As described inline, Bionic hosts will build invalid Trusty images.
Hack around this by disabling metadata_csum in the ext4 mkfs.
Change-Id: Ibd67d58ca830a9e60605d0700ee2b17906c804e6
This reverts commit a3e9e7f89e.
We still have some issues with vhd creation on RAX
In short, it appears that images fail to resize unless they have a
specific "creator" field. Revert this while we consider the options.
[1] https://bugs.launchpad.net/nova/+bug/862653
Change-Id: I2b6a3bfbfe28432fbb6a2ce4a0211939d224b8d5
The vhdutil utility is completely dead; the whole subsystem it relies
on was removed with [1] so it's not even vaguely possible to keep it
up-to-date.
I took the .raw images on a nb and used the qemu-img there (so Xenial)
and generated some VPC images; uploaded them to rackspace and the all
seemed to boot fine. If there was a problem, maybe it's been fixed on
either the qemu or RAX side in the previous few years.
Thus swith to qemu-img to generate the vhd images too.
[1] https://xenbits.xen.org/gitweb/?p=xen.git;a=commit;h=5c883cf036cf5ab8b1b79390549e2475f7a568dd
Change-Id: I3099d2ebb958370fcec623087a093b2c8dbdc6c4
Add a new environment variable $DIB_GZIP_BIN allowing builders to
specify a different gzip (such as pigz) to be used when compressing
tgz images.
Change-Id: Ifb617568140a149e2fda241e07ff8a59429e6697
As noted in the change, 7fd52ba841
increased the size of the EFI partition considerably. This has meant
that our padding upwards of the disk size is insufficient and EFI
builds (arm64 in particular) is failing due to out-of-disk errors
during final image operations like installing kernels.
Similar to the discussion we had in
I65fa13a088eecdfe61636678578577ea2cfb3c0c, this feels a bit ugly
because we're mixing logic here with sizes specified in block-device
config files. But it boils down to the same problem; we are
calculating the disk size here and passing it to the block-layer, so
unless we want to make large changes to the status quo about where
these sizes are calculated, small adjustments here are the most KISS
solution.
Thus we check if we have selected the EFI bootloader element, and thus
assume there will be a large system EFI partition and expand the disk
size accordingly.
Change-Id: Ifa05366c2f2b95259f3312e4dde8c85347075ba1
I want to use the new --image-extra-size flag[1] but my use-case
calls for megabyte granularity of this value. Rather than adding
60% to an 800MB image, maybe I only want to add 100 or 200MB, etc.
[1] https://review.opendev.org/#/c/655127/
Change-Id: I8fb9685d60ebb1260d5efcf03c5c23c561c24384
Currently diskimage-builder supports two ways to specify the image
size. One is defining a fixed image size using DIB_IMAGE_SIZE, the
other one is auto-detection while adding a security margin of 60% as
free space. This means when building larger images (e.g. >100GB) with
unknown size upfront we end up with much wasted space, IO and network
traffic when uploading the images to several cloud providers. This can
be optimized by adding a third way by defining DIB_IMAGE_EXTRA_SIZE to
specify the free space in GB. This makes it possible to easily build
images of varying sizes while still minimizing the overhead by keeping
the free space constant to e.g. 1GB.
Change-Id: I114c739d11d0cfe3b8d8abc6df5ff989edfb67f2
In many cases, the statically sized 64MB journal is far below the
e2fstools default calculation[0] which calls for a 64MB journal only
on filesystems smaller than 16GB. On bare metal in particular, the
correct default journal size will often be in the 512MB-1GB range.
Since we cannot know what the target system is, this should be a
tunable parameter that the user can set depending on the intended
image usage.
Add a DIB_JOURNAL_SIZE envvar and --mkfs-journal-size parameter
to the image creation so users can override the default journal
size.
[0] https://git.kernel.org/pub/scm/fs/ext2/e2fsprogs.git/tree/lib/ext2fs/mkjournal.c#n333
Change-Id: I65fa13a088eecdfe61636678578577ea2cfb3c0c
I'm not really sure why I originally had --logfile also log to stdout
in I202e1cb200bde17f6d7770cf1e2710bbf4cca64c, but it seem
counter-intuitive (indeed, I just tripped myself up thinking that in a
devstack job "--logfile" would put the logs into a separate file and
avoid the stdout logging, and I wrote it!).
Make it so specifying a --logfile puts dib into quiet mode for stdout.
Explicitly overriding DIB_QUIET will allow both if someone wants that.
Change-Id: I3279c9253eee1c9db69c958b87a0ce73efc0be9b
While trying to get docker image pre-caching to work we couldn't get a
docker daeomon to run within the chrooted environment. However we got
docker running with the help of bwrap outside of the chrooted
environment. The only option so far for this is the block-device.d
phase. But this has the problem that it runs after the image size has
been calculated. This leads to broken builds if the docker images
being pulled are big.
This can be solved by adding a post-root.d phase that runs outside the
chroot but before the image size calculation.
Change-Id: I36c2a81e2d9f5069f18ce5b0d52c5f1c7212c3ae
In exploring Gentoo caching, it was realised that we have no way to
bind mount the cache into the finalised image for the finalise.d
phases.
By adding a pre-finalise.d phase that runs outside the chroot, we can
mount outside things into the hierarchy at $TMP_BUILD_DIR/mnt which
are then seen by the in-chroot finalise.d phase.
This is similar to the pre-install phase
Change-Id: I9d782994843383ddf90f62c40498af9925fd9558
Redirecting our output through outfilter.py is inherently a bit racy,
since the disk-image-create process will exit, and then you might get
outfilter.py flushing any remaining output as it closes.
On an interactive prompt this might lead to final output overwriting
the prompt, etc. This can be a bit confusing when you start running
things in a loop.
If we save the original fd, then on the exit path close the redirected
fd's and wait a little bit for final output (as a result of the
close), we get a more consistent output.
Change-Id: I8efe57ab421c1941e99bdecab62c6e21a87e4584
Strip everything before "site-packages" in the output filename for the
PS4 prompt. This makes the line in debug logs significantly shorter
as we don't have the full virtualenv path every single time. The
important thing -- the file being called in the lib/ dir, is retained.
Change-Id: I00706b6f6c0425c7795f997c08ceda3374dc84b5
In a couple of places we use flock for critical sections, but we leave
lockfiles around in various locations which can be confusing.
Introduce DIB_LOCKFILES global (under ~/.cache/dib/lockfiles) and
write lockfiles in there.
Fix up removal of the lockfile in the yum path; we just want to make
sure we cleanup the .rpmmacros file, but we don't need to remove the
lockfile as well.
Co-Authored-By: Andreas Florath <andreas@florath.net>
Change-Id: Ie810b2836be521325afe923708d046112e1e1e20
The call to fstrim in disk-image-create is currently useless, because
at the time this is called, the file systems were already umounted by
the block device layer.
The current implementation of the block-device mount plugin does not
call fstrim at all: resulting in larger image sizes.
This patch removes the useless fstrim call from the disk-image-create
script and moves this into the block-device mount.py.
The resulting image might be much smaller. Example: Ubuntu Xenial
with some elements; once with and once without this patch:
-rw-r--r-- 1 dib dib 475661824 Sep 16 06:43 ubuntu-xenial-without-fstrim.qcow2
-rw-r--r-- 1 dib dib 364249088 Sep 16 09:30 ubuntu-xenial-with-fstrim.qcow2
Change-Id: I4e21ae50c5e6e26dc9f50f004ed6413132c81047
Signed-off-by: Andreas Florath <andreas@florath.net>
This adds a devstack-inspired output filter to standardise
timestamping.
Currently, python tools timestamp always (timestamp setup in
logging_config.py) but all the surrounding bash does not.
We have extra timestamps added in run_functests.sh for our own
purposes to get the bash timestamps; but this ends up giving us
double-timestamps for the python bits. Additionally, callers such as
nodepool capture our output and put their own timestamps on it, and
again have the double-timestamps.
This uses a lightly modified outfilter.py from devstack to standardise
this.
All output is run through this filter, which will timestamp it. I
have removed the places where we double-timestamp -- logging_config.py
and the prefix in dib-run-parts.
An env option is added to turn timestamps off completely (does not
seem worth taking up a command-line option for). For callers like
nodepool, they can set this and will just have their own timestamps as
they collect the lines.
Since all logging is going through outfilter, it's easy to add a
--logfile option. I think this will be quite handy; personally I'm
always redirecting dib runs to files for debugging.
I've also added a "quiet" option. I think this could be useful in
run_tests.sh if we were to start logging the output of each test to
individual files. This would be much easier to deal with than the
very large log files we get (especially if we wanted to turn on
parallel running...)
Change-Id: I202e1cb200bde17f6d7770cf1e2710bbf4cca64c
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
Instead, either use the bash built-in of type to ensure it exists. Since
which is an external dep, things can fail oddly in a constrained
environment.
Also add a dib-lint test for this.
Change-Id: I645029f5b5bfe1198c89ce10fd3246be8636e8af
Signed-off-by: Jesse Keating <omgjlk@us.ibm.com>
With new block device definition, where content of the image
can be mounted on different partitions, is not enough with
executing setfiles on root directory. Instead of that, expose
all the mountpoints on the image, and apply setfiles on them.
Change-Id: I153f979722eaec49eab93d7cd398c5589b9bfc44
This patch finalizes the block device refactoring. It moves the three
remaining levels (filesystem creation, mount and fstab handling) into
the new python module.
Now it is possible to use any number of disk images, any number of
partitions and used them mounted to different directories.
Notes:
* unmount_dir : modified to only unmount the subdirs mounted by
mount_proc_sys_dev(). dib-block-device unmounts
$TMP_MOUNT_PATH/mnt (see I85e01f3898d3c043071de5fad82307cb091a64a9)
Change-Id: I592c0b1329409307197460cfa8fd69798013f1f8
Signed-off-by: Andreas Florath <andreas@florath.net>
Closes-Bug: #1664924
Move argument parsing to subparsers, rather than positional arguments.
This better reflects the tool's role as a driver and allows
sub-commands to deal with arguments in a natural way.
Change-Id: Iae8c368e0f3fe47abfddb9e0a1558bd5b3423aee
DIB_BLOCK_DEVICE_PARAMS_YAML should be exported, and the
dib-block-device will take this as the value of --params. Remove this
to simplify the command-line
Change-Id: I6764ed223ecd36f9d24e19f164b6a927380b410f
As part of the final steps, refactor the bits belonging
to block device and functions. This is a partial refactor
from I3600c6a3d663c697b59d91bd3fbb5e408af345e4
Change-Id: I7aa4fe0466e44846d8fa3194575d446fe4b5b2e6
Co-Authored-By: Andreas Florath <andreas@florath.net>
It seems that the redhat nodepool job is quite reliably geting a
"floating point" error during centos image build. This happens after
03-yum-cleanup which is pruning the locales. This might be a
red-herring, since the logs are full of
/bin/bash: warning: setlocale: LC_ALL: cannot change locale (C.UTF-8)
I think in our recent de-puppetisation of hosts, something might have
changed that is setting LC_ALL=C.UTF-8 for the jenkins user, at least
on Ubuntu. This is a problem for centos, as it doesn't have C.UTF-8
locale. I then think using the invalid locale is what leads the the
floating-point error when doing some maths in dib-run-parts to
calculate runtimes.
We are currently overriding LANG, but we really want LC_ALL to ensure
this applies globally.
Change-Id: I8e7cae093c4b32e0d20b73ae0086f14c7cc6a9cb
Add a new getval call that allows to retrieve values
from the block device. Also isolating the block device
information into a 'blockdev' dictionary entry, to better
return it with the getval command.
This is a refactor from the original code at
I3600c6a3d663c697b59d91bd3fbb5e408af345e4.
Change-Id: I93d33669a3a0ae644eab9f9b955bb5a9470cadeb
Co-Authored-By: Andreas Florath <andreas@florath.net>
The original approach was to pass each and every command
line parameter to the block device. While the block device
functionality gets extended, this is not any longer practical.
Instead of passing in all the parameters separately this patch
collects these in a YAML file that is passed in to the block device
layer.
Change-Id: I9d07593a01441b62632234468ac25a982cf1a9f0
Signed-off-by: Andreas Florath <andreas@florath.net>
This change move "do_extra_package_install" from pre-install to install
phase.
Extra packages are added by user request using the flag "-p", This
package should not be something the elements depend on.
The reason behind this patch is to move the extra package install to
a proper phase, Also more reasonable if base element run package update
to be before we install extra packages.
Change-Id: I68cc773aba9aa01743f0dda9f4e635e4cac2a282
DIB_PYTHON_EXEC was added in I5fab0e192c3a2dad8f60e821c184479e24e33bcd
to export the python that disk-image-create is running under. If dib
is installed with python3 then just calls "python" (python2) to run
sub-scripts, it fails to find itself.
This has been tested in devstack gates that use py3x
Change-Id: Ia1028972bfc0517b468b279aab9decdbcd7424ca
Now that the main partitioning refactor patch is merged, there is
a small relict of handling partitions still in the disk-image-create
main.
This patch moves the functionality from disk-image-create to the
block-device/partitioning module: it is mostly a rewrite of the
original bash code in python.
Change-Id: Ia73baeca74180a7bc9ea487da03ff56d6a3070ce
Signed-off-by: Andreas Florath <andreas@florath.net>
During the creation of a disk image (e.g. for a VM), there is the need
to create, setup, configure and afterwards detach some kind of storage
where the newly installed OS can be copied to or directly installed
in.
This patch implements partitioning handling.
Change-Id: I0ca6a4ae3a2684d473b44e5f332ee4225ee30f8c
Signed-off-by: Andreas Florath <andreas@florath.net>
We do not have the concept of "not installed" in v2, so remove the
obsolete code looking for a now non-existent variable.
Also, log the version at startup. This can help when
debugging from logs
Change-Id: I964c4cf207c10666afc5bc7ab9f2bfb9b1897c1e
Remove the x bit from lib/disk-image-create; because it's called
directly by the entry-point, it doesn't need to be exectuable.
This should also be clearer that you're not supposed to run it
by hand.
Remove some boilerplate from old file
Change-Id: Ibb6cdae613e6c9cf21dd6aecc8e1f739bc3a2643
It has always been a weird thing that dib is a python package, but
is totally driven by the disk-image-create script. It creates this
strange division that is hard to explain.
This moves disk-image-create to a regular python entry-point
Currently, this simply exec()s the original disk-image-create script.
However, we now have a (private) interface between disk-image-create
written in python and the driver shell script. Here's some things we
could do, for example:
* Argument parsing is generally nicer in Python, and then end result
is mostly just setting environment variables to flag different things
in the shell script. I could see us moving the argument-parsing into
diskimage_builder.disk_image_create:main() and just setting things in
os.environ before the exec()).
* I7092e1845942f249175933d67ab121188f3511fd sets IMAGE_ELEMENT_YAML in
disk-image-create by calling-back to element-info. We can just call
element_dependencies.find_all_elements() in here an export is to
os.environ before disk-image-create starts.
* remove need for ramdisk-image-create symlink by just exporting
IS_RAMDISK based on sys.argv[1] value
* you could even unit test some of this :)
Change-Id: I69ca3d26fede0506a6353c077c69f735c8d84d28
