Skip to content

pytest terraform plugin with fixtures and offline replay support

Notifications You must be signed in to change notification settings

cloud-custodian/pytest-terraform

Repository files navigation

Introduction

CI codecov

pytest_terraform is a pytest plugin that enables executing terraform to provision infrastructure in a unit/functional test as a fixture.

This plugin features uses a fixture factory pattern to enable paramterized construction of fixtures via decorators.

Usage

pytest_terraform provides a terraform decorator with the following parameters:

Argument Required? Type Default Description
terraform_dir yes String Terraform module (directory) to execute.
scope no String "function" Pytest scope - should be one of: function, or session. Other scopes like class, module, and package should work but have not been fully tested.
replay no Boolean True Use recorded resources instead of invoking terraform. See Replay Support for more details.
name no String None Name used for the fixture. This defaults to the terraform_dir when None is supplied.
teardown no String "default" Configure which teardown mode is used for terraform resources. See Teardown Options for more details.

Example

from boto3 import Session
from pytest_terraform import terraform


# We use the terraform decorator to create a fixture with the name of
# the terraform module.
#
# The test function will be invoked after the terraform module is provisioned
# with the results of the provisioning.
#
# The module `aws_sqs` will be searched for in several directories, the test
# file directory, a sub directory `terraform`.
#
# This fixture specifies a session scope and will be run once per test run.
#
@terraform('aws_sqs', scope='session')
def test_sqs(aws_sqs):
    # A test is passed a terraform resources class containing content from
    # the terraform state file.
    #
    # Note the state file contents may vary across terraform versions.
    #
    # We can access nested datastructures with a jmespath expression.
    assert aws_sqs["aws_sqs_queue.test_queue.tags"] == {
        "Environment": "production"
    }
   queue_url = aws_sqs['test_queue.queue_url']
   print(queue_url)


def test_sqs_deliver(aws_sqs):
   # Once a fixture has been defined with a decorator
   # it can be reused in the same module by name, with provisioning
   # respecting scopes.
   #
   sqs = Session().client('sqs')
   sqs.send_message(
       QueueUrl=aws_sqs['test_queue.queue_url'],
       MessageBody=b"123")


@terraform('aws_sqs')
def test_sqs_dlq(aws_sqs):
   # the fixture can also referenced again via decorator, if redefined
   # with decorator the fixture parameters much match (ie same session scope).

   # Module outputs are available as a separate mapping.
   aws_sqs.outputs['QueueUrl']

Note the fixture name should match the terraform module name

Note The terraform state file is considered an internal implementation detail of terraform, not per se a stable public interface across versions.

Marks

All tests using terraform fixtures have a terraform mark applied so they can be run/selected via the command line ie.

pytest -k terraform tests/

to run all terraform tests only. See pytest mark documentation for additional details, https://docs.pytest.org/en/stable/example/markers.html#mark-examples

Options

You can provide the path to the terraform binary else its auto discovered

--tf-binary=$HOME/bin/terraform

Terraform modules referenced by fixtures are looked up in a few different locations, directly in the same directory as the test module, in a subdir named terraform, and in a sibling directory named terraform. An explicit directory can be given which will be looked at first for all modules.

--tf-mod-dir=terraform

This plugin also supports flight recording (see next section)

--tf-replay=[record|replay|disable]

Teardown Options

pytest_terraform supports three different teardown modes for the terraform decorator. The default, pytest_terraform.teardown.ON will always attempt to teardown any and all modules via terraform destory. If for any reason destroy fails it will raise an exception to alert the test runner. The next mode, pytest_terraform.teardown.IGNORE, will invoke terraform destroy as with teardown.ON but will ignore any failures. This mode is particularly help if your test function performs destructive actions against any objects created by the terraform module. The final option is pytest_terraform.teardown.OFF which will remove the teardown method register all together. This should generally only be used in very specific situations and is considered an edge case.

There is a special pytest_terraform.teardown.DEFAULT which is what the teardown parameter actually defaults to.

Teardown options are available, for convenience, on the terraform decorator. For example, set teardown to ignore:

from pytest_terraform import terraform


@terraform('aws_sqs', teardown=terraform.TEARDOWN_IGNORE)
def test_sqs(aws_sqs):
    assert aws_sqs["aws_sqs_queue.test_queue.tags"] == {
        "Environment": "production"
    }
   queue_url = aws_sqs['test_queue.queue_url']
   print(queue_url)

Hooks

pytest_terraform provides hooks via the pytest hook implementation. Hooks should be added in the conftest.py file.

pytest_terraform_modify_state

This hook is executed after state has been captured from terraform apply and before writing to disk. This hook does not modify state that's passed to the function under test. The state is passed as the kwarg tfstate which is a TerraformStateJson UserString class with the following methods and properties:

  • TerraformStateJson.dict - The deserialized state as a dict
  • TerraformStateJson.update(state: str) - Replace the serialized state with a new state string
  • TerraformStateJson.update_dict(state: dict) - Replace the serialized state from a dictionary

Example

def pytest_terraform_modify_state(tfstate):
    print(str(tfstate))

Example AWS Account scrub

import re

def pytest_terraform_modify_state(tfstate):
    """ Replace potential AWS account numbers with 'REDACTED' """
    tfstate.update(re.sub(r'([0-9]+){12}', 'REDACTED', str(tfstate)))

Flight Recording

The usage/philosophy of this plugin is based on using flight recording for unit tests against cloud infrastructure. In flight recording rather than mocking or stubbing infrastructure, actual resources are created and interacted with with responses recorded, with those responses subsequently replayed for fast test execution. Beyond the fidelity offered, this also enables these tests to be executed/re-recorded against live infrastructure for additional functional/release testing.

https://cloudcustodian.io/docs/developer/tests.html#creating-cloud-resources-with-terraform

Replay Support

By default fixtures will save a tf_resources.json back to the module directory, that will be used when in replay mode.

Replay can be configured by passing --tf-replay on the cli or via pytest config file.

Recording

Passing the fixture parameter replay can control the replay behavior on an individual test. The default is to operate in recording mode.

@terraform('file_example', replay=False) def test_file_example(file_example): assert file_example['local_file.bar.content'] == 'bar!'

XDist Compatibility

pytest_terraform supports pytest-xdist in multi-process (not distributed) mode.

When run with python-xdist, pytest_terraform treats all non functional scopes as per test run fixtures across all workers, honoring their original scope lifecycle but with global semantics, instead of once per worker (xdist default).

To enable this the plugin does multi-process coodination using lock files, a test execution log, and a dependency mapping of fixtures to tests. Any worker can execute a module teardown when its done executing the last test that depends on a given fixture. All provisioning and teardown are guarded by atomic file locks in the pytest execution's temp directory.

Root module references

terraform_remote_state can be used to introduce a dependency between a scoped root modules on an individual test, note we are not attempting to support same scope inter fixture dependencies as that imposes additional scheduling constraints outside of pytest native capabilities. The higher scoped root module (ie session or module scoped) will need to have output variables to enable this consumption.