Currently we have all our elements and library files in a top-level
directory and install them into
<root>/share/diskimage-builder/[elements|lib] (where root is either /
or the root of a virtualenv).
The problem with this is that editable/development installs (pip -e)
do *not* install data_files. Thus we have no canonical location to
look for elements -- leading to the various odd things we do such as a
whole bunch of guessing at the top of disk-image-create and having a
special test-loader in tests/test_elements.py so we can run python
unit tests on those elements that have it.
data_files is really the wrong thing to use for what are essentially
assets of the program. data_files install works well for things like
config-files, init.d files or dropping documentation files.
By moving the elements under the diskimage_builder package, we always
know where they are relative to where we import from. In fact,
pkg_resources has an api for this which we wrap in the new
diskimage_builder/paths.py helper [1].
We use this helper to find the correct path in the couple of places we
need to find the base-elements dir, and for the paths to import the
library shell functions.
Elements such as svc-map and pkg-map include python unit-tests, which
we do not need tests/test_elements.py to special-case load any more.
They just get found automatically by the normal subunit loader.
I have a follow-on change (I69ca3d26fede0506a6353c077c69f735c8d84d28)
to move disk-image-create to a regular python entry-point.
Unfortunately, this has to move to work with setuptools. You'd think
a symlink under diskimage_builder/[elements|lib] would work, but it
doesn't.
[1] this API handles stuff like getting files out of .zip archive
modules, which we don't do. Essentially for us it's returning
__file__.
Change-Id: I5e3e3c97f385b1a4ff2031a161a55b231895df5b
In shade, we use both md5 and sha256 checksums to help validate the
integrity of an image. Rather then having nodepool do this each time
for every time, have diskimage-builder create these files when we
build the image.
We've added a flag (disabled by default) to toggle this functionality.
Change-Id: I5815ba69b7d477f1e91dc8ec0c69c86168770964
Signed-off-by: Paul Belanger <pabelanger@redhat.com>
Block device handling can be somewhat complex - especially
when taking things like md, lvm or encryption into account.
This patch factors out the creation and deletion of the local
loop image device handling into a python library.
The main propose of this patch is to implement the needed
infrastructure. Based on this, more advanced functions can be added.
Example: (advanced) partitioning, LVM, handling different boot
scenarios (BIOS, UEFI, ...), possibility of handling multiple images
(local loop image, iSCSI, physical hard disk, ...), handling of
different filesystems for different partitions / LVs.
Change-Id: Ib626b36a00f8a5dc3dbde8df3e2619a2438eaaf1
Signed-off-by: Andreas Florath <andreas@florath.net>
Storing the du output in a variable seemed convenient, but I didn't
realise just how big it could get especially with things like infra
images -- there's something like 100MiB of text being stored in a bash
variable here.
Convert this to work with a temporary file
Change-Id: I6a6d22c2142e0f199490c39cca8c94769e4b0232
Since the ironic-agent element builds the ramdisk and extracts the
kernel itself, there's no need to actually generate an image at the
end of the process. Previously the unnecessary image was being
deleted, but this wastes a bunch of time compressing and converting
the image. It's better to just not create the image at all.
This change adds a noop element called no-final-image that
disk-image-create looks for in the element list and, if found, will
cause it to skip the final image generation. This is more flexible
than the previous ironic-agent-specific method that would have
required changes to disk-image-create for every element that wanted
to behave similarly.
Note that this cannot be done using an environment variable, because
element environments.d entries do not propagate out to
disk-image-create. It also doesn't make sense as a user option
because it should be set by the element author, not the user.
Change-Id: I168feb18f0d578b3babbe4784d3ef75e755e1ebd
Under some systems this leads to an error if the oder of parameters
does not comply exactly with the way it is specified.
Change-Id: Ie1ff871dfffecaf95e7ac467b18543561aaa0ceb
The ironic-agent element doesn't care about the final qcow2/raw/
whatever image the disk-image-create command normally creates, so
previously it was deleting it at the end of the process. This is
a pretty significant waste of time when building those images, and
instead we can just skip creating the image when building
ironic-agent.
Change-Id: If48f575e795a823c777891f193ebf8bd943aa296
For something fairly simple, I went back-and-forward with this a bit.
Firstly, I realise calling readlink constantly sucks. Due to the way
we call dib and source various files, you end up with the source-file
from "caller" being usually a very ugly path including levels of "../"
indirection. Cleaning this up to something canonical is the only sane
way to present it.
Because we evaluate _ps4() from a sub-shell in the PS4 string, there's
no way for it to do something like build a global in-memory cache in
an associative array or similar. It could write out a temp file or
some other side-band method, but the overheads of managing this don't
seem any different to just calling readlink. If anyone can think of a
bash-hack around this that doesn't involve a fork() I'm interested.
We could potentially strip some of the leading paths in the assumption
you know what they are; but it gets complex when things are split
across /usr/bin & /usr/lib and external elements, etc. I thought
about arbitrarily shortening it (e.g. just take last 20 characters)
which gives you enough of an idea of the file, but looks a bit ugly.
Or we could just leave the file-name out all together and assume the
function name is unique enough; this also seemed a bit ugly.
Obviously it's a matter of taste in the output. It is certainly
wider, but it also adds a lot of information. It also makes it fairly
clear where there are things we can make less verbose,
e.g. I1e39822f218dc0322e2490a770f3dc867a55802c disables tracing in
run-parts which is just noise. There's a few other frequently used
loops that we could disable tracing for by default to benefit
signal:noise.
tl;dr : take a look at the logs. I think it is a step in the right
direction of making the logs more usable for debugging.
Change-Id: I8054a3050415fcb527baeb7012bf133e5c864bf3
We have some test cases which attempt to build docker images, therefore
we need docker.
Fix a few bugs that showed up when we run docker tests - we need to
docker rm with sudo and docker images don't always have a /tmp so check
before unmounting it.
Change-Id: I147d0ef3f2ea83f35bac568214573a6bde0b1967
This was not well tested. Build the argument into a variable which
can be eval()ed to produce the final output.
Add the flag so we test this during functional tests. Add "-x" to dib
invocations so we can more easily debug failures.
Change-Id: Ifdc82627c520379b4124ccb9a4c2fe806c52c75c
We don't want the output of "du" run on the image spammed into the
logs with "set -x". Swizzle it off around the sensitive commands.
Change-Id: I687e77275f9a49e7934211835aba8610e88cdca6
If you check logs like [1] it's literally thousands of lines of the
same thing over-and-over as the git caching happens. It is basically
all just noise unless you're debugging it specifically. Up this to
tracing level 2 ("-x -x") to see it. Add a note in the help about
multiple flags, which has always been intended but not documented.
Image builds should continue to run with single "-x", but we could
probably greatly increase signal:noise ratio in the logs with a little
more judicial use of this to turn down some of the very noisy &
repetitive parts.
[1] anything in http://nodepool.openstack.org/
Change-Id: I91c5e55814ba9840769357261d203f4850e2eba6
This cuts the image size down alot, esspecially if there were lots of
small file deletes.
The fstrim utility is in the util-linux package and should be on
most all systems. fstrim also works with XFS, ext4, btrfs, etc
prodiving the kernel is new enough.
A reduction of 25% or more in size is common.
Change-Id: I269b4416be450369616f9b8e030f84c30e329804
In the common case of not specifying a size, we are already running
"du" over the image to figure out how big it is. Leverage that by
saving it's output and displaying a pruned list of big files when
requested.
We add a flag to show a summarised option (files >10MiB) and another
to show full output, should you wish that level of detail.
"Invocation" documentation is updated (and formatted a little better
while we're here).
Change-Id: I255800790a62fed1c82fcd311f1cc29c9867766d
Being able to discover DIB's version from the command itself is
convenient. This patch adds a --version option to the disk-image-create
command, failing gracefully if diskimage-builder is not installed.
This adds an explicit dependency on pbr to the requirements since this
is required to run diskimage_builder/version.py outside of a test
environment.
This patch consciously chooses to only provide the long-form option
and no '-v' to allow for the future possibility that a '-v' might
indicate '--verbose' in the future.
Change-Id: I9fc084774d6c7a39a944b07680b3eb8be8e34f9c
This checks the profile, if it has hardened in it's name it needs xattr support
unfortunately xattr support cannot yet be relied on everywhere, so it needs to
be disabled for hardened profile builds to correctly pax-mark.
Change-Id: I7fb855249a9e6c9b6497ab5061b4ea3c014f5081
Closes-Bug: 1537177
This patch fixes the calculation of the resultant image size
when building an image with diskimage-builder on ext4 a
filesystem.
Prior to this, using the '--image-size 2' (2GB) setting would
generate an image that would not boot under a 2GB nova flavor.
Change-Id: I7a753bdef84c6300ccea73ae4a92bf330dcd77cb
Closes-Bug: #1513622
Patch adds support for using decimal values for $DIB_IMAGE_SIZE.
This allows for creating images that are <1G in size.
Change-Id: I945644a8e77fecfb0b83efa282dc00bb29514e0b
Closes-Bug: #1366909
As described in the comments, inspect the installation to see if we
have been installed with "pip -e" and, if so, make sure we reference
the scripts from the source location rather than the
system-installations.
Update the documentation with a terse but helpful quick-start to show
an easy way to start developing a change using this.
Closes-Bug: #1491035
Change-Id: I0460061b834a2b854175f8c9be2be8d38c540c9d
App containers are a format used by rocket and are specified at the
following url:
https://github.com/appc/spec/
Change-Id: I8ac24f0194c4bf53dffd6c47e0587bc413101698
We can already produce tarballs, which is the input format docker import
expects. This makes it trivial to add docker as an output format.
Co-Authored-By: Dan Prince <dprince@redhat.com>
Change-Id: Ib60db3b717d33d4cf3181d70fe0ffbfa86fd5d02
Split the cleanup_dirs function in two, i.e. cleanup of the build dir
and the image dir, and use the former to cleanup the temporary build
subdirs after their unmount, before the conversion to other disk
formats; they are not needed anyway at that point, and allows to save
disk space during the conversion phase.
Change-Id: Ie30d7e6033613d6979148423326ae7e17a7342e7
$DIB_DEFAULT_INSTALLTYPE is propogated into the chroot by writing out an
export statement to a file under environment.d. However, that export
statement does not account for other elements that may choose to define
$DIB_DEFAULT_INSTALLTYPE earlier on in their own environment.d files.
Change-Id: I5865b1859c080c328251e8dce072c0634ef75c16
We do not have any testing inside DIB for testing disk-image-create
logic. Lets do some smoke testing for all our supported image formats.
Also adding a run_functests.sh so we can extend this later without editing
the jenkins job.
Change-Id: Ie491e27f00bde54f73af6b47c9696ec04d973b14
Our logic to determine needed image size doesnt fully account for the
in-filesystem journal. This only shows up when creating images that are
very small relative to the FS journal size.
Change-Id: Ic3c2bcd31ec4fee6bcd9f67767842eb3fbe20d3a
At present, MKFS_OPTS is closed for modification. The ability
to extend the set of MKFS_OPTS adds a great deal of power for
knowledgeable end-users. (And in some specific circumstances,
it is vital to success, as in the case of building RHEL/CentOS
6 images from RHEL/CentOS 7 hosts, in which case -O ^64bit is
required in order for the image to boot.)
Change-Id: I714e86a5a413779e63f598fbbb5a79d23cf6d8c3
We currently use qemu-img convert with a raw source and dest when
building raw images. We can just mv the file for increased speed.
Change-Id: I3da095cb9ecad7224a121a434a9fb204132bf6df
There is a wide variety of tracing options through the various shell
scripts. Some use "set -eux", others explicity set xtrace and others
do nothing. There is a "-x" option to bin/disk-image-create but it
doesn't flow down to the many scripts it calls.
This adds a global integer variable set by disk-image-create
DIB_DEBUG_TRACE. All scripts have a stanza added to detect this and
turn on tracing. Any other tracing methods are rolled into this. So
the standard header is
---
if [ "${DIB_DEBUG_TRACE:-0}" -gt 0 ]; then
set -x
fi
set -eu
set -o pipefail
---
Multiple -x options can be specified to dib-create-image, which
increases the value of DIB_DEBUG_TRACE. If script authors feel their
script should only trace at higher levels, they should modify the
"-gt" value. If they feel it should trace by default, they can modify
the default value also.
Changes to pachset 16 : scripts which currently trace themselves by
default have retained this behaviour with DIB_DEBUG_TRACE defaulting
to "1". This was done by running [1] on patch set 15. See the thread
beginning at [2]
dib-lint is also updated to look for the variable being matched.
[1] https://gist.github.com/ianw/71bbda9e6acc74ccd0fd
[2] http://lists.openstack.org/pipermail/openstack-dev/2014-November/051575.html
Change-Id: I6c5a962260741dcf6f89da9a33b96372a719b7b0
When building the ramdisk we don't cleanup the temporary
directories after ourselves. This leaves /tmp/image.* directories
mounted and /tmp/image.* directories on the system.
Also the ramdisk-functions duplicate, from what I can see,
the cleanup function from common-functions. So when a job
is killed off it ends up leaving /tmp/image.* directories
on the system.
Change-Id: I2d73aabd0eb176027b4e7368580db08902e2b6ab
Unlike most of the options to disk-image-create, this one didn't
document its default value, and it also didn't have any context for
why one might use it.
Change-Id: I57ea529cf387ab8ec8497d35be19ac64198df47a
Related-Bug: 1397073
Avoid a warning when multiple arguments are passed to disk-image-create
like:
bin/disk-image-create: 146: [: aa: unexpected operator
Change-Id: I619d7bf2e4b56885d629019b80b249bc80b8f43e
Instead of charging ahead and failing because we lack dib-run-parts we
can fail in this case with a useful error message.
Change-Id: Ie10176b5a7019a0385796984ebbaf7742876925a
This element creates kernel and ramdisk files based on Fedora,
example:
disk-image-create -a i386 -o test fedora ironic-agent disable-selinux
Change-Id: Ifa133d1680b81cb87d32a405aa7d7b40fe91f835
Adds new disk-image-create --install-type option which
can be used to enable alternate install types. The
default install type is 'source'.
The motivation is to eliminate elements like
enable-package-installs which require coupling and also
don't work with elements in multiple element repositories.
This patch does not prevent you from using the previous
DIB_INSTALLTYPE_ variables to customize the install type,
rather it just changes the default so you don't have to
set it so often when using things like packages.
Change-Id: Icee98440fc2251728f2dca30e7c4789a0fd89b93
