[Tracker] memory leaking after launching theScheduler

[diver]

I'm not really sure what component it should be.

After starting this 32bit app on 64bit(!) Haiku hrev53013 I observe high CPU usage in these two threads:

registrar -> main_mime
Tracker -> launch looper

​http://www.pinc-software.de/files/theScheduler-1.1.0.zip

[axeld]

Have you tried other 32-bit apps, if they behave in the same way? There shouldn't be anything special with this application, at least.

And you say "memory leaking" in the title, and high CPU usage in the summary -- is both happening? Is the CPU usage staying high all the time after that? Is memory continually decreasing? Or does it stop after a while?

[diver]

Replying to axeld:

Have you tried other 32-bit apps

Just tried 43bit BePDF and it fails to start as expected with "Not an executable" message. (Side note: would be nice to detect architecture and show "You're trying to run 32bit executable which is not supported.")

And you say "memory leaking" in the title, and high CPU usage in the summary -- is both happening?

Yes, sorry for confusion.

Is the CPU usage staying high all the time after that?

Yes.

Is memory continually decreasing? Or does it stop after a while?

No and no.

See attached screenshot:

[diver]

Restarting Tracker fixes both CPU usage of registrar + Tracker and memory leaking.

[CodeforEvolution]

I believe I've experienced an issue like this too. When trying to open an app with incompatible symbols with the version of Haiku being used, I see the app continuously appear and disappear from Process Monitor along with a high CPU usage from Tracker and registrar. The only way to stop this loop for me is to kill registrar itself. When registrar is finally killed, an error message of why the program couldn't be opened appears (not related to the registrar not being available). Of course, this forces me to reboot, as once registrar is closed, it can't be restarted, and the system is unable to launch any more programs. So, I definitely believe this is some sort of conflict between registrar and runtime_loader.

[diver]

I managed to reproduce it by running /system/apps/Qt5/qdoc from Tracker but only once.

[waddlesplash]

I just realized hrev52968 may be related here, but I didn't investigate too closely.

[waddlesplash]

#11582 may also tie into this somehow.

[waddlesplash]

Fixed in hrev53399.

[nielx]

Assign tickets with status=closed and resolution=fixed within the R1/beta2 development window to the R1/beta2 Milestone