When support for osbuild result was added into osbuild-composer it was in
a bit hacky way - localtarget's location was reused as a path for the
result. This didn't make much sense because we want to store the result
even when image build has no localtarget.
Several past commits made store less dependant on the localtarget. The
responsibility for "holding the paths" to build artifacts was gradually
switched from the localtarget to the store while still maintaining
backwards compatibility - localtarget.Location still pointed at the
correct location.
This commit finishes the switch: local target now has no Location field.
The store is now fully responsible for managing the artifacts and paths
to them. LocalTarget is now just a simple "switch" - if image build has it,
then worker uploads an image into the store and it's then available for
download using the weldr API.
In #221 Compose was refactored: Now it can have multiple image builds. More
image builds result in more jobs. Each job has its own result (logs from
osbuild). Additionally, also targets are now a part of image build. With
local target this effectively means we can have multiple images per compose.
However, these artifacts (images & results) were stored only per compose
prior this commit, thus rendering the behaviour of composes with multiple
image builds undefined and racy.
This commit fixes it by storing all the artifacts per image build instead of
per compose. To achieve this feature, getComposeDirectory and
getImageBuildDirectory methods were created to centralize the path assembly.
Paths to artifacts prior this commit:
${COMPOSER_STATE_DIR}/outputs/${COMPOSE_ID}/*
Paths to artifacts after this commit:
${COMPOSER_STATE_DIR}/outputs/${COMPOSE_ID}/${IMAGE_BUILD_ID}/*
Everything that this field contained can be computed in another way:
- path: just lookup the local target and read the path from there
- mime: can be derived from distribution and compose output type
- size: can be derived from the path
Therefore it imho doesn't make much sense to store these information multiple
times.
All the preceding changes in jobqueue, compose package, common types
etc. were a base for refactoring the store so that a compose can handle
multiple image builds. This commit introduces the change in a backwards
compatible way, so that the weldr api don't change.
When a user does not define the image size for a compose the default
image size of that image type is used. In order to properly store the
compose's image size even if the default is used the store calls the
distro function GetSizeForOutputType. This function accepts an output
format and image size. If the image size is 0 then the default
value for the output format will be returned. Also, for vhd images the
size must be rounded. This is now handled in the distro function instead
of the api.
When a use defines the image size for a compose this size is stored in
the compose struct so that the virtual image size can be returned by the
api instead of the file size of the image.
In the future remote workers will be introduced. Obviously, the remote worker
cannot support the local target. Unfortunately, the current implementation of
storing the osbuild result is dependant on it.
This commit moves the responsibility of storing osbuild result to the
composer process instead of the worker process. The result is transferred from
a worker to a composer using extended HTTP API.
Generating a SHA-1 based on time is not safe. A collision can easily be
generated, and if parallel operations are used they will eventually
collide. This reads random bytes and uses them for the SHA-1 hash. It
will return an error if the rand.Read() fails.
When creating a pipeline the assembler includes an image size. This
image size can be set when creating the pipeline but if it is 0 then a
default image size will be used. The default is 2 GB except for ami
images which are 6 GB.
Commit b1c5ef2a introduced support for retrieving logs from osbuild.
This commit finishes the second part - actually returning the logs
from /compose/logs route.
When restarting composer, we were not handling the compose states
correctly.
This resolves that as follows:
* any running composes are marked as failed,
* any waiting composes are put back in the pending jobs queue
As a consequence of needing the ability to reinitialize the job
queue, we must include the depsolved pipeline in the compose object.
This is the correct thing to do, as the semantics we currently
adhere to is that pipelines are depsolved when the compose is
started (and restarting composer should not affect this by for
instance re-depsolve the pipeline).
Resolves rhbz#1784062.
Signed-off-by: Tom Gundersen <teg@jklm.no>
The pipeline generation now takes the architecture as an argument.
Currently only x86_64 is supported. The architecture is detected
at start-up, and passed down to each pipeline translation.
For osbuild-pipeline we now requrie the architecture to be passed
in.
Signed-off-by: Tom Gundersen <teg@jklm.no>
Instead of having a static repository checksum, set it dynamically from
the metadata that osbuild-composer last saw. This is implemented in
dnf-json, which returns the checksums for each repository on every call.
This enables the use of repositories that change over time, such as
fedora-updates. Note that the osbuild pipeline will break when such a
repository changes. This is intentional: pipelines have to be
reproducible.
Prior to this commit outputs directory used by local target was owned by root.
This made impossible for osbuild-composer to delete images. (osbuild-composer
doesn't run as root).
This commit introduces state directory in which osbuild-composer creates
outputs directory. Because this directory is owned by osbuild-composer, it's
able to delete files inside.
Prior to this commit blueprint getters looked like C-style API with output
parameters. This commit refactors them to more conventional multiple return
values API.
As a part of f4991cb1 ComposeEntry struct was removed from store package.
This change made sense because this struct is connected more with API than
with store - store uses its own Compose struct. In addition, converters
between Compose and ComposeEntry were added. Unfortunately, ComposeEntry
contains ImageSize which was not stored in Compose but retrieved from store
using GetImage method. This made those converters dependent on the store,
which was messy.
To solve this issue this commit adds image struct into Compose struct.
The content of image struct is generated on the worker side - when the worker
sets the compose status to FINISHED, it also sends Image struct with detailed
information about the result.
This commit introduces basic support for upload API. Currently, all the routes
required by cockpit-composer are supported (except for /compose/log).
Also, ComposeEntry struct is moved outside of the store package. I decided
to do it because it isn't connected in any way to store, it's more connected
to API. Due to this move there's currently a known bug that image size is
not returned. This should be solved by moving Image struct inside Compose
struct by follow-up PR.
The return value hasn't been predictable prior to this commit. This is
probably caused by internal implementation of Golang map. This commit
fixes it by sorting the output.
Make osbuild-composer use FromHost() directly. Everywhere else needs to
specify the distro explicitly.
Also don't panic when a distro doesn't exist. Instead, return nil. Make
sure all callers check for that.
Automatically registering on `init()` is clever, but a bit too magical
and easy to get wrong, because every binary must include all distros
somewhere.
Flip this inside out: distros now have a `New()`, which returns
something that implements the `Distro` interface. The distro package
explicitly creates all of them.
This means that distros cannot import distro itself anymore, because go
forbids import cycles. This only affected `InvalidOutputFormatError`.
Return a generic error for now.
Add function to get a single blueprint change given the blueprint name
and the commit hash. Update the PushBlueprint function to allow a commit
message to be passed to it which will be used when adding a change.
When a blueprint is pushed to the store it will also add a change to the
BlueprintChanges map. Each blueprint has a list of changes mapped to
their commit. This commmit is a hex string based off of the sha1 hash of
the current time so every commit is unique. The message and timestamp
follow the format of lorax-composer.
lorax-composer uses the host's repositories for building images. This is
prone to accidental configuration errors and duplicates functionality
(adding custom repositories via the source API is much more explicit).
However, blueprints don't specify the distribution they're based on.
This is something they should do in the future to enable cross-distro
image builds. Until then, read `/etc/os-release` to detect the host OS
and use that as the base distro for a blueprint.
Detection works by concatenating the ID field with a `-` and the
VERSION_ID field. This mandates how additional distributions should
register themselves to the `distro` package.
If composer cannot build the detected distro, fall back to fedora-30.
Use this detection everywhere that fedora-30 was hard-coded before.
Introduce the `distro` package, which contains an interface for OS
implementations. Its main purpose is to convert a blueprint to a
distro-specific pipeline.
Also introduce the `distro/fedora30` package. It is the first
implementation of the distro interface. Most of its code has been copied
with minimal modifications from the blueprint package.
The `blueprint` package is now back to serving a single purpose:
representing a weldr blueprint. It does not depend on the `pipeline`
package anymore.
Change osbuild-composer and osbuild-pipeline to use the new API,
hard-coding "fedora-30". This looks a bit weird now, but is the same
behavior as before.
All test cases now also take an "distro" key in the "compose" object.
The api is initialized with a base repo but users may want to add their
own custom repos. Based on lorax's implementation, the store now
supports sources. The source object is similar to but not the same as
the repos used by dnf. The store can now add, delete, and get info about
sources.
There was missing condition for the case when compose push fails.
As the previous commit introduces customizations which can return even more
errors it is important to report such errors.
For now the API is returning only 500 HTTP code. In future we should
distinguish between failures and blueprint validation errors.
All composes now include the image size in their status response. The
image size will be 0 except for finished composes where it will be the
file length in bytes.
Alongside File, Name, and Mime type, each Image in the store will
contain the image size. This will default to 0 unless the compose's
status is FINISHED. Then, it will be the length in bytes of the file.
A job is now in "WAITING" state exactly when it is in the channel,
once it is popped it enters "RUNNING" state. It is only possible
to update the state of a job that is in the running state.
This mean that updating to "RUNNING" is entirely optional, but in
the future we may want to use this as a watchdog logic, and require
the worker to update at regular intervals to avoid being restarted.
The job queue API is updated to require a POST followed by one
or several PATCH messages to the returned ID. If a patch is sent
to an ID before the POST it is as if the object does not exist
(regarldess of it being in the queue in WAITING state or not).
Once a job has been POSTed it can be PATCHed to update it zero or
more times with (still) RUNNING before exactly oncee with either
FINISHED or FAILED.
Signed-off-by: Tom Gundersen <teg@jklm.no>
This was only used internally in the store package, so moved the
one type that was still used and remove the rest.
Signed-off-by: Tom Gundersen <teg@jklm.no>
This hides the state hanlding in the store package. As before, it
can be disabled by passing `nil` instead of the path to the state
file.
Signed-off-by: Tom Gundersen <teg@jklm.no>
Wrap the channel in Pop and Push methods, so it is not exposed to
the callers. PushCompose replaces the old AddCompose for consistency,
and PopCompose simply reads from the other end of the channel.
Signed-off-by: Tom Gundersen <teg@jklm.no>
