On Fri, 2017-01-20 at 17:33 +0000, Fedora compose checker wrote:
x86_64 Workstation-boot-iso memory_check peak memory: 717MiB
x86_64 Workstation-boot-iso memory_check@uefi peak memory: 715MiB
So this is another shiny new openQA / check-compose feature: there's
now a test which just runs an install with the `debug` kernel param,
which triggers anaconda to record memory usage throughout the install
process, then uploads the usage data file.
check-compose then notices jobs which do this, grabs the memory usage
data file, and includes the peak usage during installation in the
report. Tomorrow, it should include a comparison to today's usage (we
don't get the comparison today, because today's the first compose we
have the data for). If the usage changes significantly and we want to
know why, we can go check the raw data file for more information (I
guess I should enhance check-compose to link to it).
For the moment we're just running this test on the Workstation boot
ISO, as it's likely to be the most memory-intensive install in openQA's
set (network installs use more memory than DVD installs, and the
Workstation network install has more packages than the Server network
As for the test results, there is a whole bunch of stuff going on, too
much to break out test-by-test. Here are the highlights:
1) There is some kind of systemd bug which essentially means that
occasionally, the system boots to emergency mode instead of booting
properly. This affects about a half dozen tests every day, on average -
never the same ones, the bug is quasi-random, it just...happens
sometimes. That's https://bugzilla.redhat.com/show_bug.cgi?id=1414904
I'm currently trying to pin that one down.
2) There is an AVC that appears on boot on just about all installs of
Rawhide at present:
this causes the large number of soft failures, because many openQA
tests now run a check for any AVCs or crash notifications after
installation, and if any are found, it's considered a soft failure.
3) All i686 installs (and live image boots...really, Rawhide is just
entirely broken on i686) are broken and have been since (IIRC) systemd
v232. I think someone pointed me to a bug which may be the cause of
this, but I've lost the reference. It's on my list to figure out soon.
4) KDE upgrades are running into a dependency issue: 'kde-runtime-
drkonqi-16.08.3-3.fc24.x86_64 requires kde-runtime = 16.08.3-3.fc24,
but none of the providers can be installed'. I haven't reported this
properly yet, sorry, I should.
5) Russian installs are failing because of an anaconda bug:
6) The desktop_browser test needs a needle update, I'll take care of
7) All the 'check if all services start correctly' tests for various
images fail because of this SELinux/systemd issue:
It occurs to me that we have kind of a recurring story: 'new systemd
release lands in Rawhide, stuff breaks because it does stuff SELinux
doesn't expect it to'. It might be nice if we could organize some kind
of co-ordination between systemd and SELinux folks such that the
appropriate SELinux permissions get added *before* the new systemd
release lands. Of course, it would make things easier if we could
provide upstream systemd with a nice easy way to test their git master
on Fedora Rawhide, I guess.
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net