The Gating spec[1] used to recommend having one type of
YAML stanza for master, one for released branches. We
just discovered today that this breaks when master
forks into a release branch: the string 'stable'
needs to be manually changed to 'testing', and
nobody is going to remember to do that, and I can't
think of a (non-tedious) way to add automated checks.
As of 2019-08 (09d7d4f), though, the Gating spec suggests
having *both* stanzas together in the same gating.yaml
file. @pingou himself concurs[2]. Although this is
horrible duplication -- two hundred characters, repeated
with only six of those different -- it seems cleaner
than the maintenance burden of having to update
release branches.
[1] https://docs.fedoraproject.org/en-US/ci/gating/
[2] https://src.fedoraproject.org/rpms/podman/pull-request/33
Oh, also, add a new 'subject_type' line apparently "required"
even though things seem to have been working fine without it.
(Source: same commit, 09d7d4f, on gating.adoc).
Signed-off-by: Ed Santiago <santiago@redhat.com>
The first document I went with[1] specified a magic
value containing 'package.test.functional'. I used
that, and my bodhi results[2] looked weird: first,
none of the usual results include 'test.functional';
so, second, there's now a new test.functional line
under 'undefined' in red and with an asterisk next
to it, and another red one without an asterisk.
All four lines seem to link to the same results.
[1] https://docs.fedoraproject.org/en-US/ci/gating/
[2] https://bodhi.fedoraproject.org/updates/FEDORA-2019-144ed359e9
A different document[3] specifies a magic value that
more closely conforms to the ones already shown
in the 'undefined' section of older bodhi runs.
That's what I'm trying now.
[3] https://docs.fedoraproject.org/en-US/rawhide-gating/optin/
Unfortunately neither document has a timestamp so it's
impossible to know which (if any) is up-to-date and/or
correct.
Signed-off-by: Ed Santiago <santiago@redhat.com>
Purpose: prevent a build from being pushed to stable
unless tests/tests.yml pass; require manual override
on failure.
Counterargument: this is going to fail effective now.
Rawhide has been highly unstable, with kernel and
systemd bugs one right after the other, all preventing
podman tests from passing. (Current blocker: rhbz1740664
in systemd-udev, in which /dev/net/tun is mode 600
which causes slirp4netns to fail).
Countercounterargument: even if all these failures aren't
podman's fault, we still want to catch them and at least
report them early. I am monitoring test runs but I don't
want to be the single point of failure.
Signed-off-by: Ed Santiago <santiago@redhat.com>
