Salt comes with a powerful integration and unit test suite allowing for the fully automated run of integration and/or unit tests from a single interface. It uses the combination of pytest, nox and Kitchen Salt to run these tests. Nox is used to manage all of the test python dependencies. When you run the test runner with nox, you will be installing the same python dependencies that we use to run our test suite on PRs and branch tests. Kitchen Salt is used to spin up our virtual machines based off of golden images. These virtual machines use the salt-jenkins sls states to configure any system dependencies.
To learn the basics of how Salt's test suite works, be sure to check out the Salt's Test Suite: An Introduction tutorial.
Nox is used to manage all of the python dependencies used in the test suite
and spins up the different nox sessions. You can look at the
in the salt repo to view all of the current nox configurations. In that file
you will notice various nox sessions. When creating each of these sessions,
nox will create a virtualenv with the specified interpreter. Once the virtualenv
is created it will also install all of the required python dependencies
required for that session and run the tests.
For example if you want to run all of the tests using the zeromq transport on python3 you would need to specify the zeromq transport and python3.
nox -e 'test-zeromq-3(coverage=False)'
And because zeromq is the default transport, the following nox session can also be used:
nox -e 'test-zeromq-3(coverage=False)'
To run all the tests but on the tcp transport, you would need to specify the tcp session.
nox -e 'test-tcp-3(coverage=False)'
As a contrast, when using the deprecated
runtests.py test runner, the
command would be:
nox -e 'runtests-tcp-3(coverage=False)'
You can view all available sessions by running:
For the most part you will only need nox to run the test suite, as this tool
will install the exact same python dependencies we use to run on our test runs.
The exception to this is when a system dependency is required, for example
These system dependencies are installed with sls states managed in the salt-jenkins
repo or you can manually install the dependency yourself.
The system dependencies are installed from the salt-jenkins repo. The
golden-images-provision state is what is run to determine what dependencies
to install on which platform.
We run this state only when we want to update our current VM images with new
We also use Kitchen Salt to spin up the VM's used for testing. You can view the kitchen-salt getting started for instructions on how to install and set it up. Kitchen Salt uses Test Kitchen to spin up the VM or container in the configured provider. Once the VM is spun up, Kitchen Salt can install salt and run a particular set of states. In the case of our branch and PR tests we create "Golden Images" which run the salt-jenkins states and install salt system dependencies beforehand. We only update these "Golden Images" when we need to upgrade or install a system dependency. You can view the kitchen-salt jenkins setup docs for instructions on how to set up Kitchen Salt similar to the jenkins environment we use to run branch and PR tests.
Salt's test suite is located in the
tests/ directory in the root of
With the migration to PyTest, Salt has created a separate directory for tests
that are written taking advantage of the full potential of PyTest. These are
As for the old test suite, it is divided into two main groups:
Within each of these groups, the directory structure roughly mirrors the
structure of Salt's own codebase. Notice that there are directories for
output, and more in each testing
The files that are housed in the
modules directory of either the unit
or the integration testing factions contain respective integration or unit
test files for Salt execution modules.
The PyTest only tests under
tests/pytests should, more or less, follow the
same grouping as the old test suite.
The Integration section of Salt's test suite start up a number of Salt daemons to test functionality in a live environment. These daemons include two Salt Masters, one Syndic, and two Minions. This allows the Syndic interface to be tested and Master/Minion communication to be verified. All of the integration tests are executed as live Salt commands sent through the started daemons.
Integration tests are particularly good at testing modules, states, and shell commands, among other segments of Salt's ecosystem. By utilizing the integration test daemons, integration tests are easy to write. They are also SaltStack's generally preferred method of adding new tests.
The discussion in the Integration vs. Unit section of the testing tutorial is beneficial in learning why you might want to write integration tests vs. unit tests. Both testing arenas add value to Salt's test suite and you should consider adding both types of tests if possible and appropriate when contributing to Salt.
Unit tests do not spin up any Salt daemons, but instead find their value
in testing singular implementations of individual functions. Instead of
testing against specific interactions, unit tests should be used to test
a function's logic as well as any
Unit tests also rely heavily on mocking external resources.
The discussion in the Integration vs. Unit section of the testing tutorial is useful in determining when you should consider writing unit tests instead of, or in addition to, integration tests when contributing to Salt.
There is only one requirement to install, to quickly get started
running salt's test suite:
pip install nox
Once this requirement is installed, you can use the
nox binary to run
all of the tests included in Salt's test suite:
nox -e 'test-3(coverage=False)'
For more information about options you can pass the test runner, see the
nox -e 'test-3(coverage=False)' -- --help
Instead of running the entire test suite all at once, which can take a long time, there are several ways to run only specific groups of tests or individual tests:
Run unit tests only:
'test-3(coverage=False)' -- tests/unit/.
Run unit and integration tests for states:
'test-3(coverage=False)' -- tests/unit/states/ tests/integration/states/.
Run integration tests for an individual module:
nox -e 'test-3(coverage=False)' --
Run unit tests for an individual module:
nox -e 'test-3(coverage=False)' --
Run an individual test by using the class and test name (this example is for the
test_default_kvm_profile test in the
nox -e 'test-3(coverage=False)' --
For more specific examples of how to run various test subsections or individual tests, please see the pytest documentation on how to run specific tests or the Running Specific Tests section of the Salt's Test Suite: An Introduction tutorial.
Since the unit tests do not require a master or minion to execute, it is often useful to be able to run unit tests individually, or as a whole group, without having to start up the integration testing daemons. Starting up the master, minion, and syndic daemons takes a lot of time before the tests can even start running and is unnecessary to run unit tests. To run unit tests without invoking the integration test daemons, simply add the unit directory as an argument:
nox -e 'test-3(coverage=False)' -- tests/unit/
All of the other options to run individual tests, entire classes of tests, or entire test modules still apply.
Salt is used to change the settings and behavior of systems. In order to effectively test Salt's functionality, some integration tests are written to make actual changes to the underlying system. These tests are referred to as "destructive tests". Some examples of destructive tests are changes may be testing the addition of a user or installing packages. By default, destructive tests are disabled and will be skipped.
Generally, destructive tests should clean up after themselves by attempting to restore the system to its original state. For instance, if a new user is created during a test, the user should be deleted after the related test(s) have completed. However, no guarantees are made that test clean-up will complete successfully. Therefore, running destructive tests should be done with caution.
Running destructive tests will change the underlying system. Use caution when running destructive tests.
To run tests marked as destructive, set the
nox -e 'test-3(coverage=False)' -- --run-destructive
Salt's testing suite also includes integration tests to assess the successful creation and deletion of cloud instances using Salt-Cloud for providers supported by Salt-Cloud.
The cloud provider tests are off by default and run on sample configuration files
tests/integration/files/conf/cloud.providers.d/. In order to run
the cloud provider tests, valid credentials, which differ per provider, must be
supplied. Each credential item that must be supplied is indicated by an empty
string value and should be edited by the user before running the tests. For
example, DigitalOcean requires a client key and an api key to operate. Therefore,
the default cloud provider configuration file for DigitalOcean looks like this:
digitalocean-config: driver: digitalocean client_key: '' api_key: '' location: New York 1
As indicated by the empty string values, the
client_key and the
must be provided:
digitalocean-config: driver: digitalocean client_key: wFGEwgregeqw3435gDger api_key: GDE43t43REGTrkilg43934t34qT43t4dgegerGEgg location: New York 1
When providing credential information in cloud provider configuration files, do not include the single quotes.
Once all of the valid credentials for the cloud provider have been supplied, the cloud provider tests can be run like:
nox -e 'test-cloud-3(coverage=False)'
SaltStack maintains a Jenkins server to allow for the execution of tests across supported platforms. The tests executed from Salt's Jenkins server create fresh virtual machines for each test run, then execute destructive tests on the new, clean virtual machine.
SaltStack's Jenkins server continuously runs the entire test suite, including destructive tests, on an array of various supported operating systems throughout the day. Each actively supported branch of Salt's repository runs the tests located in the respective branch's code. Each set of branch tests also includes a pylint run. These branch tests help ensure the viability of Salt code at any given point in time as pull requests are merged into branches throughout the day.
In addition to branch tests, SaltStack's Jenkins server also runs tests on pull requests. These pull request tests include a smaller set of virtual machines that run on the branch tests. The pull request tests, like the branch tests, include a pylint test as well.
When a pull request is submitted to Salt's repository on GitHub, the suite of pull request tests are started by Jenkins. These tests are used to gauge the pull request's viability to merge into Salt's codebase. If these initial tests pass, the pull request can then merged into the Salt branch by one of Salt's core developers, pending their discretion. If the initial tests fail, core developers may request changes to the pull request. If the failure is unrelated to the changes in question, core developers may merge the pull request despite the initial failure.
As soon as the pull request is merged, the changes will be added to the next branch test run on Jenkins.
For a full list of currently running test environments, go to https://jenkins.saltproject.io.
The salt testing infrastructure is divided into two classes of tests, integration tests and unit tests. These terms may be defined differently in other contexts, but for Salt they are defined this way:
Unit Test: Tests which validate isolated code blocks and do not require
external interfaces such as
salt-call or any of the salt daemons.
Integration Test: Tests which validate externally accessible features.
Salt testing uses unittest2 from the python standard library and MagicMock.
Any function in either integration test files or unit test files that is doing
the actual testing, such as functions containing assertions, must start with
def test_user_present(self): ...
When functions in test files are not prepended with
test_, the function
acts as a normal, helper function and is not run as a test by the test suite.
Which branch of the Salt codebase should new tests be written against? The location of where new tests should be submitted depends largely on the reason you're writing the tests.
If you are adding new functionality to Salt, please write the tests for this new
feature in the same pull request as the new feature. New features should always be
submitted to the
If you have already submitted the new feature, but did not write tests in the original
pull request that has already been merged, please feel free to submit a new pull
request containing tests. If the feature was recently added to Salt's
branch, then the tests should be added there as well. However, if the feature was
master some time ago and is already present in one or more release
branches, please refer to the Tests for Entire Files or Functions section below
for more details about where to submit tests for functions or files that do not
already have tests.
If you are writing tests for code that fixes a bug in Salt, tests will be required before merging the PR. A great option for most bugfixes is to adopt a TDD style approach:
reproduce the issue
write a test that exhibits the behavior
write the bugfix
This helps ensure that known issues are not reintroduced into the codebase.
Sometimes entire files in Salt are completely untested. If you are writing tests for a file that doesn't have any tests written for it, write your test against the earliest supported release branch that contains the file or function you're testing.
Once your tests are submitted in a pull request and is merged into the branch in
question, the tests you wrote will be merged-forward by SaltStack core engineers and
the new tests will propagate to the newer release branches. That way the tests you
wrote will apply to all current and relevant release branches, and not just the
branch, for example. This methodology will help protect against regressions on older
files in Salt's codebase.
There may be times when the tests you write against an older branch fail in the merge-forward process because functionality has changed in newer release branches. In these cases, a Salt core developer may reach out to you for advice on the tests in question if the path forward is unclear.
If tests are written against a file in an older release branch and then merged forward, there may be new functionality in the file that is present in the new release branch that is untested.It would be wise to see if new functionality could use additional testing once the test file has propagated to newer release branches.
If you need access to module or global level variables, please use a pytest fixture. The use of module and global variables can introduce mutable global objects and increases processing time because all globals are evaluated when collecting tests. If there is a use case where you cannot use a fixture and you are using a type of string, integer, or tuple you can use global/module level variables. Any mutable types such as lists and dictionaries must use pytest fixtures. For an example, if all of your tests need access to a string variable:
FOO = "bar" def test_foo_bar(): assert FOO == "bar" def test_foo_not(): assert not FOO == "foo"
We recommend using a pytest fixture:
import pytest @pytest.fixture() def foo(): return "bar" def test_foo_bar(foo): assert foo == "bar" def test_foo_not(foo): assert not foo == "foo"
If you need a class to mock something, it can be defined at the global scope, but it should only be initialized on the fixture:
class Foo: def __init__(self): self.bar = True @pytest.fixture def foo(): return Foo()
Several Salt-specific helpers are available. A full list is available by inspecting functions exported under tests/support/*.py.
@pytest.mark.expensive_test -- Designates a test which typically requires a relatively costly external resource, like a cloud virtual machine. This decorator is not normally used by developers outside of the Salt core team.
@pytest.mark.destructive_test -- Marks a test as potentially destructive. It
will not be run unless the
--run-destructive flag is expressly passed.
@pytest.mark.requires_network -- Requires a network connection for the test to operate successfully. If a network connection is not detected, the test will not run.
These are just a small preview of the supported marker. For a full listing, please run:
nox -e 'test-3(coverage=False)' -- --markers