The current spot price could be limiting the available instance pool
significantly. ARM instances specifically are experiencing a lot of
capacity errors.
The current path sometimes launches two instances, which is problematic
because the rest of the secure instance code expects exactly one
instance. A security group could be attached to both instances, and
would block the worker from launching any more SIs, as it tries to
delete the old security group first, which is still held by one of the
surplus SIs which didn't get terminated.
Only retry if:
- on "UnfulfillableCapacity" or "InsufficientInstanceCapacity" error codes;
- there wasn't an instance launched anyway.
If either of these checks fail, do not try to launch another one, and
just fail the job.
The errors returned by create fleet are not entirely clear. It seems it
also returns `InsufficientInstanceCapacity` in addition to
`UnfulfillableCapacity`. Let's just retry three times regardless of the
create fleet error, that way there's no need to chase error codes which
aren't clearly defined.
In case the on demand option failed as well, retry one more time across
availability zones. This significantly increases the pool of available
instances, but increases network related costs, as transferring data
between AZs is not free.
Extend the API unit test for Koji composes, to verify that the newly
added /sboms endpoint works correctly.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
Extend the unit test for regular (non-Koji) composes, to verify that
the newly added /sboms endpoint works correctly.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
Add a new /sboms API endpoint, for getting SBOM documents for a given
compose ID. The endpoint returns an array of SBOM documents for each
image built as part of the compose. For each image, there is an SBOM
document for each osbuild pipeline, which installs RPM packages. This is
usually one 'buildroot' and one 'image' pipeline.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
Extend the `manifestJobResultsFromJobDeps()` function to also return the
manifest `JobInfo`. This will be useful to inspect the job dependencies
and eliminate the need to add a specialized function for getting only
the `JobInfo`.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
If the Koji target result contains information about any uploaded SBOM
documents, import them to Koji as part of the finalize task.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
For Koji composes, all files are uploaded to Koji as part of the osbuild
job (specifically as part of handling the Koji target). So in order to
be able to upload SBOM documents to Koji as part of Koji compose, the
osbuild job needs to to be able to access the depsolve job result, which
contains the SBOM documents. For this, the osbuild job must depend on
the depsolve job.
For Koji composes, make sure that osbuild job depends on the depsolve
job and set the DepsolveDynArgsIdx.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
Adjust all paces that call `Solver.Depsolve()`, to cope with the changes
that enabled SBOM support.
Fix loading of testing repositories in the CloudAPI unit tests.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
gce-rhui is now gone from RHEL 9 [1] and the old name simply aliases to
gce. gcp-rhui in the cloudapi now resolves to 'gce' in RHEL 9 and
'gce-rhui' in RHEL 8.
[1] https://github.com/osbuild/images/pull/857
The sshkey customization in osbuild/images has been dropped. In
osbuild-composer we maintain it for backwards compatibility, converting
each to a user customization, which is a superset of the sshkey.
When the osbuild worker cannot register itself with the server
on startup the worker will "crash". This is inconsistent with the
existing behavior in `workerHeartbeat()` which deals with connectivity
or other server issue gracefully and retries periodically.
To unify the behavior this commit changes the behavior and only
issues a `logrus.Warnf` instead of the previous `Falalf` when
the registration fails.
Co-authored-by: Florian Schüller <florian.schueller@redhat.com>
Also adds the policy id to the blueprint, this doesn't have any effect
on the openscap step, it just puts in place the rhsm fact so instances
registered to insights will appear under that policy.
Previously, the worker was determining the GCE image guest OS Features
on its own, based on the OS name. This caused problems, in case the
osbuild-composer was of a newer version than the worker.
Example:
osbuild-composer contained support for c10s GCE image type and its
implementation also contained the proper guest OS Features list for it.
However, when the worker got the osbuild job, it built it and tried to
fetch the guest OS Features for the distro. Since its implementation was
too old, it didn't contain the code that added the actual support for
c10s GCE images and got no guest OS features list (which is the default
for unsupported distros). The image was successfully uploaded and
shared, but it does not boot in GCP, because it does not know that it
should use UEFI to boot it.
This behavior could be considered a bug. The worker should be dumb. It
should not be making decisions about the image features, but instead it
should take them from the upload target options. And composer should be
the authoritative source of truth for this. Because otherwise, we
basically have two components that need to be updated in sync to add
support for GCE images on a new distro.
Move the GCE image guest OS features to the GCP upload target options.
The worker will just take what is specified there and use it when
importing the image to GCP. As a compatibility layer for the case when
the composer would be older than the worker (unlikely, but still),
worker will try to determine the image guest OS features in case the
list in the upload target options is empty.
Extend the GCP functional tests to check that the imported image has at
least some guest OS features set.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
I find it slightly eaiser to read this code when
`api.BasePath = conf.BasePath` is right at the top as it's
unrelated to the parsing code below.
Note that the code itself is problematic:
- api.BasePath is global but client is not, this means that
multiple client with different configs will result in
api.BasePath being potentially wrong
- api.BasePath is set in a non-thread safe manner
Changing is a bigger job but we might consider it (IMHO).
The current name `Client.server` feels a bit misleading as it is
unclear if this is an abstraction for a "server" object or an
URL. This rename makes this unambiguous.
Add support for RHSM customizations, which currently allow various
aspects of RHSM. Specifically enabling / disabling DNF plugins shipped
by subscription-manager package and setting a subset of options in the
rhsm.conf.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
Add support for RPM customizations, which currently allow to import RPM
GPG key from a file installed in the image. This is e.g. done for the
Azure RHUI image type.
Signed-off-by: Tomáš Hozza <thozza@redhat.com>
