createhdds.py

createhdds.py creates and maintains the set of pre-rolled hard disk images needed for some of the Fedora openQA tests.

Requirements

createhdds requires libvirt-daemon-kvm, libvirt-python3, python3-libguestfs, python3-fedfind, qemu-kvm, virt-install, and (for ppc64 images) powerpc-utils. You will need somewhere around 100-150GB of disk space for the full set of base images for all supported arches.

Usage

Most usage information can be seen in the help text - just run createhdds.py -h for an overview of the subcommands available, and createhdds.py (subcommand) -h for help on a subcommand. To put it simply, the most common usage is simply to run createhdds.py all -c. This will create all the currently-expected images (that are arch-compatible with the host you are running createhdds on) that have not already been created, and recreate any that need recreating (images can have a 'maximum age' causing them to be rebuilt by all when they're older than that age, and images also have a 'version' - if the image's 'version' is bumped by the maintainers, all will rebuild it). It will also remove any image files that are present that aren't expected to be present - usually images for old releases that are no longer tested, or images we've simply stopped using. In a typical deployment of a Fedora openQA instance, the admin should set things up so the git checkout is updated and createhdds.py all -c is run regularly - say, once a day (and probably not while tests are being run).

createhdds.py check will just check whether all expected images are present and up-to-date. If all images are present but some are outdated, it will exit 1. If some images are entirely missing, it will exit 2. This can be handy for use with things like Ansible (so you can run the check to decide whether you need to run the creation, and thus avoid spurious 'changed' statuses).

There are also individual subcommands for each of the named 'image groups', allowing you to create just the image(s) from that group. For image groups which usually generate multiple images, the subcommand will have arguments that let you restrict creation to just a subset of those images (and for virt-install type images, you can create the image(s) for a different release than would usually be the case, too).

In all mode, and in single-image mode if you do not pass --release, createhdds can decide what releases to build images for, for those image groups that include an installed Fedora release (the virt-install type images). A virt-install type image group can specify the releases to build images for absolutely (by giving the release numbers as positive integers), or relative to the next pending release (by giving the release numbers as negative integers). When it encounters one of these 'relative' release numbers, createhdds uses fedfind to discover the 'current' release, and adds 1 to that (to find the 'pending' release). Just in case anything goes wrong with this, or you need to override it for some reason, the --nextrel argument is available for relevant subcommands to explicitly specify the 'next release'.

Specifying images / 'image groups': hdds.json and .commands files

All the information on what images can/should be created comes from the hdds.json file and some virt-install commands files. You can add, modify and remove image definitions without touching createhdds.py. hdds.json should define a single dictionary with three keys: guestfs, virtinstall, and renames. The meat is guestfs and virtinstall, which define 'image groups': for each image group, createhdds all will create one or more images (multiple images produced from a single group are referred to as 'variants'). Groups and variants are provided for so that if you want to create, say, three images that are identical but for their disk label, you don't have to create a whole new almost-identical entry for each one. The rules about what particular attributes of an image can be implemented as 'variants' are somewhat arbitrary and actually just taken from the old createhdds.sh; each function in that implementation became an 'image group' in this rewrite, and the attributes that can vary between variants are the same ones that could be set as function arguments in createhdds.sh. The .ks files allow for customization of virt-install images; more on this later. hdds.json and the .ks files must always be in the same folder as createhdds.py (not necessarily the same folder the disk images reside in).

renames

The value of renames is a simple list-of-lists. Each item is a pair of strings. In all mode, and optionally in check mode, createhdds will read in all the items from the renames list. For each item it will look for a file (in the working directory - createhdds always works on the working directory) named the same as the first string, and if it finds such a file, change its name to the second string. So for instance, the value of renames could be [['disk_foo.img', 'disk_bar.img'], ['disk_monkey.img', 'disk_fish.img']]; that would result in createhdds renaming disk_foo.img to disk_bar.img and disk_monkey.img to disk_fish.img. This mechanism is provided to aid in the situation where an image's expected name changes, but the existing image file is still valid; instead of forcing the user to rename it manually or re-generate it, we can list it in renames and it will get renamed automatically. Of course, if the image's content changes too, we shouldn't use this mechanism.

guestfs and virtinstall

The value for both guestfs and virtinstall is a list of dicts. Each dict defines a single 'image group'; an image group can produce just a single image, or several variants - we will learn what determines the possible variants for an image group below. The images in guestfs are produced using libguestfs - these are images that just contain a particular partition layout and perhaps some files we seed directly, but no installed operating system. The images in virtinstall are produced using virt-install - these are images containing an installed operating system. Some keys are common to both image group types:

imgver

This is the 'image version'. It is optional for both image types. By convention, it should be an integer digit string, but its practical effect is simply to be included in the image filename(s), so it can be any string valid in a filename. If omitted or set to the empty string, no imgver component will be included in the filename. This means that by changing the version you can change the expected name - which will cause the check mode to report the old file as 'unknown' and the new file as 'missing', and will cause the all mode to build the new file. Thus if you're maintaining the image set, and you make some change to an image group which would mean that existing image files for that group can no longer be used, you can change imgver to cause the images to be rebuilt.

name

The image group's name (a string). Required for both image types. This is included in the image file names, of course, and it will also be the subcommand to create image(s) from this group.

size

This key is required for guestfs, optional for virtinstall. It is the size of the image. A plain digit string is a size in bytes. A digit string followed by 'M', 'MB' or 'MiB' is a size in megabytes (power-of-2). A digit string followed by 'G', 'GB' or 'GiB' is a size in gigabytes (power-of-2). For virtinstall, if size is not set, the image will be whatever size the virt-install base image it's created from is.

Some keys are specific to each type of image. These are the guestfs-specific keys:

parts

This key is required. This key's value is a list of dicts. Each dict represents a single partition that should be created on the disk. Required keys are:

  • type - the partition type: 'p' for primary, 'l' for logical, 'e' for extended
  • start - start sector
  • end - end sector

These values are just passed straight to libguestfs, so you can find further info on them in the libguestfs documentation, especially on various special values for start and end (negative values are relative to the end of the disk, for e.g.)

Optional keys are:

  • filesystem - if set, this partition will always have this filesystem. If not set, the filesystem will be determined according to the image group's filesystems value (see below).
  • label - if set, this partition will have this label

writes

This key is optional. Its value is a list of dicts. Each dict represents a single file that should be created on one of the partitions in the image. There are exactly three required keys for the dict:

  • part - the number of the partition the file should be written to, starting at 1
  • path - the path where the file should be written (root is the root of the partition it's being written to)
  • content - the actual content to write to the file (a string)

uploads

This key is optional. Its value is a list of dicts. Each dict represents a single file that should be retrieved from a web site and copied to a partition in the image. There are exactly three required keys for the dict:

  • part - the number of the partition the file should be written to, starting at 1
  • target - the path where the file should be written (root is the root of the partition it's being written to)
  • source - the URL of the file to download

There's currently no provision for uploading a local file, or any protocol besides http/https (you can use either).

labels and filesystems

These keys are optional. Each one's value is a list of strings. These keys together determine how many image variants are expected to be produced from the image group. If not set, the default value of labels is ['mbr'], and the default value of filesystems is ['ext4'] (both single-item lists). For each guestfs image group, the expected images will be the combinations of labels and filesystems. This means that if you don't set either key, or you set either key to a single item list, only a single image will be expected. If you set labels to a two-item list and filesystems to a single-item list, two images will be expected. If you set both keys to a two-item list, four images will be expected...and so on.

When multiple combinations are in play, the names of the images will include the relevant values. So if an image group has multiple entries in the labels list but not the filesystems list, the filenames will be disk_(name)_(label1).img, disk_(name)_(label2).img, and so on. If there are multiple entries in both lists, you'll get disk_(name)_(filesystem1)_(label1).img, disk_(name)_(filesystem1)_(label2).img, etc etc. The all subcommand will always create all the expected images; the image group subcommand will create all the expected images by default, but will have --label and --filesystem arguments allowing the user to restrict creation to a single item from either or both lists.

For labels, the values represent disk label types, and are passed to guestfs. The only values used at present are gpt and mbr. Obviously, gpt formats the disk with a GPT disk label, mbr formats the disk with an MBR label.

For filesystems, the values represent...filesystems. Any of the partitions defined in parts (see above) which does not specify a filesystem will be formatted with this filesystem.

Let's consider some examples!

Say an image group specifies name : "blank", labels : ["mbr", "gpt"] and does not specify filesystems. There will be two expected images: disk_blank_mbr.img and disk_blank_gpt.img. The former will have an MBR disk label, the latter a GPT disk label. In all other respects, the images will be identical. If any partition does not specify a filesystem, it will be formatted as ext4 (as the default for filesystems is ['ext4']).

Say an image group specifies name : "blank", filesystems : ["ext4", "ntfs"] and does not specify labels. There will be two expected images: disk_blank_ext4.img and disk_blank_ntfs.img. Both will have an MBR disk label (as the default for labels is ['mbr']). On the former, all partitions which do not specify a filesystem will be formatted as ext4; on the latter, all partitions which do not specify a filesystem will be formatted as ntfs. In all other respects the images will be identical. As a special note: it is nonsense to specify multiple filesystems, but also explicitly specify a filesystem for each partition in parts. This will result in the creation of multiple identical images, because none of the values from filesystems will actually do anything. However, it's perfectly reasonable to have an image group with some partitions that explicitly specify a filesystem and some that do not, and then have multiple filesystem variants - say, you want multiple variant images with the data partitions formatted using different filesystems, but you want the /boot partition in each variant to be ext4.

Finally, say an image group specifies name : "blank", filesystems : ["ext4", "ntfs"] and labels : ["mbr", "gpt"]. There will be four expected images, disk_blank_ext4_mbr.img, disk_blank_ntfs_mbr.img, disk_blank_ext4_gpt.img, disk_blank_ntfs_gpt.img. The mbr images will have MBR disk labels, the gpt images will have GPT disk labels, the ext4 images will have partitions that don't explicitly specify a filesystem formatted as ext4, and the ntfs images will have partitions that don't explicitly specify a filesystem formatted as NTFS.

Whew! That was a lot of explanation, but it's not really a super-complicated concept, you'll get it easy. OK, let's move on to the virtinstall-specific keys:

releases

This key is required. It defines the releases and arches for which images are expected; thus it determines the number of images that will be expected for this group. The value is a dict. Each key in the dict represents a release; the value for each key is a list of the arches for which images should be built for that release. The keys should be integer digit strings. Positive values indicate absolute release numbers. Negative values are relative to whatever is the pending release at the time the images are created. So a release number -1 means 'the release one before the pending release at the time the images are built'. So if the next Fedora release will be Fedora 24 at the time the images are created, and one of the dict keys is -1, an image will be expected for Fedora 23.

The filename for a virt-install type image always includes the release number and arch it's built for - disk_f(release)_(name)_(arch).img.

Let's look at an example! Say the name is minimal and the releases dict is { "-1" : ["i686", "x86_64"], "-2" : ["x86_64"] }. Three images will be expected, and the expected releases will be relevant to the pending release. Say the pending release is Fedora 24, the expected images will be disk_f23_minimal_i686.img (Fedora 23 for i686), disk_f23_minimal_x86_64.img (Fedora 23 for x86_64), and disk_f22_minimal_x86_64.img (Fedora 22 for x86_64). When time moves on and the next pending release is F25, images will be expected for Fedora 23 and Fedora 24, and the Fedora 22 images will be considered obsolete and deleted by cleanup modes of createhdds.

As with the guestfs case, the single image group subcommand will have parameters to limit creation. So in our example, the minimal subcommand will have --release and --arch parameters, each allowing just a single value. For coding simplicity, passing --arch alone is ignored (this may be fixed later) and will just result in the 'expected' images being created. If --release is passed, only a single image will be created, for whatever release is specified; by default it will be the x86_64 image, you may pass --arch (arch) to build another arch instead.

Note that createhdds attempts to only create images that are arch-compatible when run - ones where the host CPU can emulate the target arch at near-native speed. In other words, it will not attempt to build ppc64(le) images on x86_64 hosts, or aarch64 images on ppc64(le) hosts, etc. If you intend to have an openQA deployment with multiple arches that are not compatible in this way, the intent is that you ensure at least one machine of each arch group has write access to the necessary location, and run createhdds on each of those machines, so the full set of images will be created. Note this also means createhdds will ignore images for other arches entirely - it will not delete them because it thinks they're out of date or unknown, for instance. You do not need images for all arches, only for the arches you actually intend to have workers for and run tests on.

maxage

This key is optional. If not set, it defaults to 14. Its value should be an integer string. This basically indicates how often (as a number of days) the image should be rebuilt. virt-install images can go 'stale' - at build time we update the installed OS to the latest packages, but of course by the time the image is used, later updates may be available. If the test the image is used for needs all the latest packages to be installed, the test will have to install the later updates, and it's inefficient to have one or more tests doing that every day. So it makes sense to re-generate the images periodically so that the tests only have to install few if any updates. For any image group with a non-0 maxage, createhdds check and any modes will check any existing image file's age against maxage; if it exceeds the maxage check will consider it 'outdated', and all will rebuild it. To disable maxage checks for an image group, set maxage to 0.

ks files

Customization of virt-install type images, beyond the Fedora release/arch combination, is done with .ks files. These are installer kickstart files. These are how we actually define what packages to install and so on for each different image group. The logic is simple: for each virt-install image group, if there is a file named (name).ks, that file will be passed to virt-install as the install kickstart. For instance, the desktop.ks file contains the install directives for the 'desktop' virt-install image group; it installs the Workstation package group, creates a regular user, and does a few other things. The kickstart documentation explains all the possible directives.