QUdpSocket requires delays between writes

jsulm

@Kent-Dorfman said in QUdpSocket requires delays between writes:

after last reference

What reference do you mean? Since it's a raw pointer there are no references.
Do you mean you put

myUdpSocket=new QUdpSocket()

inside main()? If so, the pointer goes out of scope when main finishes. The object itself is deleted when "delete" is called.

Mwvse

@Christian-Ehrlicher

Thank you for the reply.

I am sleeping for 5 seconds before main (and thus myUdpSocket) goes out of scope.

Both '.write()' calls return the expected number of bytes written but only the first write actually gets sent.

The only difference between the example that works and the one that doesn't is a 1 second delay between calls. It's as if the Qt implementation has no internal buffering but yet the write() call indicates success.

SGaist

Hi,

Since you are not using the asynchronous nature of Qt, you should at least use the synchronous API of QUdpSocket waitForBytesWritten.

Mwvse

@SGaist
Thank you very much for the reply.

I tried that but that call merely returned false and still only the first write() made it out onto the wire.

#include <QCoreApplication>
#include <QUdpSocket>
#include <QTextStream>
#include <QThread>
#include <QHostAddress>

int main(int argc, char *argv[])
{
	QTextStream	cout(stdout);
	QUdpSocket	myUdpSocket;

	myUdpSocket.connectToHost("127.0.0.1", 50000);

	cout << myUdpSocket.write("msg 1\n") << Qt::endl;
	cout << myUdpSocket.waitForBytesWritten(1000) << Qt::endl;
	cout << myUdpSocket.write("msg two\n") << Qt::endl;

	QThread::currentThread()->sleep(5);

	return 0;
}

I did see the following note in the reference manual (I am, indeed, trying this under Windows at the moment):

bool QAbstractSocket::waitForBytesWritten(int msecs = 30000)

Reimplements: QIODevice::waitForBytesWritten(int msecs).

...

**Note: This function may fail randomly on Windows. Consider using the event loop and the bytesWritten() signal if your software will run on Windows.**

SGaist

Out of curiosity, why are you not following the rule that you should first create a QCoreApplication before any QObject based classes ?

Mwvse

@Kent-Dorfman

Same result.

#include <QCoreApplication>
#include <QUdpSocket>
#include <QTextStream>
#include <QThread>
#include <QHostAddress>

int main(int argc, char *argv[])
{
	QTextStream	cout(stdout);
	QUdpSocket*	myUdpSocket = new QUdpSocket;

	QHostAddress	addr("127.0.0.1");

	myUdpSocket->connectToHost("127.0.0.1", 50000);

	cout << myUdpSocket->write("msg 1\n") << Qt::endl;
	cout << myUdpSocket->write("msg two\n") << Qt::endl;

	QThread::currentThread()->sleep(5);

	return 0;
}

Console output shows number of bytes written for each is successful:

And wireshark shows only the 1st write made it out:

Mwvse

@SGaist

Because I am naive :-) ?? I was just trying to exercise these library classes/methods to study them a bit. I did try the following suggestion someone posted elsewhere, but still got the same problem result:

#include <QCoreApplication>
#include <QUdpSocket>
#include <QTimer>
#include <QObject>
#include <QDebug>

class Sender: public QObject {
	 Q_OBJECT
public:
	 Sender(QObject *p = nullptr):
		  myUdpSocket(new QUdpSocket(this))    {
		  myUdpSocket->connectToHost("127.0.0.1", 50000);

		  connect(myUdpSocket, &QUdpSocket::bytesWritten, this, &Sender::sentStuff);

		  QTimer::singleShot(0, this, &Sender::sendStuff);
		  // ^^^ delays sending stuff until event loop is running
		  //     and this timer event is processed
	 }
	 ~Sender() { }
private slots:
	 void sendStuff()
	 {
		  qDebug() << myUdpSocket->write("msg 1\n");
		  qDebug() << myUdpSocket->write("msg two\n");
	 }

	 void sentStuff(qint64 bytes)
	 {
		  qDebug() << "Bytes sent:" << bytes;
	 }
private:
	 QUdpSocket *myUdpSocket;
};


int main(int argc, char *argv[]) {
	 QCoreApplication app(argc, argv);  // << application object
	 Sender s;
	 return app.exec();  // << event loop
}


#include "main.moc"

Console output:

Wireshark output:

kshegunov

As you have wireshark opened, could you check a couple of things:

That there's a bound receiver to that port
If there isn't, could you check if an ARP query is being done after the first datagram goes out
If there's such a query, then that's probably the answer to your question. For a detailed explanation, look here

Also note that sending datagrams below the FastSendDatagramThreshold value in quick succession while waiting for ARP to resolve may cause datagrams to be discarded:

ARP queues only one outbound IP datagram for a specified destination address while that IP address is being resolved to a media access control address. If a User Datagram Protocol (UDP)-based application sends multiple IP datagrams to a single destination address without any pauses between them, some of the datagrams may be dropped if there is no ARP cache entry already present. An application can compensate for this by calling the iphlpapi.dll routine SendArp() to establish an ARP cache entry, before sending the stream of packets.

Mwvse

Thank you very much @kshegunov for the reply.

I have tried using netcat (under Cygwin) to listen on the port (note that only "msg 1\n" datagram arrived):

Here is unfiltered packet capture around the send of the first (and only datagram). Note the time stamps around my UDP message:

I am curious about the ARP cache as I am using the loopback device "127.0.0.1".

kshegunov

@Mwvse said in QUdpSocket requires delays between writes:

Here is unfiltered packet capture around the send of the first (and only datagram). Note the time stamps around my UDP message:

You need to expand the Info column so you and we can see what's sent before/after the datagram. An ARP query (2 from above post) is a TCP broadcast into the subnet (I can't tell from the screenshot if that's the case here).

Mwvse

@kshegunov

Sorry. I didn't consider it because the timestamps showed multiple seconds in between and thus weren't relevant (a potentially error-prone assumption):

kshegunov

@Mwvse said in QUdpSocket requires delays between writes:

Sorry. I didn't consider it because the timestamps showed multiple seconds in between and thus weren't relevant (a potentially error-prone assumption):

Nope, it's a red herring. I don't see anything strange. Currently I have no idea why the datagram may be discarded ... Could you try another something. Modify the slot like this:

void sendStuff()
{
    qDebug() << myUdpSocket->write("msg 1\n");
    QObject::connect(myUdpSocket, &QUdpSocket::bytesWritten, myUdpSocket, [myUdpSocket] () -> void  {
        qDebug() << myUdpSocket->write("msg two\n");
    });
}

Does that work as expected?

Mwvse

@kshegunov

I am curious about the ARP cache problem and being on 127.0.0.1.

Mwvse

[Edit: Sorry I deleted your post by accident. I feel real dumb right now ~kshegunov]

kshegunov

Yes, my bad:

void sendStuff()
{
    QObject::connect(myUdpSocket, &QUdpSocket::bytesWritten, this, [this] () -> void  {
        QObject::disconnect(myUdpSocket, nullptr, this, nullptr);// To prevent looping
        qDebug() << myUdpSocket->write("msg two\n");
    });
    qDebug() << myUdpSocket->write("msg 1\n");
}

Mwvse

@kshegunov

Packet capture looks the same. Console output only shows first write.

kshegunov

This is very strange. I'm currently out of ideas ... :|

Mwvse

@kshegunov

Going back to your comment:

"ARP queues only one outbound IP datagram for a specified destination address while that IP address is being resolved to a media access control address. If a User Datagram Protocol (UDP)-based application sends multiple IP datagrams to a single destination address without any pauses between them, some of the datagrams may be dropped if there is no ARP cache entry already present. An application can compensate for this by calling the iphlpapi.dll routine SendArp() to establish an ARP cache entry, before sending the stream of packets."

I did the following:

#include <QCoreApplication>
#include <QUdpSocket>
#include <QTextStream>
#include <QThread>

int main(int argc, char *argv[])
{
	QCoreApplication app(argc, argv);

	QTextStream	cout(stdout);
	QUdpSocket*	myUdpSocket = new QUdpSocket;

	myUdpSocket->connectToHost("127.0.0.1", 50000);

	cout << myUdpSocket->write("msg 1\n") << Qt::endl;
	QThread::currentThread()->sleep(1);

	for(unsigned u = 0; u < 100; u++)
		cout << myUdpSocket->write("another\n") << Qt::endl;

	QThread::currentThread()->sleep(5);

	return 0;
}

As long as I have a delay after the first write, I get all 100 of the following at full speed. So ... something in the IP stack needs to get established. That first write seems to do it although it takes some time. I am wondering if since we are on 127.0.0.1 the ARP never goes out but the net result is the ARP cache gets created and then off we go.

Anyway, thank you very much for your time and expertise. I don't have a Linux box handy at the moment but I am curious if I get similar results. If I get a chance, I will try it and post the results here just for the record.

kshegunov

@Mwvse said in QUdpSocket requires delays between writes:

As long as I have a delay after the first write, I get all 100 of the following at full speed. So ... something in the IP stack needs to get established. That first write seems to do it although it takes some time. I am wondering if since we are on 127.0.0.1 the ARP never goes out but the net result is the ARP cache gets created and then off we go.

Yes, this would seem to support that hypothesis indeed, but I didn't see a TCP broadcast in wireshark. In any case something very windows specific may be going on (which has nothing to do with Qt apparently). On the other hand if this turns out to be the case, I think Qt can do better to wait for the ARP cache to be built before sending the following datagrams.

If I get a chance, I will try it and post the results here just for the record.

This would be much appreciated.