Memory being gobbled up about qtile HOT 12 CLOSED

A little more investigation. By just running ps aux |grep qtile I can watch qtile grow by about 4 bytes/sec. So I disabled the charts (CPU/etc) and ran a Xephyr instance. It remained constant while the instance with the charts continued to grow.

I guess I'm at ~124Megs after 18 hours.

from qtile.

I'm on the trail of a similar leak. I'm not seeing a gradual growth in memory usage, but every now and agian, I come down in the morning to find qtile has grown to about a gig. My gut feeling is that this is a leak in an underlying library, which is going to make it a bugger to track down.

from qtile.

I'm assuming that the culprit may be cairo/pycairo then? I haven't looked into either. Is upstream responsive? Are there tests to run valgrind against cairo?

from qtile.

Yes, I think cairo/pycairo is the likely culprit if the problem is definitely related to the graph wigets. I haven't gone to the effort to run valgrind against cairo, so I'm not sure. Sigh, I guess it's going to be necessary to do this if we can't find the problem by carefully eliminating calls.

from qtile.

I'm still seeing ~4b/sec when I have charts enabled and no growth when they are disabled.

I tried to see if I could get a simple testcase to replicate here:

https://gist.github.com/727312

This just draws directly to the context and bypasses the drawer layer. It tries to chart a bunch of points, but it is NOT exhibiting memory growth. Note that some of the drawer methods are missing. Perhaps the drawer layer is where the leaking is occurring?

from qtile.

Also tried the following. Though it grew in memory while running, it didn't report orphaned python objects very often. So I'm assuming the Python side is behaving ok.

https://gist.github.com/734308

from qtile.

I've investigated this some time ago, and IIRC it's memory leak in xpyb library. And I've stopped at attempts to guess what xpyb is newest and right one to use :) I'll give a try once more later.

from qtile.

Other reports of pycairo memory chomping...
http://dotsony.blogspot.com/2011/06/calendar-update-part-ii-after-release.html

from qtile.

@mattharrison, this particular memory leak was fixed in qtile by making textlayout kinda singleton. I'm still sure it was xpyb leak in my case.

from qtile.

I've tracked down minimal testcase for memory leak in xpyb. Here is a bug with a testcase:

https://github.com/dequis/xpyb-ng/issues/5

BTW, Are these recomended bindings?

from qtile.

Hi all,

I believe upstream has fixed this issue. I've merged their work into xpyb-ng and posted it here: https://github.com/tych0/xpyb-ng

I'll try and get some if this stuff into xpyb-ng on the qtile org soon, but for now that repo should work and doesn't leak nearly as fast (although it seems like it still leaks, unfortunately...).

from qtile.

Can we make sure that @dmpayton's docs are updated to work with this?

from qtile.