We support auto-instrumentation for the following CI providers:

If you are using auto-instrumentation for one of these providers, you can skip the rest of the setup steps below.

If you are using a cloud CI provider without access to the underlying worker nodes, such as GitHub Actions or CircleCI, configure the library to use the Agentless mode. For this, set the following environment variables:

DD_CIVISIBILITY_AGENTLESS_ENABLED=true (Required)

Enables or disables Agentless mode.
Default: false

DD_API_KEY (Required)

The Datadog API key used to upload the test results.
Default: (empty)

Additionally, configure the Datadog site to which you want to send data.

DD_SITE (Required)

The Datadog site to upload results to.
Default: datadoghq.com

If you are running tests on an on-premises CI provider, such as Jenkins or self-managed GitLab CI, install the Datadog Agent on each worker node by following the Agent installation instructions. This is the recommended option as it allows you to automatically link test results to logs and underlying host metrics.

If you are using a Kubernetes executor, Datadog recommends using the Datadog Operator. The operator includes Datadog Admission Controller which can automatically inject the tracer library into the build pods. Note: If you use the Datadog Operator, there is no need to download and inject the tracer library since the Admission Controller can do this for you, so you can skip the corresponding step below. However, you still need to make sure that your pods set the environment variables or command-line parameters necessary to enable Test Visibility.

If you are not using Kubernetes or can’t use the Datadog Admission Controller and the CI provider is using a container-based executor, set the DD_TRACE_AGENT_URL environment variable (which defaults to http://localhost:8126) in the build container running the tracer to an endpoint that is accessible from within that container. Note: Using localhost inside the build references the container itself and not the underlying worker node or any container where the Agent might be running in.

DD_TRACE_AGENT_URL includes the protocol and port (for example, http://localhost:8126) and takes precedence over DD_AGENT_HOST and DD_TRACE_AGENT_PORT, and is the recommended configuration parameter to configure the Datadog Agent’s URL for CI Visibility.

If you still have issues connecting to the Datadog Agent, use the Agentless Mode. Note: When using this method, tests are not correlated with logs and infrastructure metrics.

To enable instrumentation of pytest tests, add the --ddtrace option when running pytest, specifying the name of the service or library under test in the DD_SERVICE environment variable, and the environment where tests are being run (for example, local when running tests on a developer workstation, or ci when running them on a CI provider) in the DD_ENV environment variable:

DD_SERVICE=my-python-app DD_ENV=ci pytest --ddtrace

If you also want to enable the rest of the APM integrations to get more information in your flamegraph, add the --ddtrace-patch-all option:

DD_SERVICE=my-python-app DD_ENV=ci pytest --ddtrace --ddtrace-patch-all

Adding custom tags to tests

To add custom tags to your tests, declare ddspan as an argument in your test:

from ddtrace import tracer

# Declare `ddspan` as argument to your test
def test_simple_case(ddspan):
    # Set your tags
    ddspan.set_tag("test_owner", "my_team")
    # test continues normally
    # ...

To create filters or group by fields for these tags, you must first create facets. For more information about adding tags, see the Adding Tags section of the Python custom instrumentation documentation.

Adding custom measures to tests

Just like tags, to add custom measures to your tests, use the current active span:

from ddtrace import tracer

# Declare `ddspan` as an argument to your test
def test_simple_case(ddspan):
    # Set your tags
    ddspan.set_tag("memory_allocations", 16)
    # test continues normally
    # ...

Read more about custom measures in the Add Custom Measures Guide.

To instrument your benchmark tests with pytest-benchmark, run your benchmark tests with the --ddtrace option when running pytest, and Datadog detects metrics from pytest-benchmark automatically:

def square_value(value):
    return value * value


def test_square_value(benchmark):
    result = benchmark(square_value, 5)
    assert result == 25

To enable instrumentation of unittest tests, run your tests by appending ddtrace-run to the beginning of your unittest command.

Make sure to specify the name of the service or library under test in the DD_SERVICE environment variable. Additionally, you may declare the environment where tests are being run in the DD_ENV environment variable:

DD_SERVICE=my-python-app DD_ENV=ci ddtrace-run python -m unittest

Alternatively, if you wish to enable unittest instrumentation manually, use patch() to enable the integration:

from ddtrace import patch
import unittest
patch(unittest=True)

class MyTest(unittest.TestCase):
def test_will_pass(self):
assert True

Plugins for pytest that alter test execution may cause unexpected behavior.

Parallelization

Plugins that introduce parallelization to pytest (such as pytest-xdist or pytest-forked) create one session event for each parallelized instance. Multiple module or suite events may be created if tests from the same package or module execute in different processes.

The overall count of test events (and their correctness) remain unaffected. Individual session, module, or suite events may have inconsistent results with other events in the same pytest run.

Test ordering

Plugins that change the ordering of test execution (such as pytest-randomly) can create multiple module or suite events. The duration and results of module or suite events may also be inconsistent with the results reported by pytest.

The overall count of test events (and their correctness) remain unaffected.

In some cases, if your unittest test execution is run in a parallel manner, this may break the instrumentation and affect test visibility.

Datadog recommends you use up to one process at a time to prevent affecting test visibility.