Bug 428048

For contrast, I haven't experienced this problem after running KSysGuard for 3+ hours.

The only difference in versions, apparently, is that here in Mageia I have Qt 5.15.1, which maybe fixes something in 5.15.0 that "breaks" KSysGuard...

(In reply to JanKusanagi from comment #1)
> For contrast, I haven't experienced this problem after running KSysGuard for
> 3+ hours.
> 
> The only difference in versions, apparently, is that here in Mageia I have
> Qt 5.15.1, which maybe fixes something in 5.15.0 that "breaks" KSysGuard...

As I am writing this, ksysguard is the process consuming most RAM on my system, standing at 673 MB. It has been open since this morning around 9, which makes it roughly 13 hours. I found this is reliably reproducible as I have experienced it ever since I updated to 5.20.0 without fails.

I can confirm,


Plasma 5.20.1
Frameworks 5.75.0
Qt 5.15.1

Memory used by KSysGuard will steadily increase and can reach 2GiB overnight.

I also suspect this to have started happening for me since version 5.20.0.

There is also https://bugs.kde.org/show_bug.cgi?id=426792 indicating the same for libksysguard, though it was originally reported with 5.19.90.

Still no issue here. KSysGuard running for 9h17m, using 32,6 MiB.

In case the other bug report isn't monitored:


** https://github.com/KDE/libksysguard/commit/40cb8d6acbb12e516c1d327a9b2574560adcb4ec seems to be working without issues for me

- https://github.com/KDE/libksysguard/commit/23852597c642a5c5303bb994d4d28e20d9c863cb will not compile for me with "error: ‘Updates’ in ‘class KSysGuard::Process’ does not name a type" - like errors

- newer commits will compile ok, but they start eating memory like described


how is it possible that some of us do and others don't experience these problems..

So, 

if a running system updates to a 'bad' version of libksysguard, the effects will not show until the system is rebooted. If a bad version gets replaced with a good one, reboot isn't necessary somehow.

I found that commit 23852597c642a5c5303bb994d4d28e20d9c863cb there might be a problem deleting the objects that are created....

I patched this on my end by changing processcore/read_procsmaps_runnable.cpp to 

-setAutoDelete(false)
+setAutoDelete(true)

I could also just git revert the entire commit and things are back to normal.

I'm having a hard time believing that some will suffer from this and others don't. Reboot if you dare ;)

A possibly relevant merge request was started @ https://invent.kde.org/plasma/libksysguard/-/merge_requests/90

>how is it possible that some of us do and others don't experience these problems..


Based on what I think it is, probably the status bar or something else doing a process list fetch.

Git commit 9b5161cb9a16eaad205631b2a4c382dbbbe2072c by David Edmundson.
Committed on 25/10/2020 at 23:00.
Pushed by davidedmundson into branch 'master'.

Handle smap read result in the correct thread

We set the context to "this" so we should auto disconnect when the
Processes object is destroyed.
But Qt::DirectConnection means that is run in the runner thread.

It is possible to have a scenario where "this" is active when the
connection is made, but destroyed before we get to this line.
Related: bug 428160

But scoping the connect to Processes existing we would have leaked the
worker if Processes got destroyed whilst the worker was running.

M  +8    -5    processcore/processes_linux_p.cpp
M  +3    -8    processcore/read_procsmaps_runnable.cpp
M  +1    -4    processcore/read_procsmaps_runnable.h

https://invent.kde.org/plasma/libksysguard/commit/9b5161cb9a16eaad205631b2a4c382dbbbe2072c

Git commit 2d522614357d6dc5a5a7b738a35574f5d8f69da5 by David Edmundson.
Committed on 26/10/2020 at 17:25.
Pushed by davidedmundson into branch 'Plasma/5.20'.

Handle smap read result in the correct thread

We set the context to "this" so we should auto disconnect when the
Processes object is destroyed.
But Qt::DirectConnection means that is run in the runner thread.

It is possible to have a scenario where "this" is active when the
connection is made, but destroyed before we get to this line.
Related: bug 428160

But scoping the connect to Processes existing we would have leaked the
worker if Processes got destroyed whilst the worker was running.


(cherry picked from commit 9b5161cb9a16eaad205631b2a4c382dbbbe2072c)

M  +8    -5    processcore/processes_linux_p.cpp
M  +3    -8    processcore/read_procsmaps_runnable.cpp
M  +1    -4    processcore/read_procsmaps_runnable.h

https://invent.kde.org/plasma/libksysguard/commit/2d522614357d6dc5a5a7b738a35574f5d8f69da5

This is similar to the comment in #6, only with some more robustness on the syncing to the main thread. 

Lets see if this resolves everything

(In reply to David Edmundson from comment #11)
> This is similar to the comment in #6, only with some more robustness on the
> syncing to the main thread. 
> 
> Lets see if this resolves everything

Thank you for your help, David. As soon as 5.20.2 is out on Neon I will let you know if the bug was solved.

*** Bug 428027 has been marked as a duplicate of this bug. ***

@David: 'more robustness' was a very kind way of understating my time-bomb ;)


commit 9b5161c does not seem to resolve the issue for me, I tried both master and 5.20.2.

Meanwhile I returned to using e5f0f06 and reverting 2385259 out of that, for a non-explosive solution.

I can confirm 5.20.2 does not fix the issue so I am marking the bug report as "reopened".

Aha, there's another obvious place for leaking.

   QFile file{m_dir + QStringLiteral("smaps_rollup")};
    if (!file.open(QIODevice::ReadOnly)) {
        return;
    }


We never hit emit finished(), no-one cleans us up.
Also explains why it's some users and not others.

A possibly relevant merge request was started @ https://invent.kde.org/plasma/libksysguard/-/merge_requests/91

Git commit ebcf87527c32db0b6685c63a450489d2c6727069 by David Edmundson.
Committed on 29/10/2020 at 11:07.
Pushed by davidedmundson into branch 'master'.

Autodelete smapsRunnable

The runnable has an early return on error conditions. This means
finished() might never be called.

Now we've removed the accessor and provide the return value in the
signal we can just set autoDelete.

M  +0    -2    processcore/processes_linux_p.cpp
M  +1    -1    processcore/read_procsmaps_runnable.cpp

https://invent.kde.org/plasma/libksysguard/commit/ebcf87527c32db0b6685c63a450489d2c6727069

Git commit 5e48a74d5f2b0cd271b4edc8308dabd8d2a6fd79 by David Edmundson.
Committed on 29/10/2020 at 15:24.
Pushed by davidedmundson into branch 'Plasma/5.20'.

Autodelete smapsRunnable

The runnable has an early return on error conditions. This means
finished() might never be called.

Now we've removed the accessor and provide the return value in the
signal we can just set autoDelete.


(cherry picked from commit ebcf87527c32db0b6685c63a450489d2c6727069)

M  +0    -2    processcore/processes_linux_p.cpp
M  +1    -1    processcore/read_procsmaps_runnable.cpp

https://invent.kde.org/plasma/libksysguard/commit/5e48a74d5f2b0cd271b4edc8308dabd8d2a6fd79

Yes, that does it, thank you very much ;)

Solves the issue. Thx.

*** Bug 424299 has been marked as a duplicate of this bug. ***

*** Bug 426792 has been marked as a duplicate of this bug. ***