CI/Standard Test Roles

Description

Package standard-test-roles provides shared Ansible roles and inventory scripts implementing the Standard Test Interface. It has support for multiple testing frameworks (such as BeakerLib or Avocado) and in this way allows to easily enable existing tests in Fedora CI.

Setup

Packages

As the first step install all necessary packages:

sudo dnf install fedpkg libselinux-python standard-test-roles

You can also install the latest version from the copr repo:

dnf copr -y enable @osci/standard-test-roles
dnf update standard-test-roles

Artifacts

Output of the test (such as the stdout/stderr output, log files or screenshots) is by default saved in the artifacts directory. Use TEST_ARTIFACTS environment variable to choose a different location if desired:

export TEST_ARTIFACTS=/tmp/artifacts

Artifacts cleanup
Before running tests make sure that all logs /tmp/artifacts/test.* are deleted.

Inventory

A test subject is what we call the thing to be tested. To turn a test subject into a launched, installed system to be tested, we use Ansible dynamic inventory. Use the following command to enable it:

export ANSIBLE_INVENTORY=$(test -e inventory && echo inventory || echo /usr/share/ansible/inventory)

As you can see from the way how the inventory is set, tests may contain their own inventory, which defines their own instructions for turning a test subject into one or more testable systems.

Testing

Classic

You can always invoke the tests locally. Many tests modify or change the system they are run against, so take that into account when looking at how to invoke tests. The following examples invoke tests against the same system that the package git repository is checked out on. Below there are further options for invoking tests against another fully formed and integrated systems, such as an Atomic Host or container image test subject.

There may be more than one test present in a package git repository, but the file tests.yml is the main entry point. To run it use the following command:

# ansible-playbook tests.yml

You can find output artifacts of the tests in an artifacts/ or specify a specific directory like this:

# ansible-playbook -e artifacts=/tmp/output tests.yml

You can filter which kinds of tests are run by providing a --tags argument. To only run tests that are suited for classic systems installed by yum or dnf you can use a command like:

# ansible-playbook --tags=classic tests.yml

When run by a CI System the tests are invoked according to the Standard Test Interface. Look here for more details and standard invocation variables.

Package

When you run the tests as above, the tests assume that the system to be tested is the same as the system invoking the tests. In particular, the test assumes that the thing to be tested is already installed.

A test subject is what we call the thing to be tested. RPMs are a particular kind of test subject. To turn a test subject into a launched, installed system to be tested, we use Ansible dynamic inventory. Lets invoke the tests with an inventory and a specific version of gzip:

curl -o gzip.rpm https://kojipkgs.fedoraproject.org//packages/gzip/1.8/2.fc26/x86_64/gzip-1.8-2.fc26.x86_64.rpm
export TEST_SUBJECTS=$PWD/gzip.rpm
ansible-playbook tests.yml

You'll notice that the RPM is installed into the testable system before invoking the tests. Some tests contain their own inventory, that is their own instructions for turning a test subject into one or more testable systems. But in this case we use the default standard-test-roles inventory in /usr/share/ansible/inventory to do this.

Container

Another example is to use a test subject of a container image. This is also a fully formed and integrated deliverable. The test subject again represents the thing to be tested. The container image is pulled from a registry and launched using docker by an Ansible dynamic inventory.

export TEST_SUBJECTS=docker:docker.io/library/fedora:26
ansible-playbook --tags=container tests.yml

If you watch closely you'll notice the image is pulled if not already local, launched as a container, and then prepared for the tests to run on. The first time this may take a little longer. Not all tests are able to function in the somewhat different environment of a container. In fact, for certain tests, the software to be tested may not be included in the container. But many of the tests for core packages should work here.

The --tags argument filters out tests that are not suitable for running in a container, either because the system functions differently, or the correct packages are not installable.

See the Debug section for instructions how to log into a running container and diagnose why the tests failed.

Additional arguments for Docker

Tests for containers are run with a help of Docker. Containers are run within default security context. For more info see Seccomp security profiles for Docker. It is possible that some tests require additional privileges. In this case specify necessary arguments for Docker using an environment variable TEST_DOCKER_EXTRA_ARGS. For this create a file inventory file in tests directory with the following content:

#!/bin/bash
export TEST_DOCKER_EXTRA_ARGS="--security-opt seccomp:unconfined"
exec merge-standard-inventory "$@"

or

#!/bin/bash
export TEST_DOCKER_EXTRA_ARGS="--privileged"
exec merge-standard-inventory "$@"

See merge-standard-inventory documentation for details.

Atomic

The former example may seem a bit contrived, but the concept of a test subject starts to make more sense when you want to test a fully formed and integrated deliverable, such as Atomic Host. The test subject again represents the thing to be tested. The test subject in this case is a QCow2 image. To turn a test subject into a launched system ready to be tested, we use Ansible dynamic inventory.

curl -Lo /tmp/atomic.qcow2 https://getfedora.org/atomic_qcow2_latest
export TEST_SUBJECTS=/tmp/atomic.qcow2
ansible-playbook --tags=atomic tests.yml

If you watch closely you'll see that the Atomic Host image is booted, and the tests run against the launched image. Not all tests are able to function in the somewhat different environment of Atomic Host, in fact, for certain cases, the software to be tested may not be included in the Atomic Host test subject. But most of the tests in core packages should work here.

Some tests contain their own inventory, that is their own instructions for turning a test subject into one or more testable systems. But in this case we use the default standard-test-roles inventory to do this.

The --tags argument filters out tests that are not suitable for running on an Atomic Host, either because the system functions differently, or the correct packages are not available on that system.

See the Debug section to learn how to diagnose why the tests failed, and log into the running Atomic Host.

Required Packages
Please note that if required_packages are specified in tests.yml for Atomic Host, additional packages will be installed using the rpm-ostree command which is affecting the test subject (it's similar as rebuilding an rpm package to be tested) so this should be used with caution and only when necessary. Also be aware that there are certain limitations for this approach (e.g. it's not possible to install different version of packages that are already part of the tree).

Required Packages
Atomic Host is shipped as a base ostree image, however you can install additional packages with a help of rpm-ostree install command. Currently (10.01.2018 ) repo with additional packages is actual only for the latest base-ostree image. Consequence: tests that install additional packages for Atomic Host can fail sometimes with: error: The following base packages would be replaced: xxx Solution: make sure you have the latest Atomic Host image. Additional information you can find rpm-ostree issue 415 and a possible solution in the feature using rpm-ostree jigdo rpm-ostree issue 1081

Debug

To increase output verbosity use option -v or -vvv:

ansible-playbook --tags=container tests.yml -v

or for full verbosity:

ansible-playbook --tags=container tests.yml -vvv

To debug tests in a running container or atomic host use the TEST_DEBUG environment variable. After the playbook runs, you'll see diagnosis information with a helpful command to log in.

export TEST_DEBUG=1

For container you'll see output like this:

DIAGNOSE: docker exec -it 56de801f0ddde36fc9770666f7be2a68f89d7f18f52b7b6fe7df7a12b193bf08 /bin/bash
DIAGNOSE: kill 18261 # when finished

For atomic host the instructions are a bit different:

DIAGNOSE: ssh -p 2222 -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null root@127.0.0.3 # password: foobar
DIAGNOSE: export ANSIBLE_INVENTORY=/tmp/inventory-cloudxyhF2M/inventory
DIAGNOSE: kill 16611 # when finished

Now you can easily connect using these commands. Use suggested kill command to finish the running instance when done with investigation.

Roles

Here's the list of currently supported roles:

• standard-test-avocado - role for executing tests written via the Avocado testing framework
• standard-test-basic
• standart-test-beakerlib
• standard-test-repo
• standard-test-rhts-rhts
• standard-test-rpm
• standard-test-scripts
• standard-test-source

standard-test-beakerlib

If you have a set of beakerlib tests, it is recommended to place each test in its own subdirectory.

Then, create tests.yml file with contents similar to the following which runs tests using the 'standard-test-beakerlib' role included in the standard-test-roles package. The list provided for the 'tests' parameter should be the list of names of your per-test subdirectories, and the 'required_packages' parameter should contain a list of additional packages that need to be installed to run the tests.

---
- hosts: localhost
  tags:
  - atomic
  - classic
  - container
  roles:
  - role: standard-test-beakerlib
    tests:
    - cmd-line-options
    required_packages:
    - which         # which package required for cmd-line-options
    - rpm-build     # upstream-testsuite requires rpmbuild command
    - libtool       # upstream-testsuite requires libtool
    - gettext       # upstream-testsuite requires gettext

Note: The 'required_packages' parameter is ignored when running on Atomic Host--since there is no way to install additional packages in that environment.

standard-test-rhts

If you have a set of tests you want to run using restraint, each test must be placed in its own subdirectory.

Then, create tests.yml file with contents similar to the following which runs tests using the 'standard-test-rhts' role included in the standard-test-roles package. The list provided for the 'tests' parameter should be the list of names of your per-test subdirectories, and the 'required_packages' parameter should contain a list of additional packages that need to be installed to run the tests.

---
- hosts: localhost
  tags:
  - classic
  - docker
  roles:
  - role: standard-test-rhts
    tests:
    - cmd-line-options
    required_packages:
    - which         # which package required for cmd-line-options
    - rpm-build     # upstream-testsuite requires rpmbuild command
    - libtool       # upstream-testsuite requires libtool
    - gettext       # upstream-testsuite requires gettext

Important Note: Tests using the 'standard-test-rhts' role are not compatible with atomic and docker environments. Please reflect that by omitting atomic and container from the <tags> section of the tests.yml playbook.

More

Links

Pagure and Copr repositories:

• https://pagure.io/standard-test-roles/
• https://copr.fedorainfracloud.org/coprs/g/osci/standard-test-roles/builds/

Contact

• Andrei Stepanov (astepano)
• Miroslav Vadkerti (mvadkert)