Multithreading was added to parallelize the
computation of image checksums. Resulting memory structures
are protected via synchronization primitives. Max number of
threads is uncapped- experiments were done to determine
whether a maximum number of threads would yield greater
efficiency and there were no gains from this.
Likewise, experiments were done to determine whether pools of
threads computed in separate processes could likewise decrease
compute-time. Evidence did not suggest that this was the
case. This indicate that the checksum operation is bounded
by I/O read/write times.
Merges: https://pagure.io/pungi/pull-request/1520
Jira: RHELCMP-5967
Signed-off-by: James Kunstle jkunstle@redhat.com
The original code could cause a source RPM to be present in two variants
that have a dependency relation.
There is always only one source repo for a variant in the final compose.
When gathering packages for a variant that depends on another variant,
we need to build a temporary lookaside repo that has similar content to
the parent variant. This lookaside only contained source RPMs for
packages present the the architecture.
This could result in duplicated SRPMs in the compose.
Example situation:
* Variant B depends on variant A.
* A contains foo.x86_64.rpm (only on x86_64)
* B pulls in subpackage foo-bar.s390x.rpm (on s390x)
Source repo for A will correctly contain foo.src.rpm. With original code
the srpm would also end up in B.src. By adding all sources to the
temporary lookaside Pungi will know that source repo for B doesn't need
to duplicate the package.
The refactoring to use a set to store the packages is meant to avoid
listing the same SRPM multiple times in the repo in the most common
situation when SRPM is listed in multiple architectures.
JIRA: RHELCMP-6002
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
It is possible to try to re-run a compose with old event. When trying to
reuse pkgset data, we must use set the bounds not based on
current/reused event, but actually check which was first.
JIRA: CWFHEALTH-495
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
- Do not reuse when there is any external lookaside repo
- Do not reuse when lookaside variant is not reused
JIRA: RHELCMP-4596
Signed-off-by: Haibo Lin <hlin@redhat.com>
The original code ended up downloading all repodata from the lookaside
repo. This could cause a lot of memory to be used.
The new code only downloads the repomd.xml and then primary record,
which is sufficient to obtain all needed information. A lot less memory
is used and the code is also significantly faster.
Here are some alternative ways of getting a list of packages from the
lookaside repo and reasons why they did not work:
* dnf repoquery - this doesn't include modular packages unless the
stream is default
* dnf reposync - requires `--urls` option to only print the names,
which is not available on RHEL 7
JIRA: RHELCMP-4761
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
OSBS will reject no scratch builds with arch_override.
When the option is not specified in Pungi, it would do `"".split(" ")`
to get list of arches, which returns a list with empty string instead of
an empty list.
With this fixed, it might be possible to have multiple images match the
spec (unless arch is used in the filter). To fix that, we can replace
arch with $basearch variable.
JIRA: RHELCMP-3824
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The name brings a different expectation than how it actually worked.
This patch makes the code work similarly to the expectation.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
When a module is skipped from the compose, we should not add it to a
mapping of module tags. If it's there, we then spend time building a
repo for the module, and it get's passed to buildinstall, despite the
packages not being supposed to be included in the compose.
If the packages are not included in any variant, they shouldn't be
available to buildinstall either.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
When a gathering thread raises an exception, it gets forwarded to the
main thread and re-raised there. However, during this transition it
loses details about exact location of the problem.
This patch creates an extended traceback in the worker, which should
make it easier to track the problem down later.
JIRA: RHELCMP-4259
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
If packages are appearing quickly in Koji, and signing them is triggered
by automation, there may be a delay between the package being signed and
compose running. In such case it may be preferable to wait for the
signed copy rather than fail the compose.
JIRA: RHELCMP-3932
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This phase runs after image-build and osbuild and can embed an image
into a container.
JIRA: RHELCMP-3820
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Rather than tracking this directly in OSBS phase, move this into Compose
object, which will allow access to this from multiple phases.
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This file can contain all Pungi specific exceptions.
It should also fix an issue encountered on Python 2.7:
AttributeError: 'module' object has no attribute 'pkgsets'
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
When adding extra modules via option *pkgset_koji_module_builds*, all
other versions of the same stream potentially available in a Brew tag
should be skipped.
JIRA: RHELCMP-3689
Signed-off-by: Haibo Lin <hlin@redhat.com>
In such case we never want to reuse the pkgset, as it risks leaking
unsigned packages. Safest option is to remove the file completely.
Fixes: https://pagure.io/pungi/issue/1480
JIRA: RHELCMP-3720
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The task can only fail as a whole, thus it only makes sense to set all
architectures (`*`) as failable. The correct value needs to be checked
though.
JIRA: RHELCMP-3412
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Instead of parsing the log file to get the NVR and then in turn
use that to get to the build info use the structured return value
from the koji task. The return value of the osbuild plugin is:
result = {
"composer": {
"server": <COMPOSER_URL>,
"id": <COMPOSE_ID>
},
"koji": {
"build": <BUILD_ID>
}
}
This means we have direct access to the koji build id, which was
returned by composer to the plugin via its status API. Using that
removes the need to parse the log file.
Adapt the test accordingly.
Merges: https://pagure.io/pungi/pull-request/1475
Signed-off-by: Christian Kellner <christian@kellner.me>
A new configuration *createiso_use_xorrisofs* is added to determine
which tool to use for creating ISOs.
By default, createiso_use_xorrisofs = False and genisoimage will be used.
When set to True, xorrisofs will be used.
JIRA: RHELCMP-2875
Fixes: https://pagure.io/pungi/issue/1130
Signed-off-by: Haibo Lin <hlin@redhat.com>
Previously, the pkg_object has been found by iterating over pkg_sets
for each package. This was quite slow given the number of RPMs in the
compose.
In this commit, the temporary dict is created which stores mapping
between path and pkg_obj and is used instead.
In this commit, the get_package_path is called just once instead of
twice in the loop which might also save some time.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
The pkgset_koji_builds influences the gather phase, but the
change itself is not a reason to not reuse old gather phase. if
new pkgset_koji_builds value leads to significant change in input
package set, we will find that later in this function when comparing
old and new package set.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
When testing if a variant has some modules,
it's no longer enough to look at `variant.modules`. That attribute
contains only names of modules configured in variants.xml.
With modules being specified in `pkgset_koji_module_builds` or
`pkgset_scratch_modules` options it is possible the code wouldn't
trigger even if there was a module.
JIRA: RHELCMP-3054
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This is similar to image-build in terms of what it does, and somewhat
similar to OSBS phase in how it's implemented.
The phase reads configuration, submits the build via XMLRPC call and
waits for the task to finish. Then it downloads the built image and
includes it in the compose metadata.
JIRA: RHELCMP-315
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
When list is used in gather_lookaside_repos, the Pungi currently
fails with an exception. This is caused by inverted condition
in the code which tries to filter-out the lookaside repos
generated during the Pungin execution pointing to different
compose variants.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
There are two thread pools for making package sets. If Pungi is being
terminated by external event and the exception is handled in the first
thread, the second one never gets to the `stop` method and the process
keeps hanging.
This patch should make sure that `stop()` is called on both pools.
JIRA: RHELCMP-2459
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
The change allows for setting the parameters as described below to Lorax.
Lorax, a program called during the buildInstall phase, creates the SquashFS
during the buildInstall phase.
The Squash filesystem is present both on the DVD and the BOOT.ISO.
squashfs_only --- (str) passes --squashfs-only option.
configuration_file --- (str or scm_dict) passes -c option to Lorax.
The final goal of this change is to allow for optimization of
the installation medium size.
This pull request is related to the Fedora change proposal, which is available
at this location:
https://fedoraproject.org/wiki/Category:Changes/OptimizeSquashFS
See the change proposal for more information about the benefits of higher
compression ratio.
Jira: RHELCMP-693
Signed-off-by: Bohdan Khomutskyi <bkhomuts@redhat.com>
This uses a previous change that added ability to retry and simply
extends the conditions when it fires.
JIRA: RHELCMP-1863
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Reusing old arch repo may fail for reasons such as arch not
available in old compose or unexpected error when copying data
from old compose.
JIRA: RHELCMP-994
Signed-off-by: Haibo Lin <hlin@redhat.com>
Logs of DepsolveError will be printed in the main log.
Fixes: https://pagure.io/pungi/issue/1399
JIRA: RHELCMP-955
Signed-off-by: Haibo Lin <hlin@redhat.com>
This patch updates the documentation to match the actual behavior, and
adds a configuration option to remove the leading prefix.
The extra wildcard is causing problems when there are two variants in
the compose and one UID is a suffix of the other (e.g. DevTools and
Tools), since multiple files will match the shorter name and an error
will be reported.
JIRA: RHELCMP-1086
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This is a workaround for modularity design issues and DNF bugs. If there
were gaps in contexts, DNF has trouble handling the upgrades. Thus we
may need to add module metadata for older versions of previously
released module streams and add the missing contexts.
JIRA: RHELCMP-982
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
This commits changes `ostree` and `ostree_installer` phases
so they can run with Koji Pungi plugin instead of the plain runroot.
It is similar to `buildinstall` phase running with Koji plugin.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
If the input packages contain e.g. `ansible`, and there's
`ansible-runner-service` in the package set, we don't want to pull it
in.
JIRA: RHELCMP-446
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
New `buildinstall.metadata` file is created once the buildinstall
phase is done. This file contains:
- list of lorax command line arguments.
- list of RPMs installed in the buildinstall buildroot.
- list of RPMs installed in the resulting boot.iso.
This file is checked in the next compose run to find out if
the result of buildinstall phase from the previous compose
can be reused. Following is checked:
- lorax commandline arguments are the same (except of expected
differences).
- The NVRAs of RPMs in the runroot_tag are the same as the ones
installed in the old buildinstall buildroot.
- The NVRAs of RPMs installed in the boot.iso are the same as
the ones in package sets in the current compose.
By its implementation, this reuse strategy is used only if
pungi_buildinstall Koji plugin is used.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
Add tests for buildinstall reuse and buildinstall_allow_reuse option.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
Move repoclosure out from test phase into its own phase and
run parallel with image building phases(osbs, imagebuild, ...)
to speed things up.
JIRA: RHELCMP-8
Signed-off-by: Haibo Lin <hlin@redhat.com>
I've analysed multiple nigthly composes and often the only difference
in the configuration between two nightly composes is different
`product_id` commit hash.
The `product_id` is used in later `createrepo` phase and does not
influence the gather phase at all. I therefore think it can be
whitelisted in gather phase reuse code.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
When modules are used, there are lot of small package sets. These
package sets have usually less than 500 packages. The createrepo
part of `MaterializedPackageSet.create` executed for such small
set of packages takes around 1 second. Most of this time
the createrepo_c runs in single thread. It does the initialization,
it writes the XML files, ...
The parts of createrepo which can be run in parallel and therefore
would use all the CPUs are quite small for very small package sets.
This commit therefore executes multiple threads with
`MaterializedPackageSet.create` for these very small package sets.
This saves around 40 seconds from pkgset phase for RHEL compose.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
- Get also requires/provides of RPMs in package set.
- Store the results of gather phase as pickle file.
- Reuse old gather phase results in case Pungi configuration
did not change, the "names" of RPMs in global package set
did not change and their requires/provides did not change.
- Add `gather_allow_reuse` option to enable this feature.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
Add gather_allow_reuse, add more tests and better handling of gather_lookaside_repos.
The current code calls `find_old_compose` followed by multiple `os.path.*`
calls to find out if particular file exists in the old compose. This
duplicates code a lot and makes it harder to read.
In this commit, new `Compose.old_compose_path` is introduced and
used instead of direct calls of `find_old_compose`.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
If Koji pungi-buildinstall is used, then the buildinstall results are
stored in the `output_dir` dir, but in "results" and "logs" subdirectories.
We need to move them to final_output_dir.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
We would like to start generating the buildinstall phase using the safer
Koji Pungi Buildinstall plugin and stop the direct use of Runroot plugin.
The plugin so far exists only as PR for Koji:
https://pagure.io/koji/pull-request/1939
This commit adds support for this plugin when `lorax_use_koji_plugin`
is set to `True`.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
There was an attempt to achieve this by customizing --version and
--release arguments for Lorax, but Pungi does not take the [general] and
[release] sections from its .treeinfo. Instead it was always using
release version.
The value from this new option will be put into .treeinfo and used for
--version and --release arguments in Lorax (unless explicitly defined in
lorax_options to another value).
JIRA: COMPOSE-4029
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
When `link_type = "symlink"` is used, the packages are in fact symlinks
to /mnt/koji. When graft points file is generated, the paths in this graft
points file point to symlinks and therefore symlinks are copied into the
generated ISO file instead of real files.
In this commit, the code to generate the graft points file is changed
so it resolves the symlink to real file stored on /mnt/koji. To make
this code safer, it does such resolving only in case the symlink points
outside of `compose.paths.compose.topdir()`. Therefore you can still
generate ISO file with symlink pointing to file stored within the ISO
file itself, although this is not done currently afaik.
The main reason for this is to be able to generate ISO files even
without hardlinks (which would need read-write access on /mnt/koji)
and without copying all the packages from /mnt/koji to local storage.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
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>
The `runroot_method` now accepts `dict` value with phase name as a key
and runroot method as a value. For backward compatibility, the `str`
value is still supported.
The new `global_runroot_method` option has been added which defines
the runroot method in case it is not set in `dict` in the `runroot_method`.
This commit allows running `createiso` phase locally while keeping the other
phases in Koji.
Signed-off-by: Jan Kaluza <jkaluza@redhat.com>
Theoretically it is possible, and this is the bare minimum of changes
needed to survive such situation. There may be other pitfalls.
Fixes: https://pagure.io/pungi/issue/1309
JIRA: COMPOSE-4016
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
Originally the list of solvables for fus was growing with each iteration
and nothing was ever removed. That later changed so that fus iterations
are only done on newly added stuff. It's great for performance, but
means that the last log is not a superset of all others.
To get all dependency problems we need to look into all log files, not
just the last one.
JIRA: COMPOSE-3964
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
When there is an exception in gathering (such as after seeing unsigned
packages in deps method), the exception was lost and the compose
continued to run until it tried to access the result and crashed on
KeyError.
Relates: https://pagure.io/releng/failed-composes/issue/587
JIRA: COMPOSE-3986
Signed-off-by: Lubomír Sedlář <lsedlar@redhat.com>
When running repoclosure as root user, it will use other dir instead of
the one returned by getCacheDir().
For yum, with --tempcache option could let the cache dir returned by
getCacheDir() always be used.
For dnf, there's no such an option and we have to handle it specially.
JIRA: COMPOSE-3922
Signed-off-by: Haibo Lin <hlin@redhat.com>
Dominik Rumian
JamesKunstle
Haibo Lin
Lubomír Sedlář
Lev Veyde
Christian Kellner
Jan Kaluza
Jiri Konecny
Bohdan Khomutskyi
Buvanesh Kumar