aarch64 cloud is blocking, but it's hitting:
https://bugzilla.redhat.com/show_bug.cgi?id=1722181
and failing the composes. Due to other issues we haven't had composes
for a while before that either, so it's getting to be a long time.
Hopefully some workaround will at least appear in a few days.
If not, we can revert this and start failing again for a while.
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
The koji runroot used for to create the OSTree installer has no access
to external networks, so use a special hostname for registry.fedoraproject.org
that suppresses the redirects to the CDN that would normally happen for
blob downloads.
Signed-off-by: Owen W. Taylor <otaylor@fishsoup.net>
Use a newly added Pungi option (https://pagure.io/pungi/pull-request/1278),
so that flatpak is in the Silverblue ostree-installer runroot, which is
needed for embedding Flatpaks in the Silverblue image.
Signed-off-by: Owen W. Taylor <otaylor@fishsoup.net>
The last change set the template variables for embedding Flatpaks,
but didn't actually include the necessary template.
Signed-off-by: Owen W. Taylor <otaylor@fishsoup.net>
A long-term goal of Silverblue has been moving applications that
were previously part of the Workstation image into Flatpaks.
At this point, we many of the applications that were removed from
Silverblue available as Fedora Flatpaks, and we have the necessary
Anaconda support, so we should update the Silverblue images to include
the runtime and applications.
This is currently x86_64 only, since we don't have a working OSBS
cluster to build containers for aarch64 or ppc64le.
Note that Fedora Flatpaks are single-stream, so this embeds the
F30 runtime and applications into the Rawhide installer image. When
the applications are released for F31, the embedded runtime will
need to be updated.
Requires: https://pagure.io/fedora-lorax-templates/pull-request/40
Signed-off-by: Owen W. Taylor <otaylor@fishsoup.net>
Partly revert commit e25f2baaa1
instead of pulling in dracut* (dracut and all it's subpackages), just
pull in the main one. This should hopefully fix
https://bugzilla.redhat.com/show_bug.cgi?id=1734179
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
Similar to what we did for Cloud (6eecb53) we need to stop building
Workstation repos so we can speed up our compose times (see [1]).
We already switched to sourcing from the Everything repo (39683dd).
Now we just need to stop building the repo. This commit makes it so
that we will no longer build the yum repo for Workstation.
[1] https://pagure.io/releng/issue/7403
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
Should fix:
WARNING: Config option pdc_url was removed and has no effect; Koji is
queried instead.
WARNING: Config option pdc_develop was removed and has no effect; Koji
is queried instead.
WARNING: Config option pdc_insecure was removed and has no effect; Koji
is queried instead.
in rawhide composes.
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
Certain applications are using the container env variable to
know if they are running in a container or not. This is making sure
we have this define in the fedora base image..
Fixes https://bugzilla.redhat.com/show_bug.cgi?id=1399507
Signed-off-by: Clement Verna <cverna@tutanota.com>
Make sure glib-networking is always installed multilib. Right now the
dependency that makes glib-networking multilib is Requires:
glib-networking%{_isa} in libsoup, but if libsoup is not in the same
repo as glib-networking (as is the case right now with the F30 updates
repo), glib-networking can drop out of multilib.
Fix this by explicitly listing it as multilib.
https://pagure.io/releng/issue/8365
Signed-off-by: Kalev Lember <klember@redhat.com>
This is to keep in sync with ostree config fedora.conf
which we ship as part of ostree update through fedora-repos-ostree
Signed-off-by: Sinny Kumari <sinny@redhat.com>
Rawhide and Branched createImage tasks are frequently failing
with "error: Curl error (16): Error in the HTTP2 framing layer"
when trying to download the zchunk metadata. We have not yet
been able to work out why this is, but it is a serious problem
causing many composes to fail and badly impeding our testing
of Rawhide and F30. For this reason we've decided to disable
zchunk metadata generation until this is fixed.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
As part of the UEFI on ARMv7 feature we build the images the same
as aarch64 using image factory. This is initiall enabling the
Minimal/Server/Workstation images to allow wider testing. Once we
know they're good we'll move all the others over and remove the old
build mechanism that uses appliance-creator.
Signed-off-by: Peter Robinson <pbrobinson@fedoraproject.org>
This is needed for uploading the image to aws using the
github.com/coreos/mantle code base. We can upload the raw
files but it is a significant waste of bandwidth to do so.
Right now aws import-image does not support qcow2 files,
otherwise we would be using that.
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
This adds the createrepo_c arguments to generate zchunked metadata for
Rawhide. For this to work, the server running createrepo_c will need
createrepo_c-0.12 and fedora-repo-dicts installed.
Signed-off-by: Jonathan Dieter <jdieter@fedoraproject.org>
Some x86_64 users wish to install ibus-gtk3.i686 and ibus-gtk2.i686
in the multilib systems but ibus.i686 is not required now.
The users could install them with `dnf install ibus-gtk3.i686` when
those modules are added to multilib_whitelist.
https://pagure.io/releng/issue/7937
Signed-off-by: Takao Fujiwara <tfujiwar@redhat.com>
There are several recommends packages that aren't getting installed in
Silverblue because the workstation comps group doesn't include them.
One of these is slirp4netns which is needed for rootless containers
using podman.
Since we are thinking about removing the separate Workstation repo
anyway[1], then maybe we can just move SB over to the Everything repo.
[1] https://pagure.io/releng/issue/7403
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
dnf does not use these, but mdapi does, so it's worth it to us to
just keep shipping them for now at least until mdapi has some other
way to get repodata.
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
This will ensure that the OSTrees get generated on each compose
so that the media created will match the version numbers of the
OSTree inside.
The caveat is that there can be empty commits that do happen.
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
We need to stop building repos and install media for the *Cloud* variant
because the only consumers of those repos/install media is the cloud
base images that could easily be built with the Everything repo/install
media. This commit removes the cloud variant (as it was previously
known) and then renames the CloudImages variant to 'Cloud'.
Summary:
- Cloud now becomes an empty variant (no repos, no install media)
- CloudImages variant goes away
- Cloud image builds use Everything repo/install media
- Atomic image builds use Everything install tree
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
In the output we will now put the AtomicHost cloud images under
the AtomicHost/ directory instead of in CloudImages.
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
Move Atomic to AtomicHost and WorkstationOstree to
AtomicWorkstation. This is mostly to make things align
more with the names of the things we actually put out.
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
Pulling these packages into Modular content is probably not the right
thing to do. The config should only list variants that get traditional
content.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Since we now include modular content, we need to tell Pungi where to get
information about them from.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Some newer subvariants / loadouts were missing from the
substitution list. This updates it, alphabetizes it, corrects
the indentation, and adds a better comment about what it's for.
Note https://pagure.io/koji/pull-request/817 , corresponding
pull request for Koji.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
With Fedora 27, s390x composes (Server and Everything) are
part of primary compose. It will be nice to have atomic, cloud
and docker composes available for s390x too.
Signed-off-by: Sinny Kumari <sinny@redhat.com>
By FESCo decision all Java packages are exempt from multilib.
See: https://pagure.io/fesco/issue/961
Signed-off-by: Mikolaj Izdebski <mizdebsk@redhat.com>
This should cause the compose to contain all the previous content plus a
separate Modular variant with packages specified by list of modules in
variants file.
The final layout of the compose should be like this:
compose/Everything/$arch/os/
compose/Modular/$arch/os/
Relates: https://pagure.io/releng/issue/7227
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
- change comps_file setting in the fedora.conf to use the new scm_dict
type from the latest pungi release
- remove the make step from the nightly shell script
Signed-off-by: Robert Marshall <rmarshall@redhat.com>
This will enable the optimizations landed in
https://github.com/rhinstaller/anaconda/pull/1119
to avoid lots of HTTP requests when installing the cloud images.
Signed-off-by: Colin Walters <walters@verbum.org>
Creating the image requires running lorax. There is however already one
lorax instance being run in the buildinstall phase. When the second one
runs, it overwrites the outputs from first one and people expecting to
boot netinst over PXE are instead getting ostree workstation.
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1477916
Relates: https://pagure.io/pungi/issue/695
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
See https://pagure.io/pungi/pull-request/727
Once that is merged, released, and deployed, this change will let us see
New and Fascinating Details about the compose process.
Signed-off-by: Ralph Bean <rbean@redhat.com>
Since not all builds are signed with f28 key yet, we will use f27 key
also and once everything is signed with f28 then we will remove f27 key
Signed-off-by: Mohan Boddu <mboddu@redhat.com>
It's failing on ppc64le, we're at 87% on x86_64. There's really no reason not to
bump this to give ourselves some room at least on 64 bit architectures.
Signed-off-by: Colin Walters <walters@verbum.org>
To match up with the Addition of Container Minimal and the renaming
of dist-git namespace to Container, rename the Docker Variant to
Container
Signed-off-by: Dennis Gilmore <dennis@ausil.us>
Currently the l10n support is moved to a single calligra-l10n package
(which is a subpackage of calligra itself) as opposed to previously
separate calligra-l10n package with a separate subpackage for each
language (which is now retired).
Due `greedy_method = "build"` the l10n subpackage will be pulled into
the compose if some package from the same build gets in (i.e. the main
calligra package required by comps).
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Since we pull from ostree repos it doesn't really make
sense to provide a yum "repo" to the image build in pungi.
Removing it since it could be confusing to have it there.
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
Add new ostree_install_repo and ostree_update_repo vars so that we
can tell lorax where to pull from during the pungi compose and also
where to configure an installed system to pull from. i.e. where a
user is going to pull from over time.
This is related to https://pagure.io/fedora-lorax-templates/pull-request/19
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
These options used to be required in pungi < 4.1.14, but now they have
default values that happen to be the same as what Fedora uses. No need
to specify them anymore.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Koji added the ability to set livemedia and image builds to
fail on some arches, the compose configs should be updated to use
that feature and able to identify the blocking artifacts and on what
arches.
Signed-off-by: Mohan Boddu <mboddu@redhat.com>
We have decided [1] that we want to have more meaningful versions
for ostree commits and the resulting images. One way to do this
is to have the ostree version match the version from the compose ID.
Pungi recently [2] gained support for embedding the label from the
compose into the ostree commit version. This change takes advantage
of that work to embed a more meaninful version.
[1] https://pagure.io/atomic-wg/issue/229
[2] https://pagure.io/pungi/pull-request/592
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
For the cloud base image there is no need to override the default
vagrant sync location. /vagrant/ works just fine and is what people
expect.
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
In koji there are ova-options and also factory-parameters.
We can pass ova-options to koji and also factory-parameters
but we can't pass ova-options as a factory-parameter. We just
need to pass in the vagrant_sync_directory as a factory
parameter.
Fixes https://pagure.io/atomic-wg/issue/181
Signed-off-by: Dusty Mabe <dusty@dustymabe.com>
Fix: #168
We didn't explicitly set the branches as master for rawhide which
created issues during branching by missing setting up the branch names.
This makes it easier to identify those branch names.
Signed-off-by: Mohan Boddu <mboddu@redhat.com>
Fixes: #6765
The default size of filesystem provided by lorax is 2GB and the
installtree size is gone up to 2.7GB which is failing to copy files
from installtree to created root filesystem.
Signed-off-by: Mohan Boddu <mboddu@redhat.com>
Fixes: #6765
The default size of filesystem provided by lorax is 2GB and the
installtree size is gone up to 2.7GB which is failing to copy files
from installtree to created root filesystem.
NOTE: Just increased for x86_64 and ppc64le since aarch64 is fine.
Signed-off-by: Mohan Boddu <mboddu@redhat.com>
Fedora Atomic WG has decided to move from the "docker" namespace to
the "container" namespace for the sake of making it more generic in
order to target all OCI Compliant container runtimes instead of
a specific implementation. This update changes what is necessary for
to the base image metadata as well as add a transitional second call
to handle both the docker and container namespace for block_retired.
Signed-off-by: Adam Miller <maxamillion@fedoraproject.org>
As there is a hard limit for volume id's on isos to be <32 characters,
we have a list of subsitutions that we maintain to make sure that
isos all have volume id's under that size. with the addition of
ppc64le and aarch64 for Atomic host installer the 7 character arch
length pushed us over the limit using AH for Atomic seems the most
logical change to make.
Signed-off-by: Dennis Gilmore <dennis@ausil.us>
For depsolving with YUM this has no effect, but will add the package to
DNF based compose.
YUM is matching the packages against NVR, so the `-*` at the end of the
pattern matches version. DNF-based Pungi is only looking at package
names.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The previous formatting for docker image metadata was getting
escaped incorrectly when passed to imagefactory. This resolves that.
Fix the license label to be MIT instead of GPLv2.
Fix the bash path
Remove atomic RUN command, it causes build failures and isn't
necessary in the base image.
Signed-off-by: Adam Miller <maxamillion@fedoraproject.org>
Currently, imagefactory's docker plugin will default an older
version of the metadata imager format that is not compatible with
the latest versions of docker. This change tells the plugin to use
the docker version "1.10.1" metadata format which is compatible with
all currently available docker releases.
Signed-off-by: Adam Miller <maxamillion@fedoraproject.org>
Currently we don't produce a docker base image that can be used with
OSBS or via the 'atomic run' command without modification. This
fixes that and will allow for an automated import of the base images
into the Fedora Container Registry
Also, fix various missing commas in dict declarations
Signed-off-by: Adam Miller <maxamillion@fedoraproject.org>
This is a rework of <https://pagure.io/pungi-fedora/pull-request/108>,
and re-applied from <https://pagure.io/pungi-fedora/pull-request/167>
since I found out that in fact `fedora-atomic.conf` isn't used in
this branch.
Note that I left a few things like git branches pointed at f26, but
I think these should be master most likely.
Signed-off-by: Colin Walters <walters@verbum.org>
The Cloud tree does not contain python-classroom comps group.
And the Lab is not part of the Cloud Variant. So, we need to
use Everything install tree for Labs.
Signed-off-by: Mohan Boddu <mboddu@redhat.com>
minimal container image says it needs more disk space so bump
size to 5G
use the Everything repo for the container images, microdnf
is not part of the cloud variant and it really does not make
ense to add it nor to make a container Variant with its own
install tree
Signed-off-by: Dennis Gilmore <dennis@ausil.us>
There was a missing },{ between the two container images. The end result
is that pungi validates the config but only finds the last section so
only makes one of the two images
Signed-off-by: Dennis Gilmore <dennis@ausil.us>
This enables the creation of the aarch64 cloud and docker images as part of the nightly compose
Signed-off-by: Peter Robinson <pbrobinson@fedoraproject.org>
