High CPU on latest nightly
Adam Williamson
adamwill at fedoraproject.org
Wed Mar 4 01:40:56 UTC 2015
On Tue, 2015-03-03 at 17:00 -0800, Adam Williamson wrote:
> On Tue, 2015-03-03 at 04:59 -0500, Kamil Paral wrote:
> > > On Mon, 2015-03-02 at 22:46 +0000, Alexander Bisogianis wrote:
> > > > > I'm not seeing anything like that on my F22 desktop.
> > > > >
> > > >
> > > > I mean while installing F22, all four CPUs on the KVM host
> > > > (F21) are constantly at 100% usage.
> > > >
> > > > Running top in TTY2, while F22 is installing, I see:
> > > >
> > > > Xorg using constantly CPU 60%-70%
> > > > 2 anaconda processes using constantly CPU 60%-70%
> > > >
> > > > I can understand anaconda, but Xorg?
> > >
> > > IIRC there's a spinner or something similar visible during
> > > install; it could well be caused by that. I even think I caught
> > > some discussion about a patch to disable it during automated
> > > test runs or something in #anaconda the other day...
> >
> > Alexander, try running the installation with this boot option:
> >
> > inst.updates=https://kparal.fedorapeople.org/tmp/no-spinner.img
> >
> > It should disable spinner animation. Then compare the CPU usage. I
> > still haven't gotten to reporting it as a bug, but it's true that
> > the spinner seems to be a *gigantic* performance hog, making all
> > our installations take much longer, especially in VMs.
>
> Now I come to notice it, GTKSpinners don't seem to spin at all in
> TC7 and TC8. They do spin on my desktop. Not sure if there's a
> package difference or it's a KVM vs. real hardware thing, but
> they're broken with at least both 'vga' and 'qxl' in a KVM. Just
> burning a USB stick to see if it's the same on bare metal. Probably
> GTK+ 3.15.9 is involved.
Huh, it's odder than that - GtkSpinners apparently don't spin in GNOME
on a KVM, but they *do* spin in the installer environment.
https://bugzilla.redhat.com/show_bug.cgi?id=1198393
--
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net
More information about the test
mailing list