As described in the comments inline, on a selinux enabled kernel (such
as a centos build host) you need to have permissions to change the
contexts to those the kernel doesn't understand -- such as when you're
building a fedora image.
For some reason, setfiles has an arbitrary limit of 10 errors before
it stops. I believe we previously had 9 errors (this mean 9
mis-labeled files, which were just waiting to cause problems).
Something changed with F26 setfiles and it started erroring
immediately, which lead to investigation. Infra builds, on
non-selinux Ubuntu kernel's, would not have hit this issue.
This means we need to move this to run with a manual chroot into the
image under restorecon.
I'm really not sure why ironic-agent removes all the selinux tools
from the image, it seems like an over-optimisation (it's been like
that since Id6333ca5d99716ccad75ea1964896acf371fa72a). Keep them so
we can run the relabel.
Change-Id: I4f5b591817ffcd776cbee0a0f9ca9f48de72aa6b
As described in the comment and associated bugzilla, the behaviour of
setfiles has changed in Fedora 26 to require "-m" situations where
labeled file-systems are mounted below non-labeled file-systems. Our
loopback/chroot system appears to trigger this nicely, leading to a
setfiles call that does nothing without this.
Change-Id: I276c6f6a4fb44f4bea5004f6b4214f94757728ae
Signed-off-by: Paul Belanger <pabelanger@redhat.com>
As described in the referenced bug, the dependency solver in yum
doesn't handle weak dependencies well and in some cases, such as
Fedora 26, can end up choosing coreutils-single (the busybox-esque
single binary) instead of actual coreutils, which then causes problems
with conflicting packages later.
Change-Id: I2907bf3b74c146986b483d52cc6ac437036330b4
On a system where the packaged pip/virtualenv is up-to-date with
upstream (such as Fedora 26 ... for now), we don't reinstall, which
then violates a bunch of assumptions later on. Force install.
Change-Id: I6ebcda0351997fa7e32f0e6e77a98b2c33764e3f
It turns out dnf argparse can't handle negative numbers without "=".
It's actually documented in the man page
--latest-limit <number> ... If <number> is negative skip <number>
of latest packages. If a negative number is used use syntax
--latest-limit=<number>
But who reads that :) This started failing with Fedora 26
Change-Id: I884af94c07fa11b010f69863047a04711b14f21e
We expect LC_ALL for non-C locales to be working inside
images, so always install glibc-locale for openSUSE.
Change-Id: I8fe92773e377539070d9d9fe2960a6202bb80a18
The code in mkfs correctly extends the command line with a '-n' for
vfat but does not currently do it for fat. This means that mkfs for
fat ends up with a '-L' which is what you'd do for everything like
ext[234].
The change just treats fat like vfat in the one place where this check
is required.
Change-Id: If65dfd949acdadff33a564640fb42ea73026a786
Closes-Bug: #1703063
The purpose of the openSUSE element is to build openSUSE distribution
based images, so an additional community repo shouldn't be pulled into
the image. In addition the dkms dependency is blacklisted for SUSE
in the dkms element anyway, so this should be a noop.
Change-Id: I0aa06d9f4f110546032f910e3361840693d02de7
On Power systems console should be added the kernel command line
in the following order: 'console=tty0 console=hvc0'.
The first one is the graphical console. The last one is the serial
console. The kernel enables all the consoles pointed through the
kernel command line. However, only the last one will receive
input/output during kernel boot. All the other consoles will be
enabled after the boot.
Change-Id: I0069f608e0ab104d3778954e033fb82ed5ea7693
We replace the base resolv.conf with an "outside" copy so that
resolving works when we're in the chroot.
Installing resolvconf package modifies the in-chroot resolv.conf to a
symlink (to /var/run) which it wants maintained in the final image.
We have the existing "immutable" check for a created resolv.conf file,
but no eqivalent for a symlink.
This adds a check to see if the resolv.conf is a symlink and leave it
alone if it is, assuming it has been re-created in the chroot.
I have tested this with ubuntu-minimal+resolvconf with
dhcp-all-interfaces and the system seems to work with resolvconf
working correctly.
Change-Id: Idd5a26e9d55979bd951577d5b098ed4bfba91ad3
The python3 package actually contains some core modules (like the xml
one) which are not present in the python3-base on which is pulled by
the python3-devel package. As such, it's best to have it installed
similar to python-xml for python2.
Change-Id: I5cd5d1127ae62d6753c2ace44965179c5400bb9a
In order to support {CentOS,RHEL}7 for building cloud images we need to
handle the differences in grub packaging from Ubuntu. We also need to
populate the defualt location for cloud images for CentOS builds.
Change-Id: Ie0d82ff21a42b08c4cb94b7a5635f80bfabf684e
When a download redirector redirects to a broken mirror, timeout
quickly rather than waiting until the overall job is being timed out.
Change-Id: If7eb63d406aaf61f71aa9203cf708c474aa63fd0
The 'packages' variable already contains the packages we need so
use it instead of duplicating the packages.
Change-Id: Id22e1862f9654e66252d03a0fed9839cf004d750
Several people have popped up in IRC recently with failures in these
elements. Without Python 2.7 available in the image they are
unsupported (OpenStack hasn't supported it for a long time). Remove
these to avoid further confusion.
The centos/centos7 DISTRO split that has happened with centos-minimal
is unfortunate but I don't think it helps to rename centos7/rhel7 ATM.
To summarise; DISTRO=centos7 means image based build,
DISTRO=centos && DIB_RELEASE=7 means the minimal build.
In the future, I think it is important that the minimal builds and
image builds set the same DISTRO. This reflects that "upper" layers
shouldn't care about the exact building of the lower layers. I see
CentOS 8 going one of two ways
1) the changes are so significant, we start separate centos8 /
centos8-minimal elements. They both set DISTRO=centos8 (and
DIB_RELEASE to point-release maybe?). This means we have to update
all "if DISTRO == centos || DISTRO == centos7" branches to also check
for "centos8". Evenually (!) "centos" goes away for versioned DISTRO
only
2) we restore centos element with DISTRO=centos and DIB_RELEASE=8, and
centos-minimal remains the same. This means we have to audit all "if
DISTRO == centos" calls to make sure they're appropriate for version 8
(stick a "&& DIB_RELEASE=7" on them all basically).
I'm not sure we can fully decide until we start to see excatly how the
distro switching/matching bits look, but (2) is consistent with Ubuntu
and probably the preferred solution.
Some "rhel" parts have been cleaned up. More could be done in
rhel-common, but given our lack of coverage of that I'd prefer to
leave it for now.
Change-Id: I6ea784116ef59ca22878c8512c963f29c815a00a
The image download tests have long been too unreliable for the gate.
We need to cache the base images similar to how devstack caches it's
testing images. Let's move them to non-voting jobs for the time
being.
This means that the gate jobs are now all based on "-minimal" and are
using infra mirrors. Unfortunately, there is still some unreliability
because we currently have issues with infra mirrors being very slow
after AFS updates, leading to job timeouts. But we're on the right
path...
Also, I noticed we don't have tests of the "ubuntu" image-download
based tests, which were tacitly being tested by apt-sources before we
moved that to -minimal. Add simple tests for these.
Change-Id: Ie33ee49656872467ef68d753210032156bb6b2cb
In a system where python2 is not installed and /usr/bin/python is not
linked then the cleanup process will fail trying to invoke the python
script. Use the previously determined DIB_PYTHON_EXEC if it's available.
Change-Id: I128292808ccef92cc1803988b35caae5aa6fa541
This was previously defined as python2-devel (which is what rhel uses),
but the actual package name is python-devel. See:
https://software.opensuse.org/package/python-devel
Change-Id: Id61e5b05772d10c32b33d3e70cb64d5ebdcba6e4
I'm uncertain as to why this is using the "fedora" element for testing
... but it requires downloading the fedora .qcow on every test which
has shown to be unreliable. An easy thing to do is to switch it to
fedora-minimal; that will only involve downloads from local mirrors in
the gate.
Add redhat-rpm-config for minimal. I admit I have not fully gone
through why this is not pulled in. It's been an issue since
I459f2203fa145049dda185da952813118193d573 and there's all sorts of
bugs.
Change-Id: I37458e3926dae32a259bd5aa9efc645561b029a0
fedora/centos-minimal don't obey DIB_DISTRIBUTION_MIRROR currently. I
don't really want them too -- we want to be able to separate the
mirrors used during the build process from those embedded into the
final image. Add DIB_YUM_MINIMAL_BOOTSTRAP_REPOS which is a directory
with repo files to use during the install.
This introduces setup-gate-mirrors.sh which is intended to setup
repo/sources/whatever files in the openstack gate that point to the
local region mirror. It pulls the info from the mirror_info.sh script
on each CI node.
The openstack-ci-mirrors element is updated to export these variables.
elements are updated to depend on it. Tests are restored
Change-Id: I7604fc4d41cb1483be16b8d628a24e8fc764f515
This adds "openstack-ci-mirrors" element which performs various
settings to get builds using local mirrors. As a first step, we
convert ubuntu-minimal jobs
The main trick is that since infra mirrors are created with rerepo
they are not signed (they are recreated, not cloned, and not signing
is seen as a feature in that it deters external use). So we need to
instruct debootstrap to ignore signing and also turn it off for
in-chroot apt. Other than that, the existing DIB_DISTRIBUTION_MIRROR
works to redirect installs.
Remove "restricted" as it's not mirrored, and I don't think we want it
in here by default.
(I think DIB_DISTRIBUTION_MIRROR is a bit of an anti-pattern, because
it leaves the mirrors in the final image -- just because you use them
to build, doesn't mean you want them at runtime). But we don't need
to fix that now, and we don't use any created images.)
This pauses fedora testing until the next change, which moves to using
local mirrors for testing on fedora/centos
Change-Id: I778bd05a1e615c27edf1c9f0a1409119a6b3a850
The gate is currently extremley unstable, and these two issues are
causing most of the problems. We need to commit them atomically so we
can get anything moving again
---
The gate is very unstable downloading the ubuntu tarballs from
upstream at the moment. Move this to ubuntu-minimal which, in a later
change will source files from our local mirror.
We need a caching mechanism for these large files to avoid this
instability. This is future work for the various image-based jobs.
---
Move debian to default skip lists
I don't know if it's mirrors being worked hard for the Stretch
release, but this is constantly failing the gate. I will move this to
the -nv extras job
I am working on having the voting job use local mirrors for
everything. Unfortunately debian infra mirrors don't have stretch yet
and we need to do some fiddling to get "stable" available. Once we
have all this, we can consider making it voting again.
Change-Id: Iaf7b3888ef06c7aef63cbf76a94b33f96bc9c5c2
We introduced the "settle" in
I90103b59357edebbac7a641e8980cb282d37561b thinking that maybe kpartx
had not finished writing the partition. This probably wasn't a bad
first assumption, since we used to have this -- but is seems
insufficient.
The other failiure here seems to be if kpartx hasn't actually seen the
updated partition table in the image, so it has correctly (in it's
mind) not mounted the partition.
Looking at strace of fdisk run manually on a loopback, it will do a
fsync on the raw device after writing and then a global sync as it
exits.
This replicates this; we flush and fsync in mbr.py in the exit handler
after writing the partition, before closing the file (i've updated one
of the unit tests to double-check the call). In the partitioning.py
caller we execute a sync call too.
Since it does seem unlikely the "-s" option of kpartx is not working,
I've removed the udev settle work-around too.
Change-Id: Ia77a0ffe4c76854b326ed76490479d9c691b49aa
Partial-Bug: #1698337
Debian Stretch released as stable recently, and the init system is
less tightly specified in the base dependencies (for some info, see
[1]). It seems, probably unintentionally, that in the previous
release systemd-sysv was brought in by debootstrap, but that is no
longer happening.
Add systemd as an early dependency of debian-minimal.
Remove the package-installs.yaml as that happens too late (other
things need to know the init system to write out service files, etc
and probe for systemd utils before package-installs). As mentioned, I
do not believe the "only install systemd on testing" idea was actually
working here, because it was being brought in during the initial
debootstrap.
Update some documentation to explain what's going on
[1] https://lists.debian.org/debian-boot/2015/05/msg00156.html
Change-Id: Id67c0cf08728407d234976f9807d3bd71d12f758
There was a race in diskimage-builder where the mkfs call after a
kpartx -avs for the loop device would fail because the device was
not yet ready. This adds a udevadm settle call after the kpartx
to make sure the udev event queue has cleared.
Change-Id: I90103b59357edebbac7a641e8980cb282d37561b
Closes-Bug: #1698337
Using the newly exposed variables from the prior change, install the
ppc bootloader to the boot partition, not the underlying loopback
device.
Change-Id: I0918e8df8797d6dbabf7af618989ab7f79ee9580
Currently we only export "image-block-device" which is the loopback
device (/dev/loopX) for the underlying image. This is the device we
install grub to (from inside the chroot ...)
This is ok for x86, but is insufficient for some platforms like PPC
which have a separate boot partition. They do not want to install to
the loop device, but do things like dd special ELF files into special
boot partitions.
The first problem seems to be that in level1/partitioning.py we have a
whole bunch of different paths that either call partprobe on the loop
device, or kpartx. We have _all_part_devices_exist() that gates the
kpartx for unknown reasons. We have detach_loopback() that does not
seem to remove losetup created devices. I don't think this does
cleanup if it uses kpartx correctly. It is extremley unclear what's
going to be mapped where.
This moves to us *only* using kpartx to map the partitions of the loop
device. We will *not* call partprobe and create the /dev/loopXpN
devices and will only have the devicemapper nodes kpartx creates.
This seems to be best. Cleanup happens inside partitioning.py.
practice. Deeper thinking about this, and more cleanup of the
variables will be welcome.
This adds "image-block-devices" (note the extra "s") which exports all
the block devices with name and path. This is in a string format that
can be eval'd to an array (you can't export arrays).
This is then used in a follow-on
(I0918e8df8797d6dbabf7af618989ab7f79ee9580) to pick the right
partition on PPC.
Change-Id: If8e33106b4104da2d56d7941ce96ffcb014907bc
Currently we pass a reference to a global "rollback" list to create()
to keep rollback functions. Other nodes don't need to know about
global rollback state, and by passing by reference we're giving them
the chance to mess it up for everyone else.
Add a "add_rollback()" function in NodeBase for create() calls to
register rollback calls within themselves. As they hit rollback
points they can add a new entry. lambda v arguments is much of a
muchness -- but this is similar to the standard atexit() call so with
go with that pattern. A new "rollback()" call is added that the
driver will invoke on each node as it works its way backwards in case
of failure.
On error, nodes will have rollback() called in reverse order (which
then calls registered rollbacks in reverse order).
A unit test is added to test rollback behaviour
Change-Id: I65214e72c7ef607dd08f750a6d32a0b10fe97ac3
Keep track of the mount-point ordering in a state variable, rather
than a global. This path is tested by existing unit tests.
Note a prior change inserted the MountNode objects directly into a
list in self.state, which makes sorting quite easy as it can just
implement __lt__. Unfortunately we still json dump the state, and
thus we can't have aribtrary objects in it (future work may be to
check keys inserted into the status object...). So we have to do a
bit of wrangling with tuple lists and comparision functions here, but
it's not too bad.
Change-Id: I0c51e0c53c4efdb7a65ab0efe09a6780cb1affa8
As we add file-systems, add them to global state and check the labels
are uniqiue. Add a unit test and remove the old global value.
Bonus fixup to the length check, and a test for that too.
Change-Id: I0f5a96f687c92e000afc9c98a26c49c4b1d3f28d
With I468dbf5134947629f125504513703d6f2cdace59 each node has a
reference to the global state object. This means it gets pickled into
the node-list, which is loaded for later calls. There is no need to
reload the state.json it and pass it for later cmd_* calls, as the
nodes can see it via the unpickled self.state
Change-Id: I9e2f8910f17599d92ee33e7df8e36d8ed4d44575
