This comes with a subtle behavioural change that we no longer
pass --nogpgcheck for upgrade tests, but we didn't really *mean*
to do that for anything but Rawhide, and it *shouldn't* be
necessary for Rawhide now, so let's see if everything is fine
without it.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
The bug number is wrong and I can't find the right one, d'oh.
We could *probably* safely remove this right now but I'm not
100% sure, I think it should be fine when F38 is EOL.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
We know this is broken and we don't want it to fail on every
update, so we need to work around the problem for now.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
As 'real' upgrades (using releases/development/rawhide on the
mirrors) do not hit this bug because it has a stale Modular
tree, it makes sense to work around the bug in testing so we can
see if upgrades are broken in any *other* way.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Remove a whole chunk of needles that haven't matched for more
than 3 months. Also move a few needles to appropriate locations,
simplify some code chunks that relied on removed needles (if
we're not matching the needles, we don't need those chunks any
more), and drop some other no-longer-needed conditionals for
older releases.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
I started out trying to fix os-release for the recent change to
add "Prerelease" tags to the VERSION and PRETTY_NAME fields, then
things spiralled. It got me thinking about the awkward DEVELOPMENT
variable we use, so I decided to get rid of it and refactor the
few things that use it. I refactored the anaconda prerelease tag
check, and wrote a new giant comment that gives details about
exactly how anaconda decides whether to show those tags, to give
context to our choices about when to expect them. This check now
uses a new LABEL variable the scheduler now sets. I also wound up
creating new UP1REL and UP2REL vars to define the 'source' release
for upgrade tests, separate from CURRREL and PREVREL, which are
now never lies - they really are the current stable and previous
stable release, even for update upgrade tests.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
This is failing on every update and that's not telling us anything
useful - we already know about the bug - so let's work around it.
Not adding a softfail as it's a bit more awkward to do that.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
The reason we have all this horrible code to use the commented-
out baseurl lines in the repo files instead of the metalinks
that are usually used is a timing issue with the metalink
system. As a protection against stale mirrors, the metalink
system sends the package manager a list of mirrors *and a list
of recent checksums for the repo metadata*. The package manager
goes out and gets the metadata from the first mirror on the
list, then checksums it; if the checksum isn't on the list of
checksums it got from mirrormanager, it assumes that means the
mirror is stale, and tries the next on the list instead.
The problem is that MM's list of checksums is currently only
updated once an hour (by a cron job). So we kept running into
a problem where, when a test ran just after one of the repos
had been regenerated, the infra mirror it's supposed to use
would be rejected because the checksum wasn't on the list - but
not because the mirror was stale, but because it was too fresh,
it had got the new packages and metadata but mirrormanager's
list of checksums hadn't been updated to include the checksum
for the latest metadata.
All this baseurl munging code was getting ridiculous, though,
what with the tests getting more complicated and errors showing
up in the actual repo files and stuff. It occurred to me that
instead of using the baseurl we can just use the 'mirrorlist'
system instead of 'metalink'. mirrorlist is the dumber, older
system which just provides the package manager a list of mirrors
and nothing else - the whole stale-mirror-detection-checksum
thing does not happen with mirrorlists, the package manager just
tries all the mirrors in order and uses the first that works.
And happily, it's very easy to convert the metalink URLs into
mirrorlist URLs, and it saves all that faffing around trying to
fix up baseurls.
Also, adjust upgrade_boot to do the s/metalink/mirrorlist/
substitution, so upgrade tests don't run into the timing issue
in the steps before the main repo_setup run is done by
upgrade_run, and adjust repo_setup_compose to sub this line out
later.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Sometimes rebooting during upgrade tests seems to take longer
than these timeouts allow, so let's bump them a bit.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
There's a problem with using `--releasever=rawhide` for upgrade
tests ATM - see #1531356 . To avoid this, we'll try using the
real Rawhide release number (which I'm adapting the scheduler
code to discover and pass in as `RAWREL`).
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Summary:
This adds an upgrade variant of the FreeIPA tests, with only
the simplest client enrolment (sssd) for now. The server test
starts from the N-1 release and deploys the domain controller
role. The client test similarly starts from the N-1 release
and, when the server is deployed, enrols as a domain client.
Then the server upgrades itself, while the client waits (as the
server is its name server). Then the client upgrades itself,
while the server does some self-checks. The server then waits
for the client to do its checks before decommissioning itself,
as usual. So, summary: *deployment* of both server and client
occurs on N-1, then both are upgraded, then the actual *checks*
occur on N.
In my testing, this all more or less works, except the role
decommission step fails. This failure seems to be a genuine one
so far as I can tell; I intend to file a bug for it soon.
Test Plan:
Run the new tests, check they work. Run the existing
FreeIPA tests (both the compose and the update variants), check
they both behave the same.
Reviewers: jsedlak, jskladan
Reviewed By: jsedlak
Subscribers: tflink
Differential Revision: https://phab.qa.fedoraproject.org/D1204
It's not really a good idea to have the comments that explain
the test_flags in *every* test, because they can go stale and
then we either have to live with them being old or update them
all. Like, now. So let's just take 'em all out. There's always
a reference in the openQA and os-autoinst docs, and those get
updated faster.
More importantly, add the new `ignore_failure` flag to relevant
tests - all the tests that don't have the 'important' or
'fatal' flag at present. Upstream killed the 'important' flag
(making all tests 'important' by default), I got it replaced
with the 'ignore_failure' flag, we now need to explicitly mark
all modules we want the 'ignore_failure' behaviour for.
Noticed in e.g. https://openqa.fedoraproject.org/tests/58798
we're doing this wrong, `boot_decrypt` was moved into utils as
a function, but we were still calling it as a method...
Summary:
This adds a couple of new exporter modules, renames main_common
to utils (this is a better name: openSUSE's main_common is
functions used in main.pm, utils is what they call their module
full of miscellaneous commonly-used functions), and moves a
bunch of utility functions that were previously needlessly
implemented as instance methods in base classes into the
exporter modules. That means we can get rid of all the annoying
$self-> syntax for calling them.
We get rid of `fedorabase` entirely, as it's no longer useful
for anything. Other base classes keep the 'standard' methods
(like `post_fail_hook`) and methods which actually need to be
methods (like `root_console`, whose behaviour is different in
anacondatest and installedtest).
Test Plan:
Do a full test suite run and check everything lines
up. There should be no functional differences from before at all,
this is just a re-org.
Reviewers: jskladan, garretraziel_but_actually_jsedlak_who_uses_stupid_nicknames
Reviewed By: garretraziel_but_actually_jsedlak_who_uses_stupid_nicknames
Subscribers: tflink
Differential Revision: https://phab.qa.fedoraproject.org/D1080
Summary:
we have a long-standing problem with all the tests that hit
the repositories. The tests are triggered as soon as a compose
completes. At this point in time, the compose is not synced to
the mirrors, where the default 'fedora' repo definition looks;
the sync happens after the compose completes, and there is also
a metadata sync step that must happen after *that* before any
operation that uses the 'fedora' repository definition will
actually use the packages from the new compose. Thus all net
install tests and tests that installed packages have been
effectively testing the previous compose, not the current one.
We have some thoughts about how to fix this 'properly' (such
that the openQA tests wouldn't have to do anything special,
but their 'fedora' repository would somehow reflect the compose
under test), but none of them is in place right now or likely
to happen in the short term, so in the mean time this should
deal with most of the issues. With this change, everything but
the default_install tests for the netinst images should use
the compose-under-test's Everything tree instead of the 'fedora'
repository, and thus should install and test the correct
packages.
This relies on a corresponding change to openqa_fedora_tools
to set the LOCATION openQA setting (which is simply the base
location of the compose under test).
Test Plan:
Do a full test run, check (as far as you can) tests run sensibly
and use appropriate repositories.
Reviewers: jskladan, garretraziel
Reviewed By: garretraziel
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D989
Summary: Pretty simple!
Test Plan:
Check the upgrade tests work the same as before the
change.
Reviewers: jskladan, garretraziel
Reviewed By: garretraziel
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D979
in the recent commit to always use nogpgcheck I inadvertently
broke the upgrade tests, by dropping the `--releasever` from
the `dnf system-upgrade download` command. So fix that.
Summary:
Except when running on the pre-upgrade release in the upgrade
tests (where GPG check should always be OK).
Currently we always need to use --nogpgcheck on Rawhide, and we
must also use it on Branched prior to the Bodhi activation
point. At present we don't really have any simple way to know
when the Bodhi activation point has kicked in. We could assume
that it's safe to do GPG checking for 'candidate' (not nightly)
composes, but even that isn't 100% safe and isn't really the
*right* thing to do. So I think for now it's best to just always
use --nogpgcheck , until we come up with a decent way to check
for Bodhi enablement, or releng figures things out so we can
rely on packages being signed in Rawhide and in Branched before
Bodhi enablement.
Test Plan:
Check the tests all still run, make sure I didn't
miss any dnf calls.
Reviewers: jskladan, garretraziel
Reviewed By: garretraziel
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D964
Summary:
This requires us to handle decryption each time we reboot in
the upgrade process, so factor that little block out into the
base class so we don't have to keep pasting it. It's also a
bit tricky to integrate into the 'catch a boot loop' code we
have to deal with #1349721, but I think this should work. There
is a matching openqa_fedora_tools diff to generate the disk
image.
Test Plan:
Run the tests, check that they work, run the other
upgrade and encrypted install tests and check they still work
properly too.
Reviewers: garretraziel
Reviewed By: garretraziel
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D922
Summary:
try to catch a boot loop after `dnf system-upgrade reboot`, if
we do, set the test to soft_fail and pass enforcing=0 to work
around it.
Test Plan:
Run the upgrade_foo tests and see that they now soft
fail instead of hard failing (unless there are any other issues).
Run the upgrade_2_foo tests and make sure they still pass (i.e.
we don't erroneously soft fail them).
Reviewers: garretraziel
Reviewed By: garretraziel
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D912
Summary:
I really just want to add the desktop_terminal test, but I think
this refactor is in order now. It splits up loading of the
various test phases (much as SUSE do it) and allows us to run
the post-install tests without the install tests, for e.g. I
tweaked things to allow the upgrade tests to use the existing
_wait_login tests for final login and combine the two upgrade
postinstall tests into one simple one.
This comes with a bit of a behaviour change to make graphical
wait login behave the same as console wait login: it will log
in unless USER_LOGIN is set to 'false'. Previously it only
logged in if both USER_LOGIN and USER_PASSWORD were set, which
I don't think ever happened in a graphical test, so we never
actually did a graphical login. The intent here is we should do
a login on the default_install tests. That's going a bit beyond
the test case, but it seems like a reasonable thing to test. We
can set USER_LOGIN to false if we don't want to do it.
Test Plan:
Do a full test run, make sure the new tests work and
no old tests break.
Reviewers: jskladan, garretraziel
Reviewed By: garretraziel
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D839
With the arrival of Pungi 4, the scheduler is no longer using
fedfind-provided BUILD and FLAVOR values, but ones derived from
Pungi properties. BUILD is now simply the Pungi compose_id.
FLAVOR is produced by joining the Pungi variant, type, and
format with '-' characters as the separators.
Pungi, unfortunately, does not treat 'Rawhide' as a release, it
synthesizes a release number for Rawhide composes and places
that in the compose ID. To cope with that, for now, the
scheduler will set RAWHIDE to '1' if the compose is a Rawhide
one. As we have to adapt all places where we parse the release
in any case, this commit consolidates them into a fedorabase
subroutine.
For the one place where we also used to parse the 'milestone'
from fedfind, there is a placeholder get_milestone subroutine
which currently returns an empty string, as I don't yet have a
good handle on how to draw the kinds of distinctions fedfind
mapped to 'milestone' from Pungi metadata.
Summary:
Instead of sitting there waiting 6000 seconds twice, when DNF
explicitly tells us it failed, just die.
This is why we haven't been getting proper compose check reports
lately; the upgrade tests are failing, waiting 6000 seconds to
time out, then being cloned and tried again, waiting another 6000
seconds. This is just barely going beyond check-compose's 8 hour
wait limit, as it's some time before the upgrade tests even get
started (they're low in the priority list). We're still going to
have that problem if the tests fail any other way, but this at
least catches that case.
Test Plan:
Run the upgrade tests and see that they fail quicker
(assuming the dependency problems they're dying on aren't fixed).
Maybe also do a 22-23 upgrade test and check it still succeeds
properly.
Reviewers: jskladan, garretraziel
Reviewed By: garretraziel
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D650
Summary:
Updating the stable release prior to doing the update can take
a long time if the image hasn't been updated for a while, and
the upgrade download process itself can take a long time. If
the screen blanks out in either case, either the following
needle match may fail (if we're waiting for a needle) or 'still
screen' may be detected early (if we're waiting for a still
screen), so let's disable screen blanking to avoid it.
Test Plan: Run the upgrade tests and see if they work.
Reviewers: garretraziel, jskladan
Reviewed By: jskladan
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D628
Summary:
This is a first cut which more or less works for now. Issues:
1) We're not really testing the BUILD, here. All the test does
is try and upgrade to the specified VERSION - so it'll be using
the latest 'stable' for the given VERSION at the time the test
runs. This isn't really that terrible, but especially for TC/RC
validation, we might want to make things a bit more elaborate
and set up the repo for the actual BUILD (and disable the main
repos).
2) We'd actually need --nogpgcheck for non-Rawhide, at one
specific point in the release cycle - after Branching but
before Bodhi activation (which is when we can be sure all
packages are signed). This won't matter until 24 branches, and
maybe releng will have it fixed by then...if not, I'll tweak
it.
3) We don't really test that the upgrade actually *happened*
for desktop, at the moment - the only thing in the old test
that really checked that was where we checked for the fedup
boot menu entry, but that has no analog in dnf. What we should
probably do is check that GUI login works, then switch to a
console and check /etc/fedora-release just as the minimal test
does.
Test Plan:
Run the tests. Note that creating the desktop disk
image doesn't work ATM, so I can't verify the desktop test
works, but the minimal one seems to (with D565). There'll be
a matching diff for openqa_fedora_tools to update the test
case names there.
Reviewers: jskladan, garretraziel
Reviewed By: jskladan, garretraziel
Subscribers: tflink
Differential Revision: https://phab.qadevel.cloud.fedoraproject.org/D567
