Small example showing QTimer slowing to a halt

CuppoJava

Hello.

I managed to isolate my bug down to a very small test case, and was wondering whether I'm doing something wrong, or whether QT has a bug.

I'm trying to set up a QTimer that just prints "tick" every 33ms. When I ran the program, it starts off doing that fine. After running for about 2 minutes, the ticks slow down to about once every 10 seconds.

I feel I'm hitting a memory-leak, or some buffer is getting really long and I'm not clearing it.

This is QT4.8 running on Mac OS-X 10.11.6.

Thanks for your help!
-Patrick

In test.h:

#ifndef TEST_H
#define TEST_H
#include <QObject>

class Driver : public QObject{
  Q_OBJECT
public:
  Driver();
public slots:
  void tick();
};

#endif

In test.cpp:

#include "test.h"
#include <iostream>
#include <QApplication>
#include <QTimer>
using namespace std;

Driver::Driver(){}
void Driver::tick(){
  cout << "tick" << endl;
}

QApplication* app;
Driver* driver;
QTimer* timer;
int main(int argc, char** argv){
  app = new QApplication(argc, argv);
  app->setQuitOnLastWindowClosed(false);

  driver = new Driver;
  timer = new QTimer;
  timer->setInterval(33);
  QObject::connect(timer, SIGNAL(timeout()), driver, SLOT(tick()));
  timer->start();
  
  return app->exec();
}

TobbY

Hello @CuppoJava
Your code is perfectly fine. Didn't find any issue with memory leak. Maybe this Qt version contain bug for Mac OS.

SGaist

Hi and welcome to devnet,

What version of Qt 4.8 ?

By the way, there's no need for all these static variables that you are using especially app.

int main(int argc, char** argv){
    QApplication app(argc, argv);
    app.setQuitOnLastWindowClosed(false);

    Driver driver;
    QTimer  timer;
    timer.setInterval(33);
    QObject::connect(&timer, SIGNAL(timeout()), &driver, SLOT(tick()));
    timer.start();
  
    return app.exec();
}

That way you don't have any memory leaks.

On a side note, Qt 4.8 has reached end of line since a couple of years now, so if possible, you should rather start your new projects with Qt 5.

CuppoJava

Thanks very much for the help. I'll upgrade to QT5 to see whether that solves it. I'm working on a client project right now that's written in QT4, so if it turns out to be a bug in QT I'll have to find a workaround.

CuppoJava

Just as an update, I upgraded to QT5 and the problem is still there. At this point, I almost think that it's a bug with the OSX terminal, except that I can't reproduce the problem using anything except QTimer. Does anyone have any suspicions that I can use as a starting point?

Thanks very much!
-Patrick

mranger90

I would put an QElapsedTimer in the driver and look at the actual interval.
If the actual interval is still at the expected time, but the display is off, then
its the terminal not being able to keep up.

J.Hilk

@mranger90 said in Small example showing QTimer slowing to a halt:

I would put an QElapsedTimer in the driver and look at the actual interval.
If the actual interval is still at the expected time, but the display is off, then
its the terminal not being able to keep up.

That's a great point, forcing a clear on the terminal may "reset" the delay

CuppoJava

That is a great tip. Thanks I'll run that experiment!

CuppoJava

Okay. I ran the experiment. I started a QElapsedTimer at the same time as my QTimer (which is set to an interval of 33 ms) and printed out the elapsed time between each call. And it prints out the following, which shows a steady interval of 33ms followed by a sudden period where the interval jumps up to about 10s.

So I guess this rules out the OSX terminal as the cause of the slowdown. I verified that I see the same behaviour using both QT4 and QT5. Do you guys have any other ideas? I'm completely stumped.

Thanks!
-Patrick

[...]
tick 889
elapsed time: 33
tick 890
elapsed time: 33
tick 891
elapsed time: 32
tick 892
elapsed time: 32
tick 893
elapsed time: 32
tick 894
elapsed time: 32
tick 895
elapsed time: 32
tick 896
elapsed time: 10033
tick 897
elapsed time: 8457
tick 898
elapsed time: 4837
tick 899
elapsed time: 38
tick 900
elapsed time: 10030
tick 901
elapsed time: 10032
tick 902
elapsed time: 10031
[...]

KonstantinosTsakalis

@CuppoJava @SGaist Hi. Thank you both for your contribution. I'm having the same problem in Qt 6.2.4 . In my application which is running on an embedded device with a Linux kernel, we use a software watchdog which must be kicked by the GUI every 5 seconds.

Along with that, the data of the GUI are in communication with another processor. For the synchronization of the two, we use a QTimer with an interval of 1 ms. The application is running ok for some minutes, but eventually fails to kick the watchdog on time and is being restarted by systemd.

Using gammaray I was able to inspect the same behavior mentioned on this topic. The 'busy' timer is missing wakeups as the time passes by, and its interval is also being increased, leading to the delay I am mentioning.

The problem seems to go away if I increase the interval to 100 ms. Is there any restrictions to as how low a QTimer's interval can be?

I think I have located the issue being that the Event Loop of the main GUI thread is flooded by the QTimer's timeouts and thus cannot cope with processing the rest of the events on time. I have also considered moving this operation to another thread, but the data synchronization is not thread safe.

Do you have any updates on this topic and maybe how you solved it?

JonB

@KonstantinosTsakalis
I am not a realtime/embedded person. But at least read Accuracy and Timer Resolution, maybe also Alternatives to QTimer. Maybe a Qt::PreciseTimer or a QBasicTimer would help?

I don't think the default QTimer is to be relied on at 1ms, it's not intended for that.

KonstantinosTsakalis

@JonB Thank you for your answer. My problem is not really the resolution/accuracy of the timer, rather the fact that this fast timeout causes the whole application to delay its execution. But your reference helped me to find out that there are more than one types of QTimers, so thank you!

J.Hilk

@KonstantinosTsakalis have you considered making it not a continuously running but a "one shot" timer. And you simply restart the timer at the end of the connected slot ?

should prevent the event queue overflow

KonstantinosTsakalis

@J-Hilk That sounds indeed like something that can work.. I will try it, thank you!

I considered stopping the timer in the beginning of the slot lambda, and restarting it at the end, but then I would be altering the sender inside the slot, and this didn't seem a good practice to me.

J.Hilk

@KonstantinosTsakalis QTimer has the singles shot property
https://doc.qt.io/qt-6/qtimer.html#singleShot-prop
simply set that to true

KonstantinosTsakalis

Thank you all for your contribution. Your comments helped me to understand better how the event loop and QThread are working.

It seems that my blocking point is not the communication with the other processor. The problem occurs when we log data to journald. The "why" is yet unclear, but for some reason logging to journald can occasionally take up to more than 10 seconds, so this is explaining why the application fails to kick the watchdog on time.

I am now exploring other ways to implement this. Currently I have moved the logging functionality on a separate thread and this seems to work fine.

I hope my description helps a future reader.

Thank you.

mliberato-pax

Posting here just in case someone comes across this -- I ~think~ I found the solution to be related to app nap, originally discovered here. My application was in python, so I was able to use appnope to ensure my python application wasn't throttled & this solved my manifestation of this problem.