Re: OpenQA critpath Bodhi tests for packages that are most likely not used during the tests
On Fri, 2023-03-03 at 22:24 +0100, Miro Hrončok wrote:
On 03. 03. 23 20:10, Adam Williamson wrote:
> On Fri, 2023-03-03 at 18:37 +0100, Miro Hrončok wrote:
> > Hello,
> > I've noticed in
> >
> > https://bodhi.fedoraproject.org/updates/FEDORA-2023-f34afe57a9
> > https://bodhi.fedoraproject.org/updates/FEDORA-2023-a61dbc3c1d
> > https://bodhi.fedoraproject.org/updates/FEDORA-2023-a61dbc3c1d
> >
> > That the update is marked as critpath (which is probably correct because
> > python3-devel Requires it and many Python package BuildRequire it) and hence
> > the OpenQA tests run. However the probably unused in those tests, considering
> > it contains RPM macros.
> >
> > Unless the OpenQA tests actually rebuild everything involved with this package
> > (which I doubt, because it's not that slow) before running the tests, what
is
> > the point of running them for this update? Should it not be in critpath? Or
> > should such tests only run if the produced package would be installed in the
> > test environment even when not explicitly installed?
>
> There is no point, but there may be nothing much we can do about it.
>
> I have been improving things in this area recently with a lot of work
> under the general heading "grouped critical path", but if you mouseover
> the critpath icon for the update, you'll see the package is in the core
> group, and the kde, gnome and server critical path groups (which means
> something explicitly listed in each of those groups ultimately depends
> on it).
>
> If something is in the core group, we kind of have to schedule all the
> openQA tests for it. Things in core could potentially break *anything*.
> It is an unavoidable hazard that, sometimes, via long dep chains, this
> results in us running the tests on things they don't really cover.
>
> Similarly, we have to run the GNOME tests for things in the GNOME
> critpath group, the KDE tests for things in the KDE critpath group, and
> the Server tests for the things in the Server critpath group. So even
> if we manage to get it out of @core , we would still have to run the
> KDE, GNOME and Server tests on it unless we could get it out of those
> groups too.
>
> So the question really is, what causes it to be in those groups? I
> suspect it's something in the dep chain of pyproject-srpm-macros , but
> at a quick glance I can't figure out what. If anyone has a good dep
> visualizer they can throw at the problem it'd help, I guess.
Thanks for the answer. I think it's:
# dnf install @critical-path-build
...
Installing group/module packages:
redhat-rpm-config
...
Installing dependencies:
pyproject-srpm-macros
...
So the question is: Are there test that test building RPM packages? If not, why
is @critical-path-build included?
No, that's a misunderstanding. critical-path-build is one of the very
new groups I created recently specifically to *help* with this problem.
critical-path-build exists precisely so that openQA can *not* run tests
on packages that are (only) in that group.
The problem here is that the package is in a bunch of critpath groups,
including the key ones I listed above - @core, @critical-path-kde,
@critical-path-gnome, @critical-path-server. We run all the openQA
tests for anything that's in @core, because, well, we kinda have to. We
also run the openQA KDE tests on anything that's in @critical-path-kde
, the GNOME tests on anything that's in @critical-path-gnome, and the
Server tests on anything that's in @critical-path-server (for obvious
reasons). So as long as it's in @core, *all* the tests will be run on
it. If we get it out of @core but it's still in the other three groups,
we'll still run the KDE, GNOME and Server tests on it (which is nearly
all of the tests).
Interestingly, this is only the case for the F38 update. Check the F39
update:
https://bodhi.fedoraproject.org/updates/FEDORA-2023-23789cd657
it's *only* in @critical-path-build, it's not in @core or the kde,
gnome or server groups. So, as you can see, no tests are run on it.
I know why this would happen for F36 or F37, but I'm surprised it's
happening for F38. F38 and F39 ought to behave basically the same so
far as this stuff is concerned. I'll poke around a bit and see if I can
figure out why.
--
Adam Williamson (he/him/his)
Fedora QA
Fedora Chat: @adamwill:fedora.im | Mastodon: @adamw(a)fosstodon.org
https://www.happyassassin.net