The host system can run out of space while running mksquashfs, and while
this is logged to program.log it isn't detected by lorax or
livemedia-creator so it will continue running, possibly reporting
unrelated errors and causing confusion.
This adds checks for the return status when calling mksquashfs, logs it
to the log, and either exits or raises an error immediately.
Previously the release package was chosen by picking the first package
that provides 'system-release' that isn't named generic. This can cause
branding issues with repos containing multiple system-release packages.
This patch changes _install_branding so that it will give preference to
a system-release package that ends with lowercase --variant name. If
there isn't one it will fall back to the previous behavior.
Resolves: rhbz#1472622
epoch wasn't included in the lorax-packages.log, but arch is. And for
the debuginfo the epoch, including 0, *is* included. This should match
the previous output.
Resolves: rhbz#1815000
A change in glibc now requires /proc be mounted in order to run mknod
which is needed in order to run rpm from runtime-postinstall.
This drops that code from the template and moves writing the package
list into run_transaction, which already has all of the needed
information to generate the list.
Resolves: rhbz#1812895
It isn't always obvious what happened when the rootfs runs out of space,
especially when using lorax via pungi. So this checks for the out of
space error string when building the runtime image and logs it to the
primary logfile and console as an error with the rootfs size.
eg.
2020-01-20 18:52:58,920: The rootfs ran out of space with size=1
startProgram is running in binary mode, so bufsize=1 is invalid. The
ExecLineReader class already breaks it up into lines using readline() so
there is no need to change the default buffering.
On some platforms (aarch64, ppc64le) toolchain limitations/optimizations
can break anaconda startup, as discussed in rhbz#1722181. The workaround
is to preload libgomp.so before starting anaconda.
The 'enabled' field in the /compose/types output now reflects whether or
not the type is supported on the current architecture. Disabled types
are not allowed to be built, and will raise an error like:
Compose type 'alibaba' is disabled on this architecture
Since we have both compose uuids and upload uuids they need to be
clearly named. This updates the upload naming to use 'upload_uuid' in
the inputs, and 'upload_id' in the output (_id instead of _uuid for
consistency with build_id naming in the status responses).
This also adds 'upload_id' to the /upload/log response.
This allows deleting a provider's profile. Pass the provider and profile
name like:
/api/v1/upload/providers/delete/azure/test-settings
A standard json response will be returned.
This extends the /compose/ route to support uploading with either an
existing profile, or with one-time use settings passed in the POST.
To select a profile include the provider and profile, as returned by
`/uploads/providers`:
"upload": {
"image_name": "My Image",
"provider": "azure",
"profile": "production-azure-settings"
}
uploads should only be included in the V1 API routes, add `api`
selection to the relevant helper functions and calls to them from v0.py
Add new V1 routes with api=1 to include the uploads information in the
results.
Also add tests to ensure that V0 requests do not include uploads.
Currently, Azure, vSphere, and OpenStack are supported. See
https://github.com/weldr/lorax/pull/826 for more details about this new
feature.
I've called the upload library "lifted" as a reference to Seuss'
The Lorax -- in the book, the Lorax lifts himself up by the seat of his
pants through a hole in the smog clouds, and they start calling him the
"Lifted Lorax."
This adds new features to the /compose route under API v1
A recipe that is valid TOML can still be an invalid recipe (eg. missing
the 'name' field) so this should also catch RecipeError.
Also added tests for this, as well as making sure commit_recipe_file()
raises the correct errors.
Resolves: rhbz#1755068
Add dmsquash-live-ntfs to the default dracut modules in livecd builds.
The omission of this is probably why:
https://bugzilla.redhat.com/show_bug.cgi?id=1449410
didn't work out as intended (I suspect it was closed unfixed).
Without this, initramfs winds up with the ntfs-3g tools lying
around, but no hooks cause them to actually get used.
AFAICT, this is a legit bugfix, not a "proposed enhancement"
so to speak. To test that this actually fixed the problem (well,
/a/ problem, at least), I used some scripts:
https://github.com/gmt/test-respin-fedora-ntfsable-iso
Signed-off-by: Greg Turner <gmt@be-evil.net>
Signed-off-by: Brian C. Lane <bcl@redhat.com>
This loads the system dnf vars from /etc/dnf/vars at startup if
system repos are enabled, and it substitutes the values in the sources
when loaded, and when a new source is added.
Also includes tests.
This changes the source 'name' field to match the DNF usage of it as a
descriptive string. 'id' is now used as the short name to refer to the
source. The v0 API remains unchanged.
Tests for v1 behavior have been added.
Now that the v1 API is in use the status message will return api: 1
This creates a tar suitable for use with the anaconda kickstart liveimg
command. It adds the kernel, grub2, and grub2-tools packages to the tar
template.
Normally the runtime image is an ext4 filesystem image that is
compressed with squashfs. dracut now supports setting up an overlayfs
when it detects a bare filesystem tree inside the squashfs.
This commit adds support for a --squashfs-only option which places the
root tree directly in the squashfs.img
In python 3 f.seek() on text doesn't work like it does in py2/C because
text is now unicode. So change read_tail to use byte mode and take
unicode into account. Also add tests for it.
The VALID_API_STRING function allows for characters that should not be
allowed in blueprint names. VALID_BLUEPRINT_NAME allows us to
specifically check if a blueprint contains a valid name.
dnf seems to have changed the default for skip_if_unavailable. Some
mock repositories are still around in later tests, which then fail
because metadata cannot be synced.
Also expose skip_if_unavailable in dnf_repo_to_file_repo(), so that
tests checking for equality of repo files continue to pass.
