openQA test failures due to typing errors
Hi folks! Another note for anyone paying attention to openQA test
results.
Since 2017-08-09 there's been kind of a flood of failures caused by
'typing errors' - that is, when the test runner is trying to type a
string into the test VM and it doesn't get through correctly (usually
due to one or more keypresses being dropped). We've always seen this
kind of failure *very occasionally* in openQA tests, but it suddenly
become massively more common (as in, half the tests for every update
were failing).
With some invaluable help from Cole Robinson I'm pretty sure we have
this figured out now; it was caused by a change to qemu which was
introduced to address a potential denial-of-service issue. For now I've
reverted qemu to the known good older version on the openQA worker
hosts (there isn't any realistic vector for anyone to cause any harm by
exploiting that DoS in the case of the openQA deployment, or any of the
other similar issues the updated qemu fixed), and we've identified some
later upstream commits that look like they may well resolve the
problem, so we should hopefully be able to put a more permanent fix in
place soon.
I'll be re-running all the tests that have run since 2017-08-09 with a
working qemu, so we have more accurate results.
Very sorry for this problem; I'd usually have noticed and addressed
this sooner, but it came at an unfortunate time.
While I'm here - there's also a consistent failure in the
'desktop_browser' test which is just caused by a screenshot that needs
updating. I'll get on that ASAP.
--
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net