There are hundreds of packages which make up the Operating System. Making sure that they all work together as a whole is not an easy task. This becomes even harder as the number of packages and their inter-dependencies grows. An extensive testing is required before a new version of the operating system is released to ensure it is stable enough. That is the past.
Imagine an Always Ready Operating System which consists of packages which are constantly kept in a good shape. Integrated and stable thanks to an extensive test coverage which is continuously executed upon changes in individual packages, in this way allowing to prepare a new release in much shorter time, or even in no time.
Imagine an operating system distribution which you could release at any moment. This is where we are heading. Here comes the CI, Continuous Integration, as an invaluable tool to ensure everything is working together as expected in every point of time.
Continuous integration aims to ensure broken changes are revealed as soon as possible and do not affect other developers, packagers, maintainers or users. The feedback that continuous integration provides is vital for fast paced agile delivery of software. Late testing, long after a change occurs, does not scale to the pace of Fedora. Learn the goals, terminology and rules for a working CI in the manifesto.
There are three main pieces of the puzzle to get this nicely working: A process which clearly defines how to discover and execute tests, a set of tools which help to efficiently implement the process and the tests themselves.
Standard Test Interface
In order to clearly distinguish test from the CI system running it the Standard Test Interface was introduced. It clearly defines essential terms such as test, test subject, test suite, test framework, test result, test artifact, test system and describes what are their responsibilities and requirements.
This general approach gives a nice flexibility as it does not enforce any specific tools or frameworks to be used. Basically it only describes how tests are discovered and where the testing results should be stored to be processed by the automation.
When a test fails, CI can prevent the broken change from affecting other packages. That gating happens in Bodhi. Greewave is used together with ResultsDB and WaiverDB to make the gating decision.
- Gating ... how to enable gating, waiving instructions
- Bodhi ... Fedora updates System
- Greewave ... service to evaluate gating policies based on test results
- ResultsDB ... results store engine
- WaiverDB ... service for recording waivers against test results
Fedora Notifications have been adjusted to notify by default every packager when any step of the CI pipeline fails on one of the package they maintain. So if you are a kernel maintainer and a commit made to the kernel dist-git repository fails to compose an OSTree, FMN will notify you of it.
Bodhi includes the CI results in its update page, just as it already includes tests results from taskotron.
Various tools involved in the CI automation are sending updates about the progress of the testing to the message bus. Consistent format of the CI messages allows to build simpler services which will provide useful features for everyone.
- Fedora CI Messages ... CI messages specification
- Datagrepper ... search sent messages (e.g. by topic)
- fedmsg bus ... messages sent by the pipeline
- topics ... list of fedmsg topics related to CI
Standard Test Roles
Standard Test Roles were implemented to enable both automation tools and developers in their local environments to easily execute tests. This set of ansible roles supports various frameworks and allows to execute tests against different test subjects (such as classic rpm package, docker container or Atomic Host).
As Standard Test Interface does not define test framework which should be used for testing it is possible to choose the most suitable framework for your project. Here are some examples:
Standard Test Interface defines only a very simple metadata for selecting which tests should be run. For more complex scenarios Flexible Metadata Format can be used:
The testing Pipeline detects tests for enabled packages, executes the test coverage and gathers the results.
Test results from the CI pipeline are displayed in Pagure web interface. See the commits page and pull request pages of respective package to see the results. Pagure integration with COPR provides auto-rebuilding and pull request/commit flagging on new changes.
The core of the CI success are reliable tests of a good quality, well selected, stable, organized and continuously maintained.
In general it makes sense to store tests as close to the upstream as possible. So what are the appropriate test types recommended for testing the Always Ready Operating System?
- Basic functionality tests
- Integration tests
For unit tests it usually makes more sense to store them directly within the upstream project repository. However, in some cases it might be worth to fetch tests for Fedora CI from the upstream repository as well.
Tests may be written all sorts of different ways, but have to be
exposed and invoked in a standard way. Tests are enabled by
tests/tests.yml file in the package
dist-git repository as defined by the Standard Test Interface.
Test code itself can be stored directly in the dist-git or fetched
from another repository. Shared tests namespace can be used for
storing test code relevant for multiple packages.
- Quick Start Guide ... quick intro for the impatient
- Tests ... how to run, write and wrap tests
- Share Test Code ... shared tests namespace
- Pull Requests ... how to create pull requests
- Examples ... real-life success stories of enabled tests
Executing a test written with the use of Standard Test Roles is as
simple as running an ansible playbook
tests.yml. However, a set of environment variables needs to
be set properly in order to execute the test against the desired
test subject. The Tests wiki contains detailed instructions
about running tests and adding new test coverage.
- Tests ... How to run, write and wrap tests
There are statistics pages to track how many Fedora dist-git repositories have tests in the Standard Test Format. These pages are updated automatically.
There is an active effort to open source existing internal Red Hat tests to Fedora called Upstream First. Tests being migrated can be found in the upstream first repository:
The ownership and maintenance of tests should be shared between QE & Devel. For tests that reside in the rpm namespace, QE can use pull requests to create/update tests. Likewise in the tests namespace, both QE and Devel will have commit rights, both QE and Devel should review and sign-off with each commit.
If you have questions or would like to get involved:
- IRC channel:
- Mailing list: firstname.lastname@example.org (archive)
- Issues: General, Atomic CI
Here's a summary of useful links:
- Standard Test Interface ... definition of the process
- Standard Test Roles ... set of ansible roles
- Workshop ... workshop materials
- Tests ... executing and adding tests
- Upstream First ... tests to be ported to dist-git
- Where do I get the latest Atomic Host images for testing?
- Why not rely on %check?
%check and the tests run in the CI pipeline are complementary. %check allows to perform a number of checks at build time but will not detect missing requirement, missing configuration file and this kind of situation which the CI pipeline will be able to test and find. We could see this as %check being unit-test (where the unit is the package) versus integration tests where the tests are run in the entire distribution just as we distribute it to our users. So the Ci pipeline will be able to find integration bugs that the %check section will never be able to.