Testing Ansible

Why test your Ansible contributions?

If you’re a developer, one of the most valuable things you can do is to look at GitHub issues and help fix bugs, since bug-fixing is almost always prioritized over feature development. Even for non-developers, helping to test pull requests for bug fixes and features is still immensely valuable.

Ansible users who understand how to write playbooks and roles should be able to test their work. GitHub pull requests will automatically run a variety of tests (e.g., Shippable) that show bugs in action. However, contributors must also test their work outside of the automated GitHub checks and show evidence of these tests in the PR to ensure that their work will be more likely to be reviewed and merged.

Read on to learn how Ansible is tested, how to test your contributions locally, and how to extend testing capabilities.

Types of tests

At a high level we have the following classifications of tests:

compile
  • Compile Tests

  • Test python code against a variety of Python versions.

sanity
  • Sanity Tests

  • Sanity tests are made up of scripts and tools used to perform static code analysis.

  • The primary purpose of these tests is to enforce Ansible coding standards and requirements.

integration
units
  • Unit Tests

  • Tests directly against individual parts of the code base.

If you’re a developer, one of the most valuable things you can do is look at the GitHub issues list and help fix bugs. We almost always prioritize bug fixing over feature development.

Even for non developers, helping to test pull requests for bug fixes and features is still immensely valuable. Ansible users who understand writing playbooks and roles should be able to add integration tests and so GitHub pull requests with integration tests that show bugs in action will also be a great way to help.

Testing within GitHub & Shippable

Organization

When Pull Requests (PRs) are created they are tested using Shippable, a Continuous Integration (CI) tool. Results are shown at the end of every PR.

When Shippable detects an error and it can be linked back to a file that has been modified in the PR then the relevant lines will be added as a GitHub comment. For example:

The test `ansible-test sanity --test pep8` failed with the following errors:

lib/ansible/modules/network/foo/bar.py:509:17: E265 block comment should start with '# '

The test `ansible-test sanity --test validate-modules` failed with the following errors:
lib/ansible/modules/network/foo/bar.py:0:0: E307 version_added should be 2.4. Currently 2.3
lib/ansible/modules/network/foo/bar.py:0:0: E316 ANSIBLE_METADATA.metadata_version: required key not provided @ data['metadata_version']. Got None

From the above example we can see that --test pep8 and --test validate-modules have identified issues. The commands given allow you to run the same tests locally to ensure you’ve fixed the issues without having to push your changed to GitHub and wait for Shippable, for example:

If you haven’t already got Ansible available, use the local checkout by running:

source hacking/env-setup

Then run the tests detailed in the GitHub comment:

ansible-test sanity --test pep8
ansible-test sanity --test validate-modules

If there isn’t a GitHub comment stating what’s failed you can inspect the results by clicking on the “Details” button under the “checks have failed” message at the end of the PR.

Rerunning a failing CI job

Occasionally you may find your PR fails due to a reason unrelated to your change. This could happen for several reasons, including:

  • a temporary issue accessing an external resource, such as a yum or git repo

  • a timeout creating a virtual machine to run the tests on

If either of these issues appear to be the case, you can rerun the Shippable test by:

  • closing and re-opening the PR

  • making another change to the PR and pushing to GitHub

If the issue persists, please contact us in the #ansible-devel irc channel on the irc.libera.chat IRC network.

How to test a PR

Ideally, code should add tests that prove that the code works. That’s not always possible and tests are not always comprehensive, especially when a user doesn’t have access to a wide variety of platforms, or is using an API or web service. In these cases, live testing against real equipment can be more valuable than automation that runs against simulated interfaces. In any case, things should always be tested manually the first time as well.

Thankfully, helping to test Ansible is pretty straightforward, assuming you are familiar with how Ansible works.

Setup: Checking out a Pull Request

You can do this by:

  • checking out Ansible

  • making a test branch off the main branch

  • merging a GitHub issue

  • testing

  • commenting on that particular issue on GitHub

Here’s how:

Warning

Testing source code from GitHub pull requests sent to us does have some inherent risk, as the source code sent may have mistakes or malicious code that could have a negative impact on your system. We recommend doing all testing on a virtual machine, whether a cloud instance, or locally. Some users like Vagrant or Docker for this, but they are optional. It is also useful to have virtual machines of different Linux or other flavors, since some features (apt vs. yum, for example) are specific to those OS versions.

Create a fresh area to work:

git clone https://github.com/ansible/ansible.git ansible-pr-testing
cd ansible-pr-testing

Next, find the pull request you’d like to test and make note of the line at the top which describes the source and destination repositories. It will look something like this:

Someuser wants to merge 1 commit into ansible:devel from someuser:feature_branch_name

Note

Only test ansible:devel

It is important that the PR request target be ansible:devel, as we do not accept pull requests into any other branch. Dot releases are cherry-picked manually by Ansible staff.

The username and branch at the end are the important parts, which will be turned into git commands as follows:

git checkout -b testing_PRXXXX devel
git pull https://github.com/someuser/ansible.git feature_branch_name

The first command creates and switches to a new branch named testing_PRXXXX, where the XXXX is the actual issue number associated with the pull request (for example, 1234). This branch is based on the devel branch. The second command pulls the new code from the users feature branch into the newly created branch.

Note

If the GitHub user interface shows that the pull request will not merge cleanly, we do not recommend proceeding if you are not somewhat familiar with git and coding, as you will have to resolve a merge conflict. This is the responsibility of the original pull request contributor.

Note

Some users do not create feature branches, which can cause problems when they have multiple, unrelated commits in their version of devel. If the source looks like someuser:devel, make sure there is only one commit listed on the pull request.

The Ansible source includes a script that allows you to use Ansible directly from source without requiring a full installation that is frequently used by developers on Ansible.

Simply source it (to use the Linux/Unix terminology) to begin using it immediately:

source ./hacking/env-setup

This script modifies the PYTHONPATH environment variables (along with a few other things), which will be temporarily set as long as your shell session is open.

Testing the Pull Request

At this point, you should be ready to begin testing!

Some ideas of what to test are:

  • Create a test Playbook with the examples in and check if they function correctly

  • Test to see if any Python backtraces returned (that’s a bug)

  • Test on different operating systems, or against different library versions

Any potential issues should be added as comments on the pull request (and it’s acceptable to comment if the feature works as well), remembering to include the output of ansible --version

Example:

Works for me! Tested on `Ansible 2.3.0`.  I verified this on CentOS 6.5 and also Ubuntu 14.04.

If the PR does not resolve the issue, or if you see any failures from the unit/integration tests, just include that output instead:

This doesn’t work for me.

When I ran this Ubuntu 16.04 it failed with the following:

```
some output
StackTrace
some other output
```

Code Coverage Online

The online code coverage reports are a good way to identify areas for testing improvement in Ansible. By following red colors you can drill down through the reports to find files which have no tests at all. Adding both integration and unit tests which show clearly how code should work, verify important Ansible functions and increase testing coverage in areas where there is none is a valuable way to help improve Ansible.

The code coverage reports only cover the devel branch of Ansible where new feature development takes place. Pull requests and new code will be missing from the codecov.io coverage reports so local reporting is needed. Most ansible-test commands allow you to collect code coverage, this is particularly useful to indicate where to extend testing. See Testing Ansible for more information.

Want to know more about testing?

If you’d like to know more about the plans for improving testing Ansible then why not join the Testing Working Group.