Executing QProcess in QThread: memory leak

[mrjj]

Hi
How do you know there is a leak ?

[sitesv]

@mrjj said in Executing QProcess in QThread: memory leak:

Hi
How do you know there is a leak ?

By system monitor...
A leak is about 4k/sec. I have been checking for a few minutes...

[mrjj]

@sitesv
Ok. it might internally allocate something
that is first freed on app termination.

If you reuse the same QProcess
(let it be a member of the TMyThread, allocate once)

Does it still then still leak ?

[sitesv]

@mrjj said in Executing QProcess in QThread: memory leak:

Does it still then still leak ?

Yes, there is still a leak.
QProcess myProcess; moved into the class.

[sitesv]

@mrjj
If I would use this code in while(1) of the 'main' function of the console app - there is NO memory leak.

[mrjj]

@sitesv
but you only create the thread once , so it should be the same ?

[sitesv]

@mrjj said in Executing QProcess in QThread: memory leak:

but you only create the thread once ,

Yes.

[mrjj]

Hi
Hmm really strange then.
Is this on windows ?
I did something similar on linux and didnt notice it would leak 4K

[Christian Ehrlicher]

@sitesv said in Executing QProcess in QThread: memory leak:

By system monitor...

This is no valid memory leak checker. Use a correct to to find leaks like e.g. valgrind.

[sitesv]

@mrjj said in Executing QProcess in QThread: memory leak:

Is this on windows ?

Linux

@Christian-Ehrlicher
Is it correct, If a memory value is increasing for a few minutes?

[Christian Ehrlicher]

@sitesv A system monitor is not a valid tool for such an investigation. It may be a hint that there is something wrong but nothing more. The system monitor only shows what the OS gave the process, but the OS may not free resources immediately but only when it needs it later on so use a correct tool for this.

[sitesv]

@Christian-Ehrlicher
Checked app by Valgrind. There is no memory leak...
Thank you. I hope there really are no errors.

[mrjj]

@sitesv
I would leave it running for like 48 hours and see it would stabilize.
If its related to actually running the process on Os level, it might only allocate up to some max.

[Christian Ehrlicher]

Since Qt cleans up a lot of stuff during exit maybe also stop the application started with valgrind with CTRL+C and take a look at the reachable memory in the valgrind output to see if there is something interesting (start valgrind with '--show-reachable=yes')

[sitesv]

@mrjj, @Christian-Ehrlicher
Thank you!

[sitesv]

Hi!
Made a simple test:

void mythread::run(){
    while(work){
        bool boolRESULT;
        QString output_str;
        QByteArray output;
        QStringList output_strlst;
        QTextCodec *codec;
        QString exe_path = "/bin/ping";;
        QStringList arguments;
        arguments << "127.0.0.1" << "-c" << "1";

        if(myProcess == nullptr) myProcess = new QProcess;
        myProcess->start(exe_path, arguments);
        myProcess->waitForFinished(500);

        output = myProcess->readAll();
        output_str = codec->toUnicode(output);
        output_strlst = output_str.split("\r\n");

        myProcess->close();
         
        boolRESULT = false;
        for (int i = 0; i < output_strlst.count(); i++){
            boolRESULT = output_strlst[i].contains("ttl",Qt::CaseInsensitive);
            if (boolRESULT) break;
        }
        emit setStatus(boolRESULT);
   
        msleep(1000);
    }
}

Valgrind output:
...
==23633== 2,400 bytes in 15 blocks are still reachable in loss record 2,049 of 2,114
==23633== at 0x4C2BBEF: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==23633== by 0x6267251: resizeSignalVector (qobject_p.h:297)
==23633== by 0x6267251: QObjectPrivate::addConnection(int, QObjectPrivate::Connection*) (qobject.cpp:330)
==23633== by 0x6268A18: QMetaObjectPrivate::connect(QObject const*, int, QMetaObject const*, QObject const*, int, QMetaObject const*, int, int*) (qobject.cpp:3453)
==23633== by 0x626FD58: QObject::connect(QObject const*, char const*, QObject const*, char const*, Qt::ConnectionType) (qobject.cpp:2911)
==23633== by 0x61BFB1F: QProcessPrivate::startProcess() (qprocess_unix.cpp:384)
==23633== by 0x61BA34D: QProcessPrivate::start(QFlagsQIODevice::OpenModeFlag) (qprocess.cpp:2246)
==23633== by 0x61BA582: QProcess::start(QString const&, QStringList const&, QFlagsQIODevice::OpenModeFlag) (qprocess.cpp:2094)
==23633== by 0x10BBBF: mythread::run() (mythread.cpp:19)
==23633== by 0x605A3B4: QThreadPrivate::start(void*) (qthread_unix.cpp:342)
==23633== by 0x6A164A3: start_thread (pthread_create.c:456)
==23633== by 0x75B1D0E: clone (clone.S:97)
...

mythread.cpp 19 line is:

myProcess->start(exe_path, arguments);

My log by system monitor:

• 16:50 4300 KB
• 17:06 4790 KB
• 17:20 5200 KB
• 17:30 5552 KB

Any ideas?

[KroMignon]

@sitesv said in Executing QProcess in QThread: memory leak:

Made a simple test:

You are aware that doing so (subclassing QThread and redefining slot run()), there will be not QEventLoop to handle signals/slots for all QObjects running in this thread?
For example QTimer::singleShot() will not work.

[JonB]

@sitesv
In addition, from @KroMignon no QEventLoop also means no deleting/disposing/freeing of Qt objects.

On a side note: nothing to do with your findings, but for the record

 arguments << "127.0.0.1" << "-c 1";

is wrong: -c & 1 are separate arguments, it ought to be << "-c" << "1". You are lucky it works as one argument.

[sitesv]

@KroMignon @JonB
Are there any methods to do this correctly?
If I do this in the main thread: the app will freeze while QProcess execution - this is unacceptable for me.

[KroMignon]

@sitesv said in Executing QProcess in QThread: memory leak:

Are there any methods to do this correctly?

I don't say this is incorrect.
I depends what you want to achieve.
If you don't need an event queue, it works. If you are using signals/slots, this will not work.

I do not know what is the purpose of this code extract, so maybe it is the best way to do.