102 lines
4.3 KiB
Markdown
102 lines
4.3 KiB
Markdown
Introduction
|
|
============
|
|
|
|
The OBS (Open Build Service) was created at a time when VCS field was still
|
|
evolving. One of the main issues not handled at the time was handling of large
|
|
files. Large files are at the core of package sources -- think, upstream
|
|
sources.
|
|
|
|
Today, this has changed. Git is the most popular and widely used VCS in history.
|
|
Entire businesses are build around providing services for Git. Git has also
|
|
ability to deal with large files via Git LFS subsystem.
|
|
|
|
Here, we'll detail how Git and Git LFS can be leveraged to provide a superior
|
|
contributor environment while increasing flexibility and transparency and
|
|
tracability in OBS project management.
|
|
|
|
|
|
Overview of current project
|
|
---------------------------
|
|
|
|
OBS is used to build projects. It doesn't build package or images, but it *only* builds
|
|
projects. And while package management is exposed directly, project management is
|
|
hidden behind APIs, legacy workflows and a monolithic codebase.
|
|
|
|
The goal of this project is to move *project* management to Git and facilitate
|
|
*project* workflow via external, adaptable helpers. As a consequence, OBS will
|
|
be used to build any project, in Git or in legacy VCS internal to OBS. But
|
|
Git-based projects will no longer be curtailed by internal OBS machinery and can
|
|
adapt any project specific workflow in a modular fashion, without the need to
|
|
change OBS sources.
|
|
|
|
The goal is to move current workflows for openSUSE:Factory, as well as SLFO,
|
|
out of OBS and into Git. OBS will still be used to build such projects, but
|
|
everything else, from approvals to maintainer definitions, to project configs,
|
|
must be moved to Git. Doing so will not only simplify workflows for package
|
|
maintainers but also will inject transparency in project history and secure our
|
|
infrastructure with modern cryptography.
|
|
|
|
|
|
How Git works
|
|
-------------
|
|
|
|
Git contains only 4 basic objects:
|
|
* blob -- file data
|
|
* tree -- directory listing, contains other trees or blob or commit, or
|
|
commits (aka, submodules)
|
|
* commit -- this contains parent commit information, tree objects, forming
|
|
unchangable, sealed history backed by a cryptographic hash function (kind of like a
|
|
Bitcoin blockchain)
|
|
* tags -- additional labels associated with commits
|
|
|
|
A good way of thinking about Git is not as a VCS, but as a multi-version file
|
|
system, where each revision is sealed by the new revisions.
|
|
|
|
Each of the objects is represented internally as part of another object via SHA256. Therefore, integrity of Git, along with entire evolution of the sources, is backed by SHA256.
|
|
|
|
In contrast, integrity function used by OBS is MD5.
|
|
|
|
|
|
|
|
Workflow of Git Projects
|
|
------------------------
|
|
|
|
OBS connects package with project. A commit to a package, updates the projects
|
|
where the instance of the package resides.
|
|
|
|
Git does not contain notion of projects and packages. It simply manages source
|
|
trees. The work associated with managing a project and its packages is now done
|
|
externally.
|
|
|
|
The basic structure of a Git managed project is below. The package repository
|
|
contains package sources, while project repository contains all the information
|
|
associated with the project, including pointers (aka, git submodules) of all the package sources.
|
|
|
|
![ProjectGit submodule points to commit in PackageGit](project.svg "ProjectGit with corresponding PackageGit")
|
|
|
|
An update in package must be represented as an update to the project. This can
|
|
happen in two ways. Either direct using `prjgit-updater`, which updates the
|
|
project git directly on pushes,
|
|
|
|
![ProjectGit submodule is updated following push to PackageGit](project-update.svg "ProjectGit update following PackgeGit push")
|
|
|
|
or indirectly via `pr-review` workflow, which updates project git via PR
|
|
workflow,
|
|
|
|
![PackageGit submodule is updated via PR rerouted through ProjectGit](project-pr-update.svg "ProjectGit PR mergesPackgeGit PR")
|
|
|
|
In all cases, the project must be updated for the changes to be built. This is
|
|
akin to OBS today, except that the project is an internal state, mostly hidden
|
|
from inspection.
|
|
|
|
Centralization of package management
|
|
------------------------------------
|
|
|
|
The proposal is to move all "official" package sources under a `/pool`
|
|
organization. Each "official" project would then have *one* branch assigned to
|
|
help with package updates.
|
|
|
|
The branches represent the current state of packages in a given project. Basic
|
|
package updates follow the `pr-review` workflow.
|
|
|