Occasional message loss executing a recv right after executing a send on a REQ socket

[paolovictor]

The scenario has two processes, one being a C server listening on a 0MQ REP socket "tcp://127.0.0.1:5555" and the other a Java client that issues requests to this server through a 0MQ REQ socket.

Right after receiving a request, the server replies with an ACK message. The server recv call is non-blocking. This way, the server and client flow would be:

C Server:
if(recv(socket, &request, ZMQ_NOBLOCK) == 0) {
// proccess reply
zmg_msg_close(&request);
send(socket, &reply, 0);
zmg_msg_close(&reply);
}

Java Client:
requestSocket.send(request_data, 0);
serverReply = requestSocket.recv(0);

The problem is that, even though the server send() call always returns 0, sometimes the client never receives the reply, staying blocked at the recv() call. Curiously, when I added a Thread.sleep(500); between the client's send and recv calls, I couldn't reproduce the problem after a couple dozen attempts.

One thing that also bugs me is that all blocking send() and recv() calls on the C server's REP socket set errno to EAGAIN, even when the call returns 0 and the message is correctly received. The zmq Manual (http://api.zeromq.org/) says that EAGAIN means "Non-blocking mode was requested and no messages are available at the moment", but I'm not even using non-blocking mode, so what gives?

[gonzus]

Have you been able to reproduce this problem with the latest versions of 0MQ and the Java binding? I would also advise to bring this up on the mailing list, in case you have not done it already.

[gonzus]

Since there have been no comments since September 2010, I will close this issue. Feel free to open a new one if necessary.

[paolovictor]

Hello.

I had a sudden detour from my project at the time, sorry for the (enormous) delay. This week I'll try to reproduce the bug with a more recent ZeroMQ version.

[paolovictor]

Just to chime in, I couldn't reproduce the bug with the git versions of ZMQ stable and jzmq.

Here's the test code:

Java client:

import org.zeromq.ZMQ;

public class ZQMRepTest {
    public static void main(String[] args) {
        ZMQ.Context ctx = ZMQ.context(1);
        ZMQ.Socket socket = ctx.socket(ZMQ.REQ);

        System.out.println("Connecting...");
        socket.connect("tcp://127.0.0.1:5555");

        System.out.println("Sending...");
        socket.send(new byte[]{3}, 0);

        System.out.println("Receiving...");
        byte[] reply = socket.recv(0);

        System.out.println("Received: " + new String(reply));
    }
}

C server:

int main(int argc, char* argv) {
void* context = zmq_init(1);
void* socket = zmq_socket(context, ZMQ_REP);

zmq_bind(socket, "tcp://127.0.0.1:5555");

while(1) {
    zmq_msg_t request;
    zmq_msg_init(&request);

    zmq_msg_t reply;
    zmq_msg_init_size(&reply, 6);
    memcpy(zmq_msg_data(&reply), "Hello\0", 6);


    if(zmq_recv(socket, &request, ZMQ_NOBLOCK) == 0) {
        zmq_send(socket, &reply, 0);
        printf("Sent reply!\n");

        zmq_msg_close(&request);
        zmq_msg_close(&reply);
    }

    sleep(1);
}

return 0;
}

[gonzus]

OK, maybe this was something that was fixed since the initial report. More the reason to leave the issue closed. Thanks for telling us about your tests.

[hamroune]

Hi every one
I have the same test in Ubuntu 10.11 whith jzmq the problem is i dont receiv anything in subscriber

code :
public static void main(String[] args) {
ZMQ.Context ctx = ZMQ.context(1);
ZMQ.Socket socket = ctx.socket(ZMQ.REQ);

    System.out.println("Connecting...");
    socket.connect("tcp://127.0.0.1:5555");

    System.out.println("Sending...");
    socket.send(new byte[]{3}, 0);

    System.out.println("Receiving...");
    byte[] reply = socket.recv(0);

    System.out.println("Received: " + new String(reply));
}

the result is simply:

Connecting...
Sending...
Receiving..

I'm beginner in zmq so may forget something .