SPServer中线程池实现部分分析

原文链接:http://blog.sina.com.cn/s/blog_67b570090100jsda.html

一次偶遇，看到了Half-Sync/Half-async相关论文，提到SPServer，以下它的英文描述：SPServer is a server framework library written on C++ that implements the Half-Sync/Half-Async andLeader/Follower patterns. It's based on libevent in order to utilize the best I/O loop on any platform。
     对SPServer，我感兴趣的还是它的线程池实现，可能是因为应用场景关系，SPServer对线程池实现相对简单，没有一些复杂的花样。
    对线程池的封装主要提供了dispatch函数，将一个将要调用的函数分配给线程池里面的线程。
class SP_ThreadPool {
public:
    typedef void ( * DispatchFunc_t )( void * );

    SP_ThreadPool( int maxThreads, const char * tag = 0 );
    ~SP_ThreadPool();

    /// @return 0 : OK, -1 : cannot create thread
    int dispatch( DispatchFunc_t dispatchFunc, void *arg );

    int getMaxThreads();

private:
    char * mTag;

    int mMaxThreads;
    int mIndex;
    int mTotal;
    int mIsShutdown;

    pthread_mutex_t mMainMutex;
    pthread_cond_t mIdleCond;
    pthread_cond_t mFullCond;
    pthread_cond_t mEmptyCond;

    SP_Thread_t ** mThreadList;

    static void * wrapperFunc( void * );
    int saveThread( SP_Thread_t * thread );
};
下面是dispatch函数代码：
int SP_ThreadPool :: dispatch( DispatchFunc_t dispatchFunc, void *arg )
{
    int ret = 0;

    pthread_attr_t attr;
    SP_Thread_t * thread = NULL;

    pthread_mutex_lock( &mMainMutex );

    if( mIndex <= 0 && mTotal >= mMaxThreads ) {
        pthread_cond_wait( &mIdleCond, &mMainMutex );
    }

    if( mIndex <= 0 ) {
        SP_Thread_t * thread = ( SP_Thread_t * )malloc( sizeof( SP_Thread_t ) );
        thread->mId = 0;
        pthread_mutex_init( &thread->mMutex, NULL );
        pthread_cond_init( &thread->mCond, NULL );
        thread->mFunc = dispatchFunc;
        thread->mArg = arg;
        thread->mParent = this;

        pthread_attr_init( &attr );
        pthread_attr_setdetachstate( &attr,PTHREAD_CREATE_DETACHED );

        if( 0 == pthread_create( &( thread->mId ), &attr, wrapperFunc, thread ) ) {
            mTotal++;
            syslog( LOG_NOTICE, "[tp@%s] create thread#%ld\n", mTag, thread->mId );
        } else {
            ret = -1;
            syslog( LOG_WARNING, "[tp@%s] cannot create thread\n", mTag );
            pthread_mutex_destroy( &thread->mMutex );
            pthread_cond_destroy( &thread->mCond );
            free( thread );
        }
    } else {
        mIndex--;
        thread = mThreadList[ mIndex ];
        mThreadList[ mIndex ] = NULL;

        thread->mFunc = dispatchFunc;
        thread->mArg = arg;
        thread->mParent = this;

        pthread_mutex_lock( &thread->mMutex );
        pthread_cond_signal( &thread->mCond ) ;
        pthread_mutex_unlock ( &thread->mMutex );
    }

    pthread_mutex_unlock( &mMainMutex );

    return ret;
}
如果当前没有空闲的线程，那么创建一个新线程来处理该事务。每个线程运行一个包裹函数
void * SP_ThreadPool :: wrapperFunc( void * arg )
{
    SP_Thread_t * thread = ( SP_Thread_t * )arg;

    for( ; 0 == thread->mParent->mIsShutdown; ) {
        thread->mFunc( thread->mArg );

        pthread_mutex_lock( &thread->mMutex );
        if( 0 == thread->mParent->saveThread( thread ) ) {
            pthread_cond_wait( &thread->mCond, &thread->mMutex );
            pthread_mutex_unlock( &thread->mMutex );
        } else {
            pthread_mutex_unlock( &thread->mMutex );
            pthread_cond_destroy( &thread->mCond );
            pthread_mutex_destroy( &thread->mMutex );

            free( thread );
            break;
        }
    }

    pthread_mutex_lock( &thread->mParent->mMainMutex );
    thread->mParent->mTotal--;
    if( thread->mParent->mTotal <= 0 ) {
        pthread_cond_signal( &thread->mParent->mEmptyCond );
    }
    pthread_mutex_unlock( &thread->mParent->mMainMutex );

    return NULL;
}
在该函数中，如果处里完事务，那么调用pthread_cond_wait等待新任务。
从上面实现中，从线程池角度来说，它并没有缓冲待处理的作务，这是因为在spserver中，采用了一个线程来接收对msgquene队列中的消息，对于接收到的每一个消息，调用threadPool中dispatch函数来处理该任务，如果线程池中没有空闲的线程而且线程数达到最大线程数，当前线程就等待。这里或许就是所谓的Leader/Followers模式应用吧。
由于任务入队列与从队列中接收队列在不同线程中，添加任务线程与接收队列线程Worker通信主要采用unix域套接字，当有任务添加到任务队列中时，通过写文件通知读端，有任务添加到任务队列中。