diskimage-builder/doc/source/user_guide/building_an_image.rst
Ian Wienand 00da1982ce Add a more generic tree->graph parser
This moves to a more generic config parser that doesn't have plugins
parsing part of the tree.

I understand why it ended up that way; we have "partitions" key which
has special semantics compared to others keys and there was a desire
to keep it isolated from core tree->graph code.  But this isn't really
isolated; you have to reverse-engineer several module-crossing
boundaries, extras classes and repetitive recursive functions.

Ultimately, plugins should have access to the node graph, but not
participate in configuration parsing.  This way we ensure that plugins
can't invent new methods of configuration parsing.

Note: unit tests produce the same tree -> graph conversion as the old
method.  i.e. this is not intended to have a functional change.

Change-Id: I8a5d62a076a5a50597f2f1df3a8615afba6dadb2
2017-05-26 10:13:14 +10:00

516 lines
15 KiB
ReStructuredText

Building An Image
=================
Now that you have diskimage-builder properly :doc:`installed <installation>`
you can get started by building your first disk image.
VM Image
--------
Our first image is going to be a bootable vm image using one of the standard
supported distribution :doc:`elements <../elements>` (Ubuntu or Fedora).
The following command will start our image build (distro must be either
'ubuntu' or 'fedora'):
::
disk-image-create <distro> vm
This will create a qcow2 file 'image.qcow2' which can then be booted.
Elements
--------
It is important to note that we are passing in a list of
:doc:`elements <../elements>` to disk-image-create in our above command. Elements
are how we decide what goes into our image and what modifications will be
performed.
Some elements provide a root filesystem, such as the ubuntu or fedora element
in our example above, which other elements modify to create our image. At least
one of these 'distro elements' must be specified when performing an image
build. It's worth pointing out that there are many distro elements (you can even
create your own), and even multiples for some of the distros. This is because
there are often multiple ways to install a distro which are very different.
For example: One distro element might use a cloud image while another uses
a package installation tool to build a root filesystem for the same distro.
Other elements modify our image in some way. The 'vm' element in our example
above ensures that our image has a bootloader properly installed. This is only
needed for certain use cases and certain output formats and therefore it is
not performed by default.
Output Formats
--------------
By default a qcow2 image is created by the disk-image-create command. Other
output formats may be specified using the `-t <format>` argument. Multiple
output formats can also be specified by comma separation. The supported output
formats are:
* qcow2
* tar
* tgz
* squashfs
* vhd
* docker
* raw
Disk Image Layout
-----------------
The disk image layout (like number of images, partitions, LVM, disk
encryption) is something which should be set up during the initial
image build: it is mostly not possible to change these things later
on.
There are currently two defaults:
* When using the `vm` element a MBR based partition layout is created
with exactly one partition that fills up the whole disk and used as
root device.
* When not using the `vm` element a plain filesystem image, without
any partitioning, is created.
The user can overwrite the default handling by setting the environment
variable `DIB_BLOCK_DEVICE_CONFIG`. This variable must hold YAML
structured configuration data.
The default when using the `vm` element is:
.. code-block:: yaml
DIB_BLOCK_DEVICE_CONFIG='
- local_loop:
name: image0
- partitioning:
base: image0
label: mbr
partitions:
- name: root
flags: [ boot, primary ]
size: 100%
mkfs:
mount:
mount_point: /
fstab:
options: "defaults"
fsck-passno: 1'
The default when not using the `vm` element is:
.. code-block:: yaml
DIB_BLOCK_DEVICE_CONFIG='
- local_loop:
name: image0
mkfs:
name: mkfs_root
mount:
mount_point: /
fstab:
options: "defaults"
fsck-passno: 1'
There are a lot of different options for the different levels. The
following sections describe each level in detail.
General Remarks
+++++++++++++++
In general each module that depends on another module has a `base`
element that points to the depending base. Also each module has a
`name` that can be used to reference the module.
Tree-Like vs. Complete Digraph Configuration
++++++++++++++++++++++++++++++++++++++++++++
The configuration is specified as a digraph_. Each module is a
node; a edge is the relation of the current element to its `base`.
Because the general digraph_ approach is somewhat complex when it comes
to write it down, the configuration can also be given as a tree_.
.. _digraph: https://en.wikipedia.org/wiki/Directed_graph
.. _tree: https://en.wikipedia.org/wiki/Tree_(graph_theory)
Example: The tree like notation
.. code-block:: yaml
mkfs:
name: root_fs
base: root_part
mount:
mount_point: /
is exactly the same as writing
.. code-block:: yaml
mkfs:
name: root_fs
base: root_part
mount:
name: mount_root_fs
base: root_fs
mount_point: /
Non existing `name` and `base` entries in the tree notation are
automatically generated: the `name` is the name of the base module
prepended by the type-name of the module itself; the `base` element is
automatically set to the parent node in the tree.
In mostly all cases the much simpler tree notation can be used.
Nevertheless there are some use cases when the more general digraph
notation is needed. Example: when there is the need to combine two or
more modules into one new, like combining a couple of physical volumes
into one volume group.
Tree and digraph notations can be mixed as needed in a configuration.
Limitations
+++++++++++
There are a couple of new modules planned, but not yet implemented,
like LVM, MD, encryption, ...
To provide an interface towards the existing elements, there are
currently three fixed keys used - which are not configurable:
* `root-label`: this is the label of the block device that is mounted at
`/`.
* `image-block-partition`: if there is a block device with the name
`root` this is used else the block device with the name `image0` is
used.
* `image-path`: the path of the image that contains the root file
system is taken from the `image0`.
Level 0
+++++++
Module: Local Loop
..................
This module generates a local image file and uses the loop device to
create a block device from it. The symbolic name for this module is
`local_loop`.
Configuration options:
name
(mandatory) The name of the image. This is used as the name for the
image in the file system and also as a symbolic name to be able to
reference this image (e.g. to create a partition table on this
disk).
size
(optional) The size of the disk. The size can be expressed using
unit names like TiB (1024^4 bytes) or GB (1000^3 bytes).
Examples: 2.5GiB, 12KB.
If the size is not specified here, the size as given to
disk-image-create (--image-size) or the automatically computed size
is used.
directory
(optional) The directory where the image is created.
Example:
.. code-block:: yaml
local_loop:
name: image0
local_loop:
name: data_image
size: 7.5GiB
directory: /var/tmp
This creates two image files and uses the loop device to use them as
block devices. One image file called `image0` is created with
default size in the default temp directory. The second image has the
size of 7.5GiB and is created in the `/var/tmp` folder.
Level 1
+++++++
Module: Partitioning
....................
This module generates partitions on existing block devices. This
means that it is possible to take any kind of block device (e.g. LVM,
encrypted, ...) and create partition information in it.
The symbolic name for this module is `partitioning`.
Currently the only supported partitioning layout is Master Boot Record
`MBR`.
It is possible to create primary or logical partitions or a mix of
them. The numbering of the primary partitions will start at 1,
e.g. `/dev/vda1`; logical partitions will typically start
with `5`, e.g. `/dev/vda5` for the first partition, `/dev/vda6` for
the second and so on.
The number of logical partitions created by this module is theoretical
unlimited and it was tested with more than 1000 partitions inside one
block device. Nevertheless the Linux kernel and different tools (like
`parted`, `sfdisk`, `fdisk`) have some default maximum number of
partitions that they can handle. Please consult the documentation of
the appropriate software you plan to use and adapt the number of
partitions.
Partitions are created in the order they are configured. Primary
partitions - if needed - must be first in the list.
There are the following key / value pairs to define one partition
table:
base
(mandatory) The base device where to create the partitions in.
label
(mandatory) Possible values: 'mbr'
This uses the Master Boot Record (MBR) layout for the disk.
(There are currently plans to add GPT later on.)
align
(optional - default value '1MiB')
Set the alignment of the partition. This must be a multiple of the
block size (i.e. 512 bytes). The default of 1MiB (~ 2048 * 512
bytes blocks) is the default for modern systems and known to
perform well on a wide range of targets. For each partition
there might be some space that is not used - which is `align` - 512
bytes. For the default of 1MiB exactly 1048064 bytes (= 1 MiB -
512 byte) are not used in the partition itself. Please note that
if a boot loader should be written to the disk or partition,
there is a need for some space. E.g. grub needs 63 * 512 byte
blocks between the MBR and the start of the partition data; this
means when grub will be installed, the `align` must be set at least
to 64 * 512 byte = 32 KiB.
partitions
(mandatory) A list of dictionaries. Each dictionary describes one
partition.
The following key / value pairs can be given for each partition:
name
(mandatory) The name of the partition. With the help of this name,
the partition can later be referenced, e.g. when creating a
file system.
flags
(optional) List of flags for the partition. Default: empty.
Possible values:
boot
Sets the boot flag for the partition
primary
Partition should be a primary partition. If not set a logical
partition will be created.
size
(mandatory) The size of the partition. The size can either be an
absolute number using units like `10GiB` or `1.75TB` or relative
(percentage) numbers: in the later case the size is calculated
based on the remaining free space.
Example:
.. code-block:: yaml
- partitioning:
base: image0
label: mbr
partitions:
- name: part-01
flags: [ boot ]
size: 1GiB
- name: part-02
size: 100%
- partitioning:
base: data_image
label: mbr
partitions:
- name: data0
size: 33%
- name: data1
size: 50%
- name: data2
size: 100%
On the `image0` two partitions are created. The size of the first is
1GiB, the second uses the remaining free space. On the `data_image`
three partitions are created: all are about 1/3 of the disk size.
Level 2
+++++++
Module: Mkfs
............
This module creates file systems on the block device given as `base`.
The following key / value pairs can be given:
base
(mandatory) The name of the block device where the filesystem will
be created on.
name
(mandatory) The name of the partition. This can be used to
reference (e.g. mounting) the filesystem.
type
(mandatory) The type of the filesystem, like `ext4` or `xfs`.
label
(optional - defaults to the name)
The label of the filesystem. This can be used e.g. by grub or in
the fstab.
opts
(optional - defaults to empty list)
Options that will passed to the mkfs command.
uuid
(optional - no default / not used if not givem)
The UUID of the filesystem. Not all file systems might
support this. Currently there is support for `ext2`, `ext3`,
`ext4` and `xfs`.
Example:
.. code-block:: yaml
- mkfs:
name: mkfs_root
base: root
type: ext4
label: cloudimage-root
uuid: b733f302-0336-49c0-85f2-38ca109e8bdb
opts: "-i 16384"
Level 3
+++++++
Module: Mount
.............
This module mounts a filesystem. The options are:
base
(mandatory) The name of the filesystem that will be mounted.
name
(mandatory) The name of the mount point. This can be used for
reference the mount (e.g. creating the fstab).
mount_point
(mandatory) The mount point of the filesystem.
There is no need to list the mount points in the correct order: an
algorithm will automatically detect the mount order.
Example:
.. code-block:: yaml
- mount:
name: root_mnt
base: mkfs_root
mount_point: /
Level 4
+++++++
Module: fstab
.............
This module creates fstab entries. The following options exists. For
details please consult the fstab man page.
base
(mandatory) The name of the mount point that will be written to
fstab.
name
(mandatory) The name of the fstab entry. This can be used later on
as reference - and is currently unused.
options
(optional, defaults to `default`)
Special mount options can be given. This is used as the fourth
field in the fstab entry.
dump-freq
(optional, defaults to 0 - don't dump)
This is passed to dump to determine which filesystem should be
dumped. This is used as the fifth field in the fstab entry.
fsck-passno
(optional, defaults to 2)
Determines the order to run fsck. Please note that this should be
set to 1 for the root file system. This is used as the sixth field
in the fstab entry.
Example:
.. code-block:: yaml
- fstab:
name: var_log_fstab
base: var_log_mnt
options: nodev,nosuid
dump-freq: 2
Filesystem Caveat
-----------------
By default, disk-image-create uses a 4k byte-to-inode ratio when
creating the filesystem in the image. This allows large 'whole-system'
images to utilize several TB disks without exhausting inodes. In
contrast, when creating images intended for tenant instances, this
ratio consumes more disk space than an end-user would expect (e.g. a
50GB root disk has 47GB avail.). If the image is intended to run
within a tens to hundrededs of gigabyte disk, setting the
byte-to-inode ratio to the ext4 default of 16k will allow for more
usable space on the instance. The default can be overridden by passing
``--mkfs-options`` like this::
disk-image-create --mkfs-options '-i 16384' <distro> vm
You can also select a different filesystem by setting the ``FS_TYPE``
environment variable.
Note ``--mkfs-options`` are options passed to the mfks *driver*,
rather than ``mkfs`` itself (i.e. after the initial `-t` argument).
Speedups
--------
If you have 4GB of available physical RAM (as reported by /proc/meminfo
MemTotal), or more, diskimage-builder will create a tmpfs mount to build the
image in. This will improve image build time by building it in RAM.
By default, the tmpfs file system uses 50% of the available RAM.
Therefore, the RAM should be at least the double of the minimum tmpfs
size required.
For larger images, when no sufficient amount of RAM is available, tmpfs
can be disabled completely by passing --no-tmpfs to disk-image-create.
ramdisk-image-create builds a regular image and then within that image
creates ramdisk.
If tmpfs is not used, you will need enough room in /tmp to store two
uncompressed cloud images. If tmpfs is used, you would still need /tmp space
for one uncompressed cloud image and about 20% of that image for working files.