When pungi-gather (or old pungi) download the packages from repos into
work/$arch/, they are linked to work/global/. This was using link_type
configuration option.
However if that is set to symlink, we get a relative symlink in
work/global/ which is later copied under the compose/ directory. Since
it's a relative symlink, it gets broken by this.
The fix is to hardlink the downloaded packages instead in the first
step. Since both the source and destination are in work/ directory, we
know it's the same volume and hardlinks should work. There is a fallback
to copy just to be sure, but it shouldn't ever be used.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
All places in the code assume the option to not be required except for
this one line.
An obsolete comment is removed as well.
Relates: https://pagure.io/releng/failed-composes/issue/477
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Instead of just checking that the repo directory exists, make sure the
repodata subdirectory is in there. If it's missing, then createrepo_c
has nothing to use anyway, and it may help avoid issues.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This should make it possible to only import the library only when it's
really needed.
DNF does not work with libmodulemd v2. If we import libmodulemd2 and
then dnf, the program will just hang forever. We only need DNF in
pungi-gather, where libmodulemd is not needed, and also where we do need
libmodulemd, we don't have DNF.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Add a configuration option to enable skipping some modules found in the
configured tag.
Fixes: https://pagure.io/pungi/issue/1260
JIRA: COMPOSE-3794
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This patch adds a new config option. This is expected to be a name of
subdirectory in the repo with module defaults. If supplied, overrides
from that location are loaded every time defaults are loaded.
This raises the minimal required version of libmodulemd to 2.8.0
JIRA: COMPOSE-3828
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This is not called anywhere anymore. Instead of merging these package
sets, they exist as separate entities.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The cache is now always saved when the repo is created on disk. The
loading procedure in Koji source is updated to look for cache for
correct tag.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The pickled pkgset was only ever used for debug mode. We don't use that
anymore. Let's stop writing files that are not used anywhere.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
With this patch, there should be a separate package set for each tag
that is consumed.
Generally each module will create a separate package set, with the
exception of -devel modules that will be in the same set as their
non-devel version.
Variants no longer need to keep their own package set objects. Instead
they now include a set of package set names that should be used for the
variant. This can replace the whitelist mechanism of deps gather method.
JIRA: COMPOSE-3620
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Once a package set repo is written to disk, let's use this object that
connects the repository path with the mapping of packages.
This change also makes it explicit where the dependency on package set
repos are. In the original codebase, any part of code could generate a
path to the repo, even if that repo has not yet been written.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This name will serve as an identifier for the group of packages.
For Koji package sets, it should the name of the tag from which the
packages come. For package sets based on repos a dummy constant name is
used.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This opens up a path to having multiple package sets in the compose. The
pkgset phase now creates a list of them (although at this time there is
always a single item in that list).
Any consumer of the package sets objects is updated to handle a list.
Generally this means an extra loop.
JIRA: COMPOSE-3620
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Both pkgset sources use the same logic to create per-arch repos. There
is no reason to have that code in both places.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The data parsed from variants.xml uses a different format that what we
added in `_add_module_to_variant`. This leads to crashes later.
JIRA: COMPOSE-3746
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This was already discouraged to not be used, and is a bad idea in
current setup anyway. Removing this can simplify the code.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
We can avoid parsing source modulemd information since we can get the
same information from the Koji build info.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Historically each variant had a list of modules. This is no longer
needed and can be dropped. We can also stop logging the modulemd since
we know it was retrieved from Koji and not modified locally.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
If the package set repo contains any modular package, the module
metadata is added there as well.
This is needed to accomodate change in DNF that refuses to work with
repo with modular packages if the metadata is not there. This DNF change
can cause issues in buildinstall phase.
Related: https://bugzilla.redhat.com/show_bug.cgi?id=1623128
The hybrid solver is modified to not create a separate repo with the
module metadata anymore, since it will be available in the repo with
packages. This also allows us to drop code to look into lookaside repos.
We still need to iterate over local modules in order to find out what
platform should be used.
JIRA: COMPOSE-3621
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This patch reuses the existing createrepo_num_threads options to limit
maximum number of parallel createrepo processes.
Fixes: https://pagure.io/pungi/issue/955
JIRA: COMPOSE-2575
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Higher protocols should be more efficient in terms of performance and
storage size. Since we don't really care about interoperability with
different python version, we can safely go to the highest version.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Since there can be multiple tags, the check must be done once for all of
them at the same time. Otherwise any module found only in some and not
all tags would raise this error.
The code builds a set of all existing patterns and then removes items
from it. If there is something left once all tags are processed, it
means such patterns were not matched by anything.
JIRA: COMPOSE-3609
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
If the package set is loaded from a pickle, it will not contain this
key, and thus it can not be again written down into a new pickle. While
this is not a common use case, it's still better not to crash in that
case.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
We need to update the source modulemd to match what we want. For most
modules the name will be correct already, but in case of devel modules
the original name will still be present there.
Fixes: COMPOSE-3596
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This is fairly similar to a package only being built for particular
arches.
Fixes: https://pagure.io/pungi/issue/1198
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The prefix checking only works if there are no streams that would share
prefixes. Let's instead check the value as a whole. There is extra
complexity from the fact that version and context may not be specified.
The stream as specified in input is processed to replace dashes (`-`)
with underscores (`_`) to match how the builds are imported into Koji.
JIRA: COMPOSE-3547
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
For modular tags we only include packages that are included in the
module. Originally the filter was taking package names from
rpms.artifacts section of MMD.
This however does not work correctly, as there can easily be module
which lists foo.src but does not want to include foo.x86_64 or other
arches.
This patch fixes this particular problem by included arch in the set of
packages to be kept.
JIRA: COMPOSE-3543
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The scheme for generating versions has changed multiple times. MBS is
careful to only modify them so that they always compare correctly. This
only works though if the versions are treated as numbers.
This should be safe in that non-numbers should never be encountered as
module version. Libmodulemd internally stores the version as int (or
some version of int).
JIRA: COMPOSE-3540
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
A module build can create packages that are tagged in the content tag,
but should not be included in the module. Originally Pungi didn't know
what exactly the module contains and so it needed to apply filters to
exclude stuff that was definitely out.
With getting the final MMD from Koji, we can actually make this a bit
more strict by only keeping packages that we know we need.
When processing each content tag, we can put into package set only
packages that are included in some module using that tag. This should
work with -devel modules as well. Both the regular and -devel modules
will contribute to the set and thus all packages will go to the package
set.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Instead of loading the "source" modulemd, always get the final file for
each architecture from Koji.
Logging the downloaded files locally is no longer necessary, when
debugging a problem we can find the files in the respective builds in
Koji.
Tests are updated to work with new code, and an obsolete test is
removed.
JIRA: COMPOSE-3147
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The file in old compose does not change, there is no need to load it
again for each tag.
One consequence of this change is that the same cache will be used by
all package sets. They add stuff to the cache as they use it. However
that should not be a problem for the same reason it's okay to use the
cache in the first place. If two packages have the same path, they are
actually the same file and it's okay to reuse the data.
JIRA: COMPOSE-3374
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Also remove the useless listTaggedRPMs call which has been used *only*
to catch the very rare error which happened from time to time when
we used PDC in Pungi few years ago. This rare error is not relevant
anymore now.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
It may make sense to break a big compose into smaller chunks that can be
done independently. This script allows describing the smaller parts,
runs them with correct dependencies and arranges the result to look like
a single big compose.
All parts use the same koji event, that is either obtained from Koji, or
from command line argument.
JIRA: COMPOSE-2654
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Stream can contain dash sign and when MBS imports such NSVC to Koji, the
dash is replaced with underscore. The current Pungi code does not
respect that and tries to use the stream from Koji directly, which
results in wrong stream being using in some Pungi internal data.
In this PR, the NSVC is taken from module metadata section of CG Koji
build, which contains real stream including the dashes.
Merges: https://pagure.io/pungi/pull-request/1072
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
By default nothing should change. This patch adds a new config option
that enables all this new craziness:
* it turns of applying module filters at pkgset level
* it creates new modules and adds them to the compose
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The pool is created in multiple places, and the process is always to
create an instance and add workers to it. Let's abstract the loop in a
method.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This PR adds new pkgset_koji_builds configuration option.
This option allows setting list of extra Koji build NVRs which will be
included in a compose. This is useful in two cases:
a) It allows generating standard composes with few packages update to
certain version to test how the compose behaves when the package is
updated for real.
b) It allows generating compose consisting only from particular builds
when pkgset_koji_tag = '' or None. This is useful when one want to
regenerate the compose with packages which are not tagged in single Koji
tag. This is very useful for ODCS when reproducing old composes.
Merges: https://pagure.io/pungi/pull-request/1049
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
This patch changes the behaviour when both module tag and NSV?C? is
specified. The NSVC are used as a whitelist and only matching modules
will be included in the compose.
Additionally this patch adds filtering based on inheritance: when
finding the latest module for each N:S combination, only the top tag in
which the module is tagged is used. Even if a newer build is available
somewhere deeper in the inheritance, it's not going to be used.
Example inheritance and tagged modules
f29-compose (foo:1:2018:cafe)
└─ f29-candidate (foo:1:2019:cafe)
The compose will use 2018 version, because it's in the topmost tag.
JIRA: COMPOSE-2685
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
In the search result, we should ignore any module build that finished
after the event that we are working with.
Fixes: https://pagure.io/pungi/issue/999
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
If the same tag is queried with the same event, Pungi can cache the
response and call the API again. Particularly for small composes this
can save up significant amount of time.
Merges: https://pagure.io/pungi/pull-request/1022
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
If a module says to filter a package out, we can do it immediately when
getting the build information from Koji.
This avoids a possible problem of something pulling the module package
in as a dependency, but it should also make the package set slightly
smaller.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
PDC is deprecated in upstream. The usecase for getting list of modules
by NS, NSV or NSVC can however be satisfied by querying modules imported
into Koji.
This makes it possible to deprecate PDC configuration.
Merges: https://pagure.io/pungi/pull-request/985
Signed-off-by: Martin Curlej <mcurlej@redhat.com>
The `add_noarch` option of `get_valid_arches` is broken and doesn't
really do anything (noarch is always present in the result).
This causes packages that have ExclusiveArch including noarch to
actually not be excluded. They should be.
Changing this globally could have a very big impact. Therefore we can
hide it behind a configuration option so that it's opt-in.
JIRA: COMPOSE-2457
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
With one compose combining traditional and modular content there might
be different requirements for tag inheritance. This patch adds a new
option that controls whether builds in modular tags should be inherited.
It defaults to False, which is the right option for current MBS
behavior.
JIRA: COMPOSE-2148
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The code checked with `assert` that there is only one module matching
given NSV. In actual package that would not do anything and we would
silently pick the first value.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This speeds up the compose quite a bit and there is no need for the
database anyway.
Merges: https://pagure.io/pungi/pull-request/922
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
The comps could potentially be different in different variants, so
instead we can create the comps repo for every variant separately and
use two repos instead of one (packages in one repository, comps in
another one).
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
We can't rely on the UID to be correctly joined with colons. There may
be historical data that still uses dashes.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This should indicate that it's a modular variant, but there is no
modular content yet. We don't want to treat that as Everything.
The end result will be an empty repository.
Fixes: https://pagure.io/pungi/issue/871
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
For the first pass we don't need to filter out exclusive architectures,
and we don't need to exclude source packages without any binary
packages. We just want to merge the two package sets as fast as
possible.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
We now have a way to select even older version of package (since the
newer one can be left out of the whitelist), so we can include multiple
versions of the same package into global package set.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
We need to check tags for the variant, not for the whole compose. This
results in merge always being done even if there is a single tag. For
f29 tag in Fedora this takes about 2 hours for each variant.
Relates: https://pagure.io/pungi/issue/860
Relates: https://bugzilla.redhat.com/show_bug.cgi?id=1551653
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
In one place, there was an explicit check if pkgset_koji_tag was set,
in another it was blindly referenced and assumed to be a list (with
accidental semi-success for a scalar.)
Signed-off-by: Owen W. Taylor <otaylor@fishsoup.net>
Use 'get_packages_to_gather' to fail early if these packages are not
signed with right key. This prevents us from having to wait for the
repo to be created and depsolving to finish. Unsigned dependencies will
still be reported later than previously.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
There can be packages in the tag that will not end up in the compose.
Instead of failing immediately with error, this patch delays the check
until after depsolving finishes and only checks packages that will
really be included.
This is not an issue for nodeps compose, as that already pulls in only
packages that will be composed and nothing else.
Merges: https://pagure.io/pungi/pull-request/843
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
Background story: if a compose is combining modular and traditional
compose, the configuration will contain multiple Koji tags to build
package set from (one tag for each module, plus at least one tag for the
traditional content). However some packages might be present in multiple
tags, and if the package set contains both, there's no way to control
which one will end up in the compose.
The solution for this is to give preference to the modular compose. If a
package with the same name exists in multiple tags, we only take the
first one we find. This relies on ordering of collected tags: modular
ones are always first, and traditional tags are at the end of the list.
If there are multiple modules that contain the same package, only one of
them will be used, which is not correct. Allegedly this should not
happen. In any case such use case does not work without this patch
either, so we're not losing anything.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Use global koji tag for populating package set even for modular
composes. To preserve backwards compatibility value "not-used" is
ignored. This should however be fixed in the configuration to simply not
specify the option at all.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
When cherry picking packages from Koji tag, we need to make sure that
for every binary package we always have a corresponding source package.
Even if it does not go into the compose, we need it to get values for
Exclusive or Exclude Arch tags.
This means we need to process the binary packages first and only then
look at source ones. Instead of sorting a potentially very long list,
let's just iterate twice.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
If buildinstall is supposed to run, we can't pick only some packages
from the tag. The temporary repo is used as a source for lorax. There
could very well be packages not meant to be shipped in the compose that
are still essential for building the bootable images.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Haibo Lin
Lubomír Sedlář
Jan Kaluza
Martin Curlej
Patrick Uiterwijk
Ondrej Nosek
Owen W. Taylor