Tweaking system settle tests in smoketesting

[Łukasz Zemczak]

In the current state of things, all the system settle tests that are visible in smoketesting are failing on our krillin devices. There might be a few reasons for that. The first thing visible is that the idle levels of krillin devices are different than mako ones - but maybe the settle tests themselves run differently than expected.

We need someone to investigate and fix this, either by tweaking the thresholds or changing the way they're ran.

[Brendan Donegan (brendan-donegan)]

<ogra_> wget -O- -q http://dev-jenkins.ubuntu-ci:8080/job/utopic-touch_stable-krillin-smoke-daily/433/artifact/clientlogs/sudoku_app/topbefore.log/*view*/|grep -v "top -c -b -d 6 -n 5"| grep PID -A3|grep -v ^PID
<ogra_> this is a better command ... filterin out top itself first
...
<ogra_> brendand, so looking at it with the ablve command (and ignoring the first three runs where the session still settles down) i see dbus-daemon, upowerd and indicator-power being the busy ones ... with an occasional spike of unity8
<ogra_> which makes me think if mako would look different that krillin is simply very chatty with upowerd
<ogra_> keeping the dbus noisy
<ogra_> thats just a guess thogh, we have to compare mako

[Brendan Donegan (brendan-donegan)]

Lets adjust the settle value to 96.5

[Łukasz Zemczak (sil2100)]

As per discussion +1 on that.

[Łukasz Zemczak (sil2100)]

Ok, so it seems we still don't know if this helped because we have scoperunner crashes for each test suite, which is causing whoopsie to change the idle levels for each test.