ZeroMQ正在吃我的数据....但有时只是

dus*_*der 3 qt zeromq

所以,我正在探索0MQ作为我的集群应用程序的消息传递协议.它提供了我需要的那种异步通信.

到目前为止,我已经初步成功制作了一个原型.我能够使用几种不同类型的模式发送和接收消息.

我目前正试图PUB/SUB通过REP/REQ"握手" 获得一个初始同步模式.ZeroMQ指南在"节点协调"下描述了这种方法.而且该方法效果很好.唯一的问题是0MQ似乎正在吃掉约50%的信息,或者至少是其中的数据.

我正在使用Qt,因此消息包含一个QString.我将字符串打包成a QByteArray使用a QDataStream.然后,我通过QByteArray"电线"并在另一侧打开包装.我经常使用这种方法通过其他协议(如tcp套接字)进行通信,并且效果很好.

我有4个"工人"连接到via PUB/SUBREQ/REP一个"经理".在大多数运行中,至少1个,最多3个"工人"同步很好.但是,当他们不这样做时,他们会收到一个空字符串.

这是4个客户的日志:

alicia:

[09:18:15.337] [Info] Logging initialized
[09:18:15.337] [Debug] Initializing client 
                       alicia
[09:18:15.337] [Debug] Attempting to receive START signal from server
[09:18:15.340] [Info] Received message: 
                      START
[    09:18:15.341] [Debug] Recieved a SYNC message
                       SYNC
[09:18:15.341] [Info] Received START signal from manager
[09:18:15.341] [Info] Received message: 
                      2
Run Code Online (Sandbox Code Playgroud)

brenda:

[09:18:15.337] [Info] Logging initialized
[09:18:15.337] [Debug] Initializing client 
                       brenda
[09:18:15.337] [Debug] Attempting to receive START signal from server
[09:18:15.340] [Info] Received message: 
                      START
[    09:18:15.340] [Debug] Recieved a SYNC message
                       SYNC
[09:18:15.340] [Info] Received START signal from manager
[09:18:15.340] [Info] Received message: 
                      START
[09:18:15.340] [Debug] Sending 
                       0
                       th message
Run Code Online (Sandbox Code Playgroud)

carlie:

[09:18:15.336] [Info] Logging initialized
[09:18:15.337] [Debug] Initializing client 
                       carlie
[09:18:15.337] [Debug] Attempting to receive START signal from server
[09:18:15.340] [Info] Received message: 
                      START
[    09:18:15.340] [Debug] Recieved a SYNC message
[09:18:15.340] [Fatal] carlie
                        Caught unhandled WASError:
                       Assertion "SOMEPATH/zmqworker.cpp" failed at SOMEPATH/zmqworker.cpp:52:virtual void was::ZMQWorker::run(): Invalid sync message
Run Code Online (Sandbox Code Playgroud)

darcie:

[09:18:15.336] [Info] Logging initialized
[09:18:15.337] [Debug] Initializing client 
                       darcie
[09:18:15.337] [Debug] Attempting to receive START signal from server
[09:18:15.340] [Info] Received message: 
                      START
[    09:18:15.341] [Debug] Recieved a SYNC message
[09:18:15.341] [Fatal] darcie
                        Caught unhandled WASError:
                       Assertion "SOMEPATH/zmqworker.cpp" failed at SOMEPATH/zmqworker.cpp:52:virtual void was::ZMQWorker::run(): Invalid sync message
Run Code Online (Sandbox Code Playgroud)

这是大部分代码:

zmqtools.cpp

void sendQString( zmq::socket_t& socket, QString& str )
{
    QByteArray package;
    QDataStream packer( &package, QIODevice::WriteOnly );
    packer << str;
    zmq::message_t msg( package.data(), package.size(), NULL );
    WAS_ASSERT_MSG( socket.send( msg ), "Failed to send QString" );
}


void recvQString( zmq::socket_t& socket, QString& str )
{
    zmq::message_t msg;
    WAS_ASSERT_MSG( socket.recv( &msg ), "Failed to receive QString" );
    QByteArray package( ( char* )msg.data(), msg.size() );
    QDataStream unpacker( &package, QIODevice::ReadOnly );
    unpacker >> str;
}
Run Code Online (Sandbox Code Playgroud)

zmqworker.cpp

    ZMQWorker::ZMQWorker( const QString& clientName, QObject *parent ) :
    QThread( parent ),
    clientName( clientName ),
    context( 1 ),
    inMessageSocket( context, ZMQ_PULL ),
    outMessageSocket( context, ZMQ_PUSH ),
        inControlSocket( context, ZMQ_SUB ),
    synchronizeSocket( context, ZMQ_REQ )
{
    qxtLog->debug() << "Initializing client " << clientName;

    inMessageSocket.connect( "tcp://localhost:9900" );
    outMessageSocket.connect( "tcp://localhost:9901" );
    inControlSocket.connect( "tcp://localhost:9902" );
    inControlSocket.setsockopt( ZMQ_SUBSCRIBE, "", 0 );
    synchronizeSocket.connect( "tcp://localhost:9903" );

    messagePoll.fd = 0;
    messagePoll.events = ZMQ_POLLIN;
    messagePoll.revents = 0;
    messagePoll.socket = inMessageSocket;

    controlPoll.fd = 0;
    controlPoll.events = ZMQ_POLLIN;
    controlPoll.revents = 0;
    controlPoll.socket = inControlSocket;
}

void ZMQWorker::run()
{
    QString message;
    // Wait for start signal from server
    bool started = true;
    do
    {
        qxtLog->debug() << "Attempting to receive START signal from server";
        recvQString( inControlSocket, message );
        qxtLog->info() << "Received message: " << message;
        started = message == "START";
    } while( !started );

    message = "SYNC";
    sendQString( synchronizeSocket, message );
    recvQString( synchronizeSocket, message );
    qxtLog->debug() << "Recieved a SYNC message" << message;
    WAS_ASSERT_MSG( message == "SYNC", "Invalid sync message" );

    qxtLog->info() << "Received START signal from manager";

    int messagesSent = 0;
    forever
    {

        zmq::poll( &messagePoll, 1, 0 );
        if( messagePoll.revents & ZMQ_POLLIN )
        {
            recvQString( inMessageSocket, message );
            qxtLog->info() << "Received message: " << message;
        }

        zmq::poll( &controlPoll, 1, 0 );
        if( controlPoll.revents & ZMQ_POLLIN )
        {
            recvQString( inControlSocket, message );
            qxtLog->info() << "Received message: " << message;
            if( message == "STOP" )
                break;
        }


        if( messagesSent < 1000 )
        {
            qxtLog->debug() << "Sending " << messagesSent << "th message";
            QString message = QString::number( messagesSent );
            sendQString( outMessageSocket, message );
            messagesSent++;
        }
    }
}
Run Code Online (Sandbox Code Playgroud)

zmqmanager.cpp

ZMQManager::ZMQManager( const QString& serverName, unsigned clientCount, QObject* parent ) :
    QThread( parent ),
    serverName( serverName ),
    clientCount( clientCount ),
    context( 1 ),
    inMessageSocket( context, ZMQ_PULL ),
    outMessageSocket( context, ZMQ_PUSH ),
    outControlSocket( context, ZMQ_PUB ),
    synchronizeSocket( context, ZMQ_REP )
{
    qxtLog->debug() << "Initializing server " << serverName;


    outMessageSocket.bind( "tcp://*:9900" );
    inMessageSocket.bind( "tcp://*:9901" );
    outControlSocket.bind( "tcp://*:9902" );
    synchronizeSocket.bind( "tcp://*:9903" );


    messagePoll.fd = 0;
    messagePoll.events = ZMQ_POLLIN;
    messagePoll.revents = 0;
    messagePoll.socket = inMessageSocket;

    synchronizePoll.fd = 0;
    synchronizePoll.events = ZMQ_POLLIN;
    synchronizePoll.revents = 0;
    synchronizePoll.socket = synchronizeSocket;
}

void ZMQManager::run()
{
    QString message;
    unsigned clientsConnected = 0;
    do
    {
        qxtLog->debug() << "Publishing START signal";
        message = "START";
        sendQString( outControlSocket, message );

        zmq::poll( &synchronizePoll, 1, 5000 );
        if( synchronizePoll.revents & ZMQ_POLLIN )
        {
            qxtLog->debug() << "Checking for response to START signal";
            recvQString( synchronizeSocket, message );
            qxtLog->debug() << "Recieved a SYNC message" << message;
            WAS_ASSERT_MSG( message == "SYNC", "Invalid sync message" );
            sendQString( synchronizeSocket, message );
            clientsConnected++;
        }

    } while( clientsConnected < clientCount );

    qxtLog->info() << "Started and syncrhonized with clients";

    unsigned messagesSent = 0;
    unsigned messagesReceived = 0;
    do
    {
            zmq::poll( &messagePoll, 1, 0 );
        if( messagePoll.revents & ZMQ_POLLIN )
        {
            recvQString( inMessageSocket, message );
            qxtLog->info() << "Received message: " << message;
            messagesReceived++;
        }

        if( messagesSent < clientCount * 1000 )
        {
            qxtLog->debug() << "Sending a message";
            message = QString::number( messagesSent );
            sendQString( outMessageSocket, message );
            messagesSent++;
        }

    } while( messagesSent < clientCount * 1000 && messagesReceived < clientCount * 1000 );

    message = "STOP";
    sendQString( outControlSocket, message );

}
Run Code Online (Sandbox Code Playgroud)

所以,最后我的问题:

  • 我在这段代码中忽略了什么吗?
  • 我是否错误地打包/解包ZeroMQ消息?
  • REQ/REP是否错误地管理了同步化?

我会很感激的人谁拥有ZMQ一些经验的一些见解,特别是与混合PUB/SUB,PUSH/PULLREQ/REP模式.

dus*_*der 8

问题是ZeroMQ在使用构造函数时不会复制数据

zmq::message_t( void*, int size );

相反,ZeroMQ只使用您提供的缓冲区作为邮件的存储空间.因此,如果在实际发送消息之前释放或覆盖了该缓冲区,则会出现垃圾或seg-fault.

解决方案相对容易.而是使用此构造函数:

zmq::message_t( int size );

它创建一个指定大小的缓冲区.然后,您只需在发送之前手动将数据手动复制到消息中.

这是更正的,现在正在工作的发送功能:

void sendQString( zmq::socket_t& socket, QString& str )
{
    QByteArray package;
    QDataStream packer( &package, QIODevice::WriteOnly );
    packer << str;
    qxtLog->debug() << "sending a message of " << package.size() << " bytes";
    qxtLog->debug() << "the message says " << str;
    zmq::message_t msg( package.size() );
    memcpy( msg.data(), package.data(), package.size() );
    ASSERT_MSG( socket.send( msg ), "Failed to send QString" );
}

void recvQString( zmq::socket_t& socket, QString& str )
{    
    zmq::message_t msg;
    ASSERT_MSG( socket.recv( &msg ), "Failed to receive QString" );
    qxtLog->debug() << "received a message of " << (int)msg.size() << " bytes";
    QByteArray package( ( char* )msg.data(), msg.size() );
    QDataStream unpacker( &package, QIODevice::ReadOnly );
    unpacker >> str;
    qxtLog->debug() << "the message says " << str;
}
Run Code Online (Sandbox Code Playgroud)

感谢freenode上#zeromq频道上的人,我终于能够发现问题所在.