Convert any image that is not a raw image, e.g. a qcow2, to a raw
image and open the partitions via loop devices. This replaces the
usage of nbd, which was racy and flaky.
Instead on relying on the kernel for the partition parsing, this
is now done manually via loop devices and start + offset taken
from sfdisk. As a result the read_partition function has been
adapted to be called at later time, after the partitions were
opened via loop devices.
Not using nbd also means that the partition table is not scanned
by the kernel anymore and udev is not triggered. As a result the
'PARTUUID' property is not present for dos/mbr partition layouts,
since it is auto-generated by udev/blkid. Relevant blkid files
and functions are:
blkid_partition_gen_uuid(par)
called from probe_dos_pt()
in file libblkid/src/partitions/dos.c line 295
defined in libblkid/src/partitions/partitions.c line 1374
which generates the uuid via snprintf using the format:
'"%.33s-%02x", par->tab->id, par->partno'
Based on https://github.com/karelzak/util-linux at ce8985cc7
NB: the loop device code is imported from osbuild, making this
tool depend on osbuild's private library.
NB: As of the image conversion, more disk space is required to
examine non-raw images.
If a compressed file is detected (xz, bzip2, gzip), uncompress it
before analyzing it. NB: this has to happen after the is_tarball
check, because a tarball would also count as compressed.
Directly iterate over the partition list entries instead if using
indices. Optimize visual flow for the happy case by immediately
continuing if a partition has no file-system.
Read the image information before opening the image. This prepares
for a change where we will use information about the image to
convert it to raw instead of using nbd.
Create a GetImageInfoCommand analogous to GetOsbuildCommand that
will adjust the PYTHONPATH for image-info on travis so that the
osbuild python module is accessible.
The time it takes to depsolve a blueprint varies widely depending on
where the job is running and which mirrors are randomly chosen based on
the data returned in the metalink XML.
Use dnf's fastestmirror plugin to choose the fastest mirror for
downloading metadata. This returns consistent results in PSI + AWS and
every depsolve completed in under 60 seconds after 25 tests in each
cloud.
Fixes#845.
Signed-off-by: Major Hayden <major@redhat.com>
List the image tests that match the distro, distro version, and
architecture before running image tests. That ensures we run all of the
tests that are appropriate for the distro/version/arch combination.
Tests can be added or removed without changing the `run_image_tests.sh`
script.
Fixes#840.
Signed-off-by: Major Hayden <major@redhat.com>
As we embark on a CI journey to other architectures, we need to be
explicit about choosing the right architecture for each CI node.
This is a portion of the work required for #834.
Signed-off-by: Major Hayden <major@redhat.com>
Now that RHEL 8.3 pulls from the beta CDN, let's stop hiding the errors
and allow them to make the pipeline fail.
Signed-off-by: Major Hayden <major@redhat.com>
Some dnf bugs existed in early releases of Fedora 31 + 32 around
repository priorities being ignored (like BZ 1733582. Ensure that we are
running the latest version of dnf before we try to install from the mock
build repo (which has a much higher priority than Fedora's default
repos).
Also, check to see if EPEL is installed already before installing it
again.
Fixes#824. 🥰
Signed-off-by: Major Hayden <major@redhat.com>
Remove the F31/F32 internal repositories since they are causing dnf
download errors from time to time.
The 8.2 repository override should have never been there in the first
place since we're using the CDN.
The 8.3 repository is no longer needed since we can register the RHEL
8.3 instances against the staging CDN and get beta content from there.
Fixes#822.
Signed-off-by: Major Hayden <major@redhat.com>
Previously, all resouces were created with a certain tag. When the cleanup
phase came, the Resources - List route[1] was used to get all resources with
the tag. Then, they were deleted in the right order.
Sadly, the Resources - List API has issues with listing disks. Sometimes,
it returns the virtual disks 15 minutes after they were created. As the
result, the disks have been left behind quite often and our bill was higher
than necessary.
This commit uses a different method - the Go code now knows all resource names
(see the previous commit), so it can delete all resources without listing them
using the "broken" API route.
[1]: https://docs.microsoft.com/en-us/rest/api/resources/resources/list
Prior this commit the resource names were generated in the deployment
template, so the Go code actually didn't know them. This commit generates
all names in the Go code, so they can be used in the future commits.
Currently, RHEL 8.3 ships lorax-composer 28.14.49, which we want to
replace. 28 <= 28.14.49, so this did not work, bump it to 29, so we do
not get tripped up by bugfix releases of lorax-composer.
In the event that this obsoletes should one day be reverted, we do not
bump it to "infinity", so a sufficiently new version of lorax, would still
be able to obolte osbuild-composer again.
Signed-off-by: Tom Gundersen <teg@jklm.no>
Sometimes dnf has issues downloading RPMs or downloading the repo XML
and this breaks the mock build. Try to run the mock builds three times
before giving up.
Signed-off-by: Major Hayden <major@redhat.com>
Now that we have minimal images built by osbuild-compsoer for CI, let's
use those for RHEL mock builds.
This PR depends on osbuild/osbuild-composer#808 to merge first.
Signed-off-by: Major Hayden <major@redhat.com>
RHEL can benefit from `fastestmirror` when it downloads metadata and
packages from the EPEL repository.
Fixes#817.
Signed-off-by: Major Hayden <major@redhat.com>
Although the vhdx format is smaller than raw, AWS often has some issues
booting instances from an imported vhdx snapshot. The raw format seems
to boot consistently at AWS.
Fixes#754.
Signed-off-by: Major Hayden <major@redhat.com>
We need to wait for cloud-init to be completely finished since it is
registering RHEL instances with RHN and preparing a swap file.
Signed-off-by: Major Hayden <major@redhat.com>
osbuild-composer is a drop-in replacement of lorax-composer, and
side-by-side installation is not supported. It is still possible to
explicitly install lorax-composer and exclude osbuild-composer, but the
default is to replace lorax with osbuild on upgrades.
This change is for now only for RHEL8.3, but we should consider doing
the same for F33.
Fixes https://bugzilla.redhat.com/show_bug.cgi?id=1836844
Signed-off-by: Tom Gundersen <teg@jklm.no>
The next release we are targeting is RHEL 8.3 Beta, switch the
repositories over.
In the future, we should distinguish between beta and non-beta based on
/etc/os-release so we don't have to hard-code this upstream.
The beta repositories are not actually available yet, so in our tests,
we have to override this again to point at the GA ones (currently 8.2).
Needless to say, this situation leaves much to be desired.
Signed-off-by: Tom Gundersen <teg@jklm.no>
Add one test-case per distro that tries to cover all the customizations
we support.
For now omit firewall customizations, as there are open questions about
how to support that on all image types.
Signed-off-by: Tom Gundersen <teg@jklm.no>
Rather than getting a set of base packages from the ImageType, and then
appending the requested packages from the blueprint, pass the blueprint
into the new Packages() function, and return the full set of packages to
be depsolved.
This allows us to also append packages based on other customizations
too, and use that to append chrony when the timezone is set. This
matches the behavior anaconda had, and there was a TODO item to do this,
which had been overlooked.
Fixes#787.
Signed-off-by: Tom Gundersen <teg@jklm.no>
The previously used snapshot was deleted. This commit changes the override to
use Internal Snapshot 2.0, which should not be deleted in the near future.
composer uses the success field to decide whether a build succeeded or failed.
This is bad.
Unfortunately, fixing this requires kinda big code changes. This commit
changes the worker's behaviour to set the osbuild success flag to false
even on errors which weren't caused by osbuild (e.g. an upload error).
This is certainly hacky but I think it's still essential to tell the user
that an error occurred.
Fixes#789
The issue was introduced in 0d3c8329c0.
The patch correctly changed the base exception class, but it didn't
change the unfortunate use of hardcoded type name. This patch uses
Python's internal `__name__` attribute to get the type (exception) name.
