHeadNode = m_pHeadNode->m_pNext;
        if( pNewHeadNode == NULL ){
            CEvent* pSignaledEvent;
            CEvent::waitForPair( m_eAtLeastOneElement, m_eFinished, pSignaledEvent );
            if( pSignaledEvent == &m_eFinished )
                return false;
            pNewHeadNode = m_pHeadNode->m_pNext;
        }
        pValue = pNewHeadNode->m_pValue;
        m_pHeadNode = pNewHeadNode;
        ::InterlockedDecrement( &m_Size );
    }
    delete pCurrentNode;
    return true;
}

它总是在调用Pop()时崩溃，在if之后的那行，上面写着:

pValue = pNewHeadNode->m_pValue

会导致cos的pNewHeadNode为NULL。但这是怎么发生的呢?

Edit:忘记初始化代码:

template <typename T>
DRA::CommonCpp::CTBlockingQueue<T>::CTBlockingQueue():
        m_HeadCriticalSection("CTBlockingQueue<T>::m_Head"),
        m_TailCriticalSection("CTBlockingQueue<T>::m_Tail"){
    CTNode* pDummyNode = new CTNode();
    m_pHeadNode = pDummyNode;
    m_pTailNode = pDummyNode;
    m_Size = 0; //Dummy node doesn't count
}

我的假设是事件设置在临界区之外，这意味着push可能会通知该事件两次。你试过在临界区内设置事件吗?

最后，我回到了我最初的，效率较低的实现，即我在这里发布的那个，添加了一个Finish()方法，以便生产者可以向消费者发出优雅的结束信号，以及一个Restart()方法，以便在不破坏和重新创建队列的情况下重新开始生产:

//Template definitions
template<class Element>
DRA::CommonCpp::CTBlockingQueue<Element>::CTBlockingQueue( unsigned int maxSize ):
    m_csFinished( "CTBlockingQueue::m_csFinished" ),
    m_csQueue( "CTBlockingQueue::m_csQueue" ),
    m_semElementCount( 0, maxSize ),
    m_bFinished(false){
}
template<class Element>
DRA::CommonCpp::CTBlockingQueue<Element>::~CTBlockingQueue(){
    Finish();
}
template<class Element>
void DRA::CommonCpp::CTBlockingQueue<Element>::Push( Element newElement ){
    {//RAII block
        CGuard g( m_csQueue );
        m_Queue.push( newElement );
    }
    m_semElementCount.Signal();
}
template<class Element>
bool DRA::CommonCpp::CTBlockingQueue<Element>::Pop( Element& element ){
    m_semElementCount.Wait();
    {//RAII block
        CGuard g( m_csFinished );
        if( m_bFinished ){
            CGuard g( m_csQueue );
            if ( m_Queue.size() == 0 )
                return false;
        }
    }
    {//RAII block
        CGuard g( m_csQueue );
        element = m_Queue.front();
        m_Queue.pop();
    }
    return true;
}
template<class Element>
void DRA::CommonCpp::CTBlockingQueue<Element>::Finish(){
    {//RAII block
        CGuard g( m_csFinished );
        m_bFinished = true;
    }
    {//RAII block
        CGuard g( m_csQueue );
        m_semElementCount.Signal();
    }
}
template<class Element>
void DRA::CommonCpp::CTBlockingQueue<Element>::Restart(){
    {//RAII block
        CGuard g( m_csFinished );
        m_bFinished = false;
    }
}

这不是最快的方法，但对我来说已经足够快了。

相关文章：