This guide will show you the way through the dense forest of available tmt features, commands and options. But don’t be afraid, we will start slowly, with the simple examples first. And then, when your eyes get accustomed to the shadow of omni-present metadata trees, we will slowly dive deeper and deeper so that you don’t miss any essential functionality which could make your life smarter, brighter and more joyful. Let’s go, follow me…

The First Steps

Installing the main package with the core functionality is quite straightforward. No worry, there are just a few dependencies:

sudo dnf install -y tmt

Enabling a simple smoke test in the continuous integration should be a joy. Just a couple of concise commands, assuming you are in your project git repository:

tmt init --template mini
vim plans/example.fmf

Open the example plan in your favorite editor and adjust the smoke test script as needed. Your very first plan can look like this:

summary: Basic smoke test
    script: foo --version

Now you’re ready to create a new pull request to check out how it’s working. During push, remote usually shows a direct link to the page with a Create button, so now it’s only two clicks away:

git add .
git checkout -b smoke-test
git commit -m "Enable a simple smoke test"
git push origin -u smoke-test

But perhaps, you are a little bit impatient and would like to see the results faster. Sure, let’s try the smoke test here and now, directly on your localhost:

tmt run --all provision --how local

If you’re a bit afraid that the test could break your machine or just want to keep your environment clean, run it in a container instead:

sudo dnf install -y tmt-provision-container
tmt run -a provision -h container

Or even in a full virtual machine if the container environment is not enough. We’ll use the libvirt to start a new virtual machine on your localhost. Be ready for a bit more dependencies here:

sudo dnf install -y tmt-provision-virtual
tmt run -a provision -h virtual

Don’t care about the disk space? Simply install tmt-all and you’ll get all available functionality at hand. Check the help to list all supported provision methods:

sudo dnf install tmt-all
tmt run provision --help

Now when you’ve met your --help friend you know everything you need to get around without getting lost in the forest of available options:

tmt --help
tmt run --help
tmt run provision --help
tmt run provision --how container --help

Go on and explore. Don’t be shy and ask, --help is eager to answer all your questions ;-)

Checking data validity

It is easy to introduce a syntax error to one of the fmf files and make the whole tree broken. You should run tmt lint before pushing changes, ideally even before you commit your changes.

You can set up pre-commit to do it for you. Add to your repository’s .pre-commit-config.yaml:

- repo:
  rev: 1.23.0
  - id: tmt-lint

This will run tmt lint --source for all modified fmf files. There are hooks to just check tests tmt-tests-lint, plans tmt-plans-lint or stories tmt-stories-lint explicitly. From time to time you might want to run pre-commit autoupdate to refresh config to the latest version.

Under The Hood

Now let’s have a brief look under the hood. For storing all config data we’re using the Flexible Metadata Format. In short, it is a yaml format extended with a couple of nice features like inheritance or virtual hierarchy which help to maintain even large data efficiently without unnecessary duplication.

The data are organized into trees. Similarly as with git, there is a special .fmf directory which marks the root of the fmf metadata tree. Use the init command to initialize it:

tmt init

Do not forget to include this special .fmf directory in your commits, it is essential for building the fmf tree structure which is created from all *.fmf files discovered under the fmf root.


As we’ve seen above, in order to enable testing the following plan is just enough:

    script: foo --version

Store these two lines in a *.fmf file and that’s it. Name and location of the file is completely up to you, plans are recognized by the execute key which is required. Once the newly created plan is submitted to the CI system test script will be executed.

By the way, there are several basic templates available which can be applied already during the init by using the --template option or the short version -t. The minimal template, which includes just a simple plan skeleton, is the fastest way to get started:

tmt init -t mini

Plans are used to enable testing and group relevant tests together. They describe how to discover tests for execution, how to provision the environment, how to prepare it for testing, how to execute tests, report results and finally how to finish the test job.

Here’s an example of a slightly more complex plan which changes the default provision method to container to speed up the testing process and ensures that an additional package is installed before the testing starts:

    how: container
    image: fedora:33
    how: install
    package: wget
    how: tmt
    script: wget

Note that each of the steps above uses the how keyword to choose the desired method which should be applied. Steps can provide multiple implementations which enables you to choose the best one for your use case. For example to prepare the guest it’s possible to use the install method for simple package installations, ansible for more complex system setup or shell for arbitrary shell commands.


Very often testing is much more complex than running just a single shell script. There might be many scenarios covered by individual scripts. For these cases the discover step can be instructed to explore available tests from fmf metadata as well. The plan will look like this:

    how: fmf
    how: tmt

Tests, identified by the required key test, define attributes which are closely related to individual test cases such as the test script, framework, directory path where the test should be executed, maximum test duration or packages required to run the test. Here’s an example of test metadata:

summary: Fetch an example web page
test: wget
require: wget
duration: 1m

Instead of writing the plan and test metadata manualy, you might want to simply apply the base template which contains the plan mentioned above together with a test example including both test metadata and test script skeleton for inspiration:

tmt init --template base

Similar to plans, it is possible to choose an arbitrary name for the test. Just make sure the test key is defined. However, to organize the metadata efficiently it is recommended to keep tests and plans under separate folders, e.g. tests and plans. This will also allow you to use inheritance to prevent unnecessary data duplication.


It’s always good to start with a “why”. Or, even better, with a story which can describe more context behind the motivation. Stories can be used to track implementation, test and documentation coverage for individual features or requirements. Thanks to this you can track everything in one place, including the project implementation progress. Stories are identified by the story attribute which every story has to define or inherit.

An example story can look like this:

    As a user I want to see more detailed information for
    particular command.
  - tmt test show -v
  - tmt test show -vvv
  - tmt test show --verbose

In order to start experimenting with the complete set of examples covering all metadata levels, use the full template which creates a test, a plan and a story:

tmt init -t full


Finally, there are certain metadata keys which can be used across all levels. Core attributes cover general metadata such as summary or description for describing the content, the enabled attribute for disabling and enabling tests, plans and stories and the link key which can be used for tracking relations between objects.

Here’s how the story above could be extended with the core attributes description and link:

    Different verbose levels can be enabled by using the
    option several times.
  - implemented-by: /tmt/
  - documented-by: /tmt/
  - verified-by: /tests/core/dry

Last but not least, the core attribute adjust provides a flexible way to adjust metadata based on the Context. But this is rather a large topic, so let’s keep it for another time. In the next chapter we’ll learn how to comfortably create new tests and plans.