pytest-random-order is a plugin for pytest that randomises the order in which tests are run to reveal unwanted coupling between tests. The plugin allows user to control the level of randomness they want to introduce and to disable reordering on subsets of tests. Tests can be rerun in a specific order by passing a seed value reported in a previous test run.

$ pip install pytest-random-order

The plugin is not enabled by default. You need to include the parameter:

$ pytest --random-order-enable

If you want to enable it by default, pytest offers ways to include arguments on every run, for example inside your pytest.ini file:

[pytest]
addopts= --random-order-enable

After adding the random-order-enable argument by default, you can randomise the order of tests within modules and shuffle the order of test modules (which is the default behaviour of the plugin), just running pytest as always:

$ pytest -v

To change the level of randomness allowed, run pytest with --random-order-bucket=<bucket-type> option where <bucket-type> can be class, module, package, or global:

$ pytest -v --random-order-bucket=package

To disable reordering of tests in a module or class, use pytest marker notation:

pytestmark = pytest.mark.random_order(disabled=True)

To rerun tests in a particular order:

$ pytest -v --random-order-seed=<value-reported-in-previous-run>

pytest-random-order plugin groups tests in buckets, shuffles them within buckets and then shuffles the buckets.

You can choose from four types of buckets:

class

Tests will be shuffled within a class and classes will be shuffled, but tests from one class will never have tests from other classes or modules run in-between them.

module

Same as above at module level. This is the default setting.

package

Same as above at package level. Note that modules (and hence tests inside those modules) that belong to package x.y.z do not belong to package x.y, so they will fall in different buckets when randomising with package bucket type.

global

All tests fall in the same bucket, full randomness, tests probably take longer to run.

If you have three buckets of tests A, B, and C with three tests 1 and 2, and 3 in each of them, then one of many potential orderings that non-global randomisation can produce could be:

c2, c1, c3, a3, a1, a2, b3, b2, b1

As you can see, all C tests are executed "next" to each other and so are tests in buckets A and B. Tests from any bucket X are guaranteed to not be interspersed with tests from another bucket Y. For example, if you choose bucket type module then bucket X contains all tests that are in this module.

By default, your tests will be randomised at module level which means that tests within a single module X will be executed in no particular order, but tests from other modules will not be mixed in between tests of module X.

The randomised reordering can be disabled per module or per class irrespective of the chosen bucket type.

It is best to start with smallest bucket type (class or module depending on whether you have class-based tests), and switch to a larger bucket type when you are sure your tests handle that.

If your tests rely on fixtures that are module or session-scoped, more randomised order of tests will mean slower tests. You probably don't want to randomise at global or package level while you are coding and need a quick confirmation that nothing big is broken.

You can disable shuffling of tests within a single module or class by marking the module or class with random_order marker and passing disabled=True to it:

pytestmark = pytest.mark.random_order(disabled=True)

def test_number_one():
    assert True

def test_number_two():
    assert True

class MyTest(TestCase):
    pytestmark = pytest.mark.random_order(disabled=True)

    def test_number_one(self):
        self.assertTrue(True)

No matter what will be the bucket type for the test run, test_number_one will always run before test_number_two.

If you discover a failing test because you reordered tests, you will probably want to be able to rerun the tests in the same failing order. To allow reproducing test order, the plugin reports the seed value it used with pseudo random number generator:

============================= test session starts ==============================
..
Using --random-order-bucket=module
Using --random-order-seed=24775
...

You can now use the --random-order-seed=... bit as an argument to the next run to produce the same order:

$ pytest -v --random-order-seed=24775

If the plugin misbehaves or you just want to assure yourself that it is not the plugin making your tests fail or pass undeservedly, you can disable it:

$ pytest -p no:random-order -v

Distributed under the terms of the MIT license, "pytest-random-order" is free and open source software