QFile and QTextStream on RPI don´t write to File

LogiSch17

Hi guys,
I am just working on a embedded Linux system on a RaspberryPi Compute Module with a small OS build up with buildroot.
I am using QCanBus and QSerialPort and everything works fine.
But when I´m tyring to stream some text to file it makes realy strange things.
I have different parts in my code where i read an write to files.
For example here i am writing a int to a textfile wich is incressed at every system start , so i can count my startups.

int HMIController::getStartCount()
{
    int count=0;
    QString logfile = "/usr/logs/SCount.log";
    QFile file(logfile);

    if(file.open(QIODevice::ReadOnly))
    {
        QTextStream in(&file);
        QString line = in.readLine();
        count = line.toInt();
        file.close();
    }
    if(file.open(QIODevice::ReadWrite | QIODevice::Truncate | QIODevice::Text))
    {
        count++;
        QTextStream out(&file);
        out << count ;
        file.close();
    }
    return count;
}

and this code workes perfectly .
But i got a other function where I am trying to save RFID keys to a text file:

void HMIController::logRfid(const QByteArray &token, int authLevel)
{
    QString filename= "/usr/logs/auth.log";
    QFile file( filename );
    if ( file.open(QIODevice::WriteOnly | QIODevice::Append) )
    {
        QTextStream stream( &file );
        stream << startCount<<":"<<token<<","<<authLevel<< endl;
        file.close();
    }
}

And this function sometimes only workes one of 5 times , or sometimes i wont work at all.
I have tried everything.. Only write one char here-> same result
tried other OpenMethods-> same results
changed rights of the text file -> same results
But the QFile always creats the file.. But i seems that QTextStream wont write crorrectly.
Does someone have a Idea what else i could try?

greetings
LogiSch

kshegunov

Does the function get executed (check with the debugger), is the file opened at all in HMIController::logRfid, as you don't log the (potential) error? What OS (kernel version), Qt (version) and filesystem are you running?

LogiSch17

Hi,
yes the function get´s exectued. When i am changed the stream message to hello world it doesnt work either.
But i already changed the function to bool where i return true after file.close(); it returns the true.
I am using this kernel :
BR2_LINUX_KERNEL_CUSTOM_REPO_URL="https://github.com/raspberrypi/linux.git"
BR2_LINUX_KERNEL_CUSTOM_REPO_VERSION="d33d0293e245badc4ca6ede3984d8bb8ea63cb1a"

and QT5:
BR2_PACKAGE_QT5=y
BR2_PACKAGE_QT5BASE_LICENSE_APPROVED=y
BR2_PACKAGE_QT5BASE_CONCURRENT=y
BR2_PACKAGE_QT5BASE_WIDGETS=y
BR2_PACKAGE_QT5BASE_EGLFS=y
BR2_PACKAGE_QT5BASE_FONTCONFIG=y
BR2_PACKAGE_QT5BASE_GIF=y
BR2_PACKAGE_QT5BASE_JPEG=y
BR2_PACKAGE_QT5BASE_PNG=y
BR2_PACKAGE_QT5SVG=y

did i missed any package?

kshegunov

I really have no idea what those variables refer to. I was asking about version of the related software. As for packages, you should only need QtCore to work with files (and streams), so no, it doesn't appear for you to have anything missing. Currently my best guess is some kind of buffering either by the kernel or the filesystem ...

LogiSch17

ok these variables are out of the buildroot defconfig ..
the kernel wich is linked is version 4.1.
and qt5 is the buildroot package wich contains the qtCore.
After some more tests i think it could be a performance problems. Cause when i let the programm run for 1 min or longer after the function should write to file i works well.
so 12 out of 15 messages were written.
When i run the same programm on my linux PC everything works perfectly.

kshegunov

Most linuxes (ext* filesystems) will defer flushing the journal (especially if the hardware is slow), so my previous supposition stands and it appears that might be the case here. What's the filesystem on your device?

LogiSch17

it´s an ext 4

kshegunov

You could try passing QIODevice::Unbuffered when opening the file and see if that makes a difference, although I wouldn't rely on that flag for production code.

Kind regards.

LogiSch17

I already tried the Unbuffered flag, makes nearly no difference for me. Also tried to flush the stream after writing with same result.

But tanks a lot for your helpt anyway
greetings

kshegunov

@LogiSch17 said in QFile and QTextStream on RPI don´t write to File:

Also tried to flush the stream after writing with same result.

This wouldn't work, as the Qt buffers are flushed implicitly when you invoke endl or close the file.

I already tried the Unbuffered flag, makes nearly no difference for me.

Well, the only other thing I could think of is to try and play a bit with the filesystem's flags, for example (assuming you have SSD storage on the device) passing journal_async_commit option when mounting might make a difference.

LogiSch17

The rpi compute module uses a eMMc flash storage. Do you have any suggestions about changing flags for that?
Cause i dont know much about this stuff.

LogiSch17

Just realised that when i open the files in vim .. at every postion he had failed to write. the file has a line of
^@^@^@^@^@^@....
but i only can see them in vim .. text editor says that file has no common codec .. and when i cat the file it shows only the "good" entries.

SGaist

Hi and welcome to devnet,

You should also add an explicit error message for when the file fails to open. You might get some clues about what is happening.

LogiSch17

Hi and thanks,
I have allready add a debug message on file.open, which never has shown an error . And i message the stream status when the file is open after writing, and QTextStream status() always return 0 for :
QTextStream::Ok 0 The text stream is operating normally.
So the file is always opened correctly and QTextStream seems to write correctly.
Because as i mentioned before, when it doesnt save the right message the file as the same among of chars in the file but it looks like
^@^@^@^@...
So as kshegunov mentioned, i looks like the buffer has not been writen to the file ..
Maybe it is because i never close my programm proberly or shutdown the system correctly and i do not have the possibility to do so.

SGaist

Why can't you shutdown your system properly ?

If you indeed always "pull the plug" that won't help keep your system clean and not only the files you write but the rest also.

What is your exact use case ?

LogiSch17

Hi,
my device is used as a controller and interface for an small kind of electric vehicle. Which is controlled via CAN Bus.
And the the system is turned on and off via one button which is also the emergency off button, so it cuts of the battery completly.

SGaist

How fast and often will you be writing these files ?
How large should they become ?
Also, should they both survive a hard restart ?

LogiSch17

These files were opened at every system startup and they are logging the initialisation of Can devices and some serial an i2c devices , as well as the log the rfid key which was used to unlock the machine.
So yes they should survive cause i would need them for debuging .

SGaist

So basically you need a read-only system but be able to write in some files.

LogiSch17

Thanks for that .
So i have now a second partiton for log files wich is read/write and one for filesystem.
Should make the system much safer.