Skip to content

Latest commit

 

History

History
614 lines (464 loc) · 36.3 KB

tests.md

File metadata and controls

614 lines (464 loc) · 36.3 KB

Python SDK testing guide

This guide walks through the setup necessary to run tests in the Azure SDK for Python, gives an overview of the shared testing infrastructure, and demonstrates how to write and run tests for a service.

Table of contents

Set up your development environment

The Azure SDK Python team creates libraries that are compatible with Python 3.8 and up. We walk through setting up a Python virtual environment for Python 3.8, but having a virtual environment for each minor version can make it easier to debug PRs locally.

C:\Users> python -m venv env
C:\Users> env\scripts\activate       # PowerShell only
C:\Users> source env\bin\activate    # Linux shell (Bash, ZSH, etc.) only
C:\Users> env\scripts\activate.bat   # Windows CMD only
(env) C:\Users>

To create virtual environment for different versions of Python use the -p flag to pass the specific Python executable you want to use

C:\Users> python -m venv -p <path/to/Python/Python38/python.exe> py38_venv

SDK root directory

In the root directory of our SDK, a number of mandatory files have been added. When creating your own SDK, these files can be copied from the sdk/template project, and modified to your needs.

  • README.md. This is the description and guidance for customers or your SDK. Please see the guide on writing a README to make sure you have the complete content requirements and formatting.
  • CHANGELOG.md. This is where you will add the summary of changes for each new release. Please see the guidance for correct formatting.
  • setup.py. This is the 'installer' for your Python SDK. Please see the guide on Python packaging for details on customizing this for a specific package.
  • setup.cfg. This is an artifact used in building the Python package. Please see the guide on Python packaging for details.
  • MANIFEST.in. This is an artifact used in building the Python package. Please see the guide on Python packaging for details.
  • dev_requirements.txt. This is for developers, and lists the packages required for running the tests and samples. See the dependency installation section below.
  • sdk_packaging.toml. This configuration is used by the packaging pipeline and no further modifications should be required.

Dependency installation

Our SDK will have dependencies on other packages in the Azure Python SDK ecosystem. In order to run our tests and samples, we will need to setup our virtual environment to be able to find these external dependencies within the repo. We use the dev_requirements.txt to list these dependencies as relative paths (along with any other external packages that should be installed from PyPI). The libraries currently listed in this file include azure-core and azure-identity as well as some internal tooling packages and our testing framework libraries. These dependencies can be installed with the following command:

(env) azure-sdk-for-python\sdk\my-service\my-package> pip install -r dev_requirements.txt

Next we will install our Python SDK to the virtual environment as an 'editable install' - this means that as we work on the implementation, we will be able to run the package as it develops, as opposed to having to periodically rebuild and reinstall.

(env) azure-sdk-for-python\sdk\my-service\my-package> pip install -e .

We should now be able to open an interactive Python terminal, and execute code from our new SDK

(env) azure-sdk-for-python\sdk\my-service\my-package> python
>>> import azure.my_package
>>> print(azure.my_package.__version__)
0.0.1

Open code in IDE

Open the directory for your package in your preferred editor, for example VSCode.

(env) azure-sdk-for-python\sdk\my-service\my-package> code .

Integrate with the pytest test framework

As a quick background, the Azure SDK uses the pytest test runner to support creating unit and functional tests for Track 2 Azure libraries. To install pytest run pip install pytest from your virtual environment, you can confirm the installation was successful by running pytest -V. The commands will run all files of the form test_*.py or *_test.py in the provided directory and its subdirectories; for more information check out the docs.

With pytest you can provide a either a directory or a specific test file to run:

(env) azure-sdk-for-python\sdk\my-service\my-package> pytest tests
(env) azure-sdk-for-python\sdk\my-service\my-package> pytest tests\<test_file.py>

In addition you can provide keywords to run specific tests within the suite or within a specific file:

(env) azure-sdk-for-python\sdk\my-service\my-package> pytest tests -k <keyword>
(env) azure-sdk-for-python\sdk\my-service\my-package> pytest <test_file.py> -k <keyword>

If you have print statements in your tests for debugging you can add the -s flag to send those print statements to standard output:

(env) azure-sdk-for-python\sdk\my-service\my-package> pytest <test_file.py> -s

Tox

The Python SDK uses the tox project to automate releases, run tests, run linters, and build our documentation. The tox.ini file is located at azure-sdk-for-python/eng/tox/tox.ini for reference. You do not need to make any changes to the tox file for tox to work with your project. Tox will create a directory (.tox) in the head of your branch. The first time you run tox commands it may take several moments, but subsequent runs will be quicker. To install tox run the following command from within your virtual environment. (env) > pip install "tox<5".

To run a tox command from your directory use the following commands:

(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e sphinx -c ../../../eng/tox/tox.ini --root .
(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e pylint -c ../../../eng/tox/tox.ini --root .
(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e mypy -c ../../../eng/tox/tox.ini --root .
(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e pyright -c ../../../eng/tox/tox.ini --root .
(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e verifytypes -c ../../../eng/tox/tox.ini --root .
(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e whl -c ../../../eng/tox/tox.ini --root .
(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e sdist -c ../../../eng/tox/tox.ini --root .
(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e samples -c ../../../eng/tox/tox.ini --root .
(env) azure-sdk-for-python\sdk\my-service\my-package> tox run -e apistub -c ../../../eng/tox/tox.ini --root .

A quick description of the nine commands above:

  • sphinx: documentation generation using the inline comments written in our code
  • lint: runs pylint to make sure our code adheres to the style guidance
  • mypy: runs the mypy static type checker for Python to make sure that our types are valid.
  • pyright: runs the pyright static type checker for Python to make sure that our types are valid.
  • verifytypes: runs pyright's verifytypes tool to verify the type completeness of the library.
  • whl: creates a whl package for installing our package
  • sdist: creates a zipped distribution of our files that the end user could install with pip
  • samples: runs all of the samples in the samples directory and verifies they are working correctly
  • apistub: runs the apistubgenerator tool on your code

The devtools_testutils package

The Azure SDK team has created some in house tools to make testing easier. These additional tools are located in the devtools_testutils package that was installed with your dev_requirements.txt. In this package is the AzureRecordedTestCase class that every service test class should inherit from. AzureRecordedTestCase provides a number of utility functions for authenticating clients during tests, naming test resources, and sanitizing credentials in recordings.

The devtools_testutils package also has other classes and functions to provide test utility, which are documented in the package README.

Write or run tests

Newer SDK tests utilize the Azure SDK Tools Test Proxy to record and playback HTTP interactions, while also automatically sanitizing sensitive information from recordings. To migrate an existing test suite to use the test proxy, or to learn more about using the test proxy, refer to the test proxy migration guide.

Set up test resources

Live Azure resources will be necessary in order to run live tests and produce recordings. There are PowerShell test resource management commands, documented in /eng/common/TestResources, that streamline this process. Both pure ARM templates (test-resources.json) and BICEP files (test-resources.bicep) are supported.

User-based authentication is preferred when using test resources. To enable this:

  • Use the -UserAuth command flag when running the New-TestResources script.
  • Choose a development tool to authenticate with by setting an AZURE_TEST_USE_*_AUTH environment variable to "true" (tests will authenticate as the tool's logged-in user). The following tools are supported, listed in the order that authentication will be attempted in if requested:
    1. Azure PowerShell: set AZURE_TEST_USE_PWSH_AUTH.
    2. Azure CLI (az): set AZURE_TEST_USE_CLI_AUTH.
    3. Visual Studio Code: set AZURE_TEST_USE_VSCODE_AUTH.
    4. Azure Developer CLI (azd): set AZURE_TEST_USE_AZD_AUTH.
  • Ensure you're logged into the tool you choose -- if you used New-TestResources.ps1 to deploy resources, you'll already have logged in with Azure PowerShell.

Important: these environment variables will only be successfully used if test credentials are fetched with the AzureRecordedTestCase.get_credential method. See Write your tests for details.

If you haven't yet set up a test-resources file for test resource deployment and/or want to use test resources of your own, you can just configure credentials to target these resources instead.

To create a test-resources file:

  1. Create an Azure Resource Management Template for your specific service and the configuration you need. This can be done in the Portal by creating a resource, and at the very last step (Review + Create), clicking "Download a template for automation".
  2. Save this template to a test-resources.json file under the directory that contains your package (sdk/<my-service>/test-resources.json) or create a test-resources.bicep file. You can refer to Key Vault's as an example.
  3. Add templates for any additional resources in a grouped "resources" section of test-resources (example).
  4. Add an "outputs" section to test-resources that describes any environment variables necessary for accessing these resources (example).

Configure test variables

Python SDK tests use a .env file to store test credentials and resource variables. This .env file should be placed at either the root of azure-sdk-for-python in your local file system, or in the directory containing the repo. The python-dotenv package is used to read this file -- documentation of the package and how to format a .env file can be found in the package's README.

If using a New-TestResources script from /eng/common/TestResources, the script should output any environment variables necessary to run live tests for the service. After storing these variables in your .env file -- formatted as VARIABLE=value on separate lines -- your credentials and test configuration variables will be set in our environment when running tests.

If you used the -UserAuth command flag to deploy test resources, choose a development tool to authenticate with by setting an AZURE_TEST_USE_*_AUTH environment variable to "true". The following tools are supported, listed in the order that authentication will be attempted in if requested:

  1. Azure PowerShell: set AZURE_TEST_USE_PWSH_AUTH.
  2. Azure CLI (az): set AZURE_TEST_USE_CLI_AUTH.
  3. Visual Studio Code: set AZURE_TEST_USE_VSCODE_AUTH.
  4. Azure Developer CLI (azd): set AZURE_TEST_USE_AZD_AUTH.

Important: these environment variables will only be successfully used if test credentials are fetched with the AzureRecordedTestCase.get_credential method. See Write your tests for details.

If your service doesn't have a test-resources file for test deployment, you'll need to set environment variables for authentication at minimum. For user-based authentication, use AZURE_TEST_USE_*_AUTH as described above.

For service principal authentication:

  1. Set the AZURE_SUBSCRIPTION_ID variable to your organization's subscription ID. You can find it in the "Overview" section of the "Subscriptions" blade in the Azure Portal.
  2. Define the AZURE_TENANT_ID, AZURE_CLIENT_ID, and AZURE_CLIENT_SECRET of a test service principal. If you do not have a service principal, use the Azure CLI's az ad sp create-for-rbac command (ideally, using your alias as the service principal's name prefix):
az login
az ad sp create-for-rbac --name "{your alias}-tests" --role Contributor

The command will output a set of credentials. Set AZURE_TENANT_ID to the value of "tenant", AZURE_CLIENT_ID to the value of "appId", and AZURE_CLIENT_SECRET to the value of "password".

Start the test proxy server

The test proxy has to be available in order for tests to work; this is done automatically with a pytest fixture.

Create a conftest.py file within your package's test directory (sdk/{service}/{package}/tests), and inside it add a session-level fixture that accepts the test_proxy fixture as a parameter (and has autouse set to True):

import pytest

# autouse=True will trigger this fixture on each pytest run, even if it's not explicitly used by a test method
# test_proxy auto-starts the test proxy
# patch_sleep and patch_async_sleep streamline tests by disabling wait times during LRO polling
@pytest.fixture(scope="session", autouse=True)
def start_proxy(test_proxy, patch_sleep, patch_async_sleep):
    return

As shown in the example, it's recommended to also request the patch_sleep and patch_async_sleep fixtures unless your tests have a unique need to use time.sleep or asyncio.sleep during playback tests (this is unusual). All of these fixtures are imported by the central /sdk/conftest.py, so pytest will automatically resolve the references.

For more details about how this fixture starts up the test proxy, or the test proxy itself, refer to the test proxy migration guide.

Deliver environment variables to tests

Refer to the documentation in devtools_testutils and use the devtools_testutils.EnvironmentVariableLoader to fetch environment variables and provide them to tests.

Write your tests

In your package's tests directory (sdk/{service}/{package}/tests), create a file with the naming pattern test_<what_you_are_testing>.py. The base of each testing file will be roughly the same (in this example we use Schema Registry for the sake of demonstration):

import functools

from devtools_testutils import AzureRecordedTestCase, EnvironmentVariableLoader, recorded_by_proxy

from azure.schemaregistry import SchemaRegistryClient

SchemaRegistryPreparer = functools.partial(
    EnvironmentVariableLoader,
    'schemaregistry',
    schemaregistry_endpoint="https://fake_resource.servicebus.windows.net",
    schemaregistry_group="fakegroup",
)

# The test class name needs to start with "Test" to get collected by pytest
class TestSchemaRegistry(AzureRecordedTestCase):

    # Start with any helper functions you might need, for example a client creation method:
    def create_schemareg_client(self, endpoint):
        credential = self.get_credential(SchemaRegistryClient)
        client = self.create_client_from_credential(SchemaRegistryClient, credential=credential, endpoint=endpoint)
        return client

    ...

    # Write your test cases, each starting with "test_":
    @SchemaRegistryPreparer()
    @recorded_by_proxy
    def test_client_creation(self, schemaregistry_endpoint):
        client = self.create_schemareg_client(schemaregistry_endpoint)
        assert client is not None

There's a lot going on in the example, so we'll take this piece by piece:

  • Import everything you will need in your tests, and include the line from devtools_testutils import AzureRecordedTestCase, EnvironmentVariableLoader, recorded_by_proxy.
  • Include any helper methods you will need at the top of your test class. Most libraries will have a client creation method to eliminate repetitive code.
  • Following your helper methods will be your actual tests. All test method names must start with "test_". The preparer defined at the top of the file should decorate your test in the fashion: @MyPreparer().
    • The signature of your test method will always contain self, and following self will be all the keys that you need from your preparer. A test does not need to accept every key from the preparer; the test framework will only pass in the parameters specifically requested in the test method signature.
  • Fetch a credential for your service client with self.get_credential. This is necessary for the credential to be usable in playback tests, and for it to make use of the authentication method selected by AZURE_TEST_USE_*_AUTH environment variables.
  • Within tests, use the assert keyword to assert conditions that you expect to be true.

If you need logging functionality for your testing, pytest also offers logging capabilities either inline through the caplog fixture or with command line flags.

If your tests use shared resources or configuration that needs to be set up at test-running time, you can refer to the Pre-test setup section of our advanced testing guide for recommended practices.

Configure live or playback testing mode

"Live" tests refer to tests that make requests to actual Azure resources. "Playback" tests require a recording for each test; the test proxy will compare the requests/responses that would be made during each test with requests/responses in the recording.

To run live tests, set the environment variable AZURE_TEST_RUN_LIVE to "true" in your environment or .env file. Live test runs will produce recordings unless the environment variable AZURE_SKIP_LIVE_RECORDING is set to "true" as well. To run tests in playback, either set AZURE_TEST_RUN_LIVE to "false" or leave it unset.

Run and record tests

First, refer to the Configure test variables section to ensure environment variables are set for test resources and authentication. With the AZURE_TEST_RUN_LIVE environment variable set to "true", use pytest to run your test(s) in live mode.

(env) azure-sdk-for-python\sdk\my-service\my-package> pytest tests

This will generate test recordings and enable playback testing. Set AZURE_TEST_RUN_LIVE to "false" and run tests with the same command to verify that playback tests pass.

Playback test errors most frequently indicate a need to add/remove sanitizers and/or add matchers (see Sanitize secrets). If you encounter any unexpected errors, refer to the test proxy troubleshooting guide.

If tests were recorded for a new library, there should now be a folder called recordings inside your package's tests directory. Each recording in this folder will be a .json file that captures the HTTP traffic that was generated while running the test matching the file's name.

The final step in setting up recordings is to move these files out of the azure-sdk-for-python and into the azure-sdk-assets repository. The recording migration guide describes how to do so. This step only needs to be completed once. Your library will have an assets.json file at its root, which stores the azure-sdk-assets tag that contains the current set of recordings.

From this point on, recordings will automatically be fetched when tests are run in playback mode -- either from a local cache (described in Update test recordings), or from azure-sdk-assets if they're not locally available.

Update test recordings

Environment prerequisites
  • The targeted library is already migrated to use the test proxy.
  • Git version > 2.30.0 is installed on the machine and in the path. Git is used by the script and test proxy.
  • Global git config settings are configured for user.name and user.email.
    • These settings are also set with environment variables GIT_COMMIT_OWNER and GIT_COMMIT_EMAIL, respectively (in your environment or your local .env file).
  • Membership in the azure-sdk-write GitHub group.

Test recordings will be updated if tests are run while AZURE_TEST_RUN_LIVE is set to "true" and AZURE_SKIP_LIVE_RECORDING is unset or "false". These recording updates will be reflected in a git-excluded .assets folder at the root of the repo.

The .assets folder contains one or more directories with random names, which each are a git directory containing recordings. To find the directory containing your package's recordings, you can use the manage_recordings.py script from azure-sdk-for-python/scripts. This script accepts a verb and a relative path to your package's assets.json file (this path is optional, and is simply assets.json by default).

For example, to view the location of azure-keyvault-keys's recordings, from a current working directory at the root of the repo, run the following command:

python scripts/manage_recordings.py locate -p sdk/keyvault/azure-keyvault-keys

The output will include an absolute path to the recordings directory; in this case:

C:/azure-sdk-for-python/.assets/Y0iKQSfTwa/python

If you cd into the folder containing your package's recordings, you can use git status to view the recording updates you've made. You can also use other git commands; for example, git diff {file name} to see specific file changes, or git restore {file name} to undo changes you don't want to keep.

After verifying that your recording updates look correct, you can use manage_recordings.py to push your recordings to the azure-sdk-assets repo:

python scripts/manage_recordings.py push -p sdk/{service}/{package}/assets.json

The verbs that can be provided to this script are "locate", "push", "show", "restore", and "reset":

  • locate: prints the location of the library's locally cached recordings.
  • push: pushes recording updates to a new assets repo tag and updates the tag pointer in assets.json.
  • show: prints the contents of the provided assets.json file.
  • restore: fetches recordings from the assets repo, based on the tag pointer in assets.json.
  • reset: discards any pending changes to recordings, based on the tag pointer in assets.json.

After pushing your recordings, the assets.json file for your package will be updated to point to a new Tag that contains the updates. Include this assets.json update in any pull request to update the recordings pointer in the upstream repo.

More details about the recording management commands and management script are documented in manage_recordings.py.

Sanitize secrets

The .json files created from running tests in live mode can include authorization details, account names, shared access signatures, and other secrets. The recordings are included in our public GitHub repository, making it important for us to remove any secrets from these recordings before committing them to the repository.

By default, the test proxy server sanitizes several common patterns of secrets, but there are additional steps you can take to ensure that any other sensitive information is removed from recordings.

There are two primary ways to keep secrets from being written into recordings:

  1. The EnvironmentVariableLoader will automatically sanitize the values of captured environment variables with the provided fake values.
  2. Additional sanitizers can be registered via add_*_sanitizer methods in devtools_testutils. For example, the general-use method for sanitizing recording bodies, headers, and URIs is add_general_string_sanitizer. Other sanitizers are available for more specific scenarios and can be found at devtools_testutils/sanitizers.py.

As a simple example of registering a sanitizer, you can provide the exact value you want to sanitize from recordings as the target in the general string sanitizer. To replace all instances of the string "my-key-vault" with "fake-vault" in recordings, you could add something like the following in the package's conftest.py file:

from devtools_testutils import add_general_string_sanitizer, test_proxy

# autouse=True will trigger this fixture on each pytest run, even if it's not explicitly used by a test method
@pytest.fixture(scope="session", autouse=True)
def add_sanitizers(test_proxy):
    add_general_string_sanitizer(target="my-key-vault", value="fake-vault")

Note that the sanitizer fixture accepts the test_proxy fixture as a parameter to ensure the proxy is started beforehand (see Start the test proxy server).

For a more advanced scenario, where we want to sanitize the account names of all storage endpoints in recordings, we could instead use add_general_regex_sanitizer:

add_general_regex_sanitizer(
    regex="(?<=\\/\\/)[a-z]+(?=(?:|-secondary)\\.(?:table|blob|queue)\\.core\\.windows\\.net)",
    value="fakeendpoint",
)

add_general_regex_sanitizer accepts a regex, replacement value, and capture group as keyword-only arguments. In the snippet above, any storage endpoint URIs that match the specified URI regex will have their account name replaced with "fakeendpoint". A request made to https://tableaccount-secondary.table.core.windows.net will be recorded as being made to https://fakeendpoint-secondary.table.core.windows.net, and URIs will also be sanitized in bodies and headers.

For more details about sanitizers and their options, please refer to devtools_testutils/sanitizers.py.

Special case: SAS tokens

In the past, it was recommended that the tests using Shared Access Signature (SAS) tokens should use the AzureRecordedTestCase.generate_sas method to generate the token and automatically register a sanitizer to keep this token out of test recordings. This method is now deprecated since the test proxy automatically sanitizes SAS tokens. If you have tests that use SAS tokens, you can remove the usage of the generate_sas method.

Opt out of specific sanitizers

Since, in some cases, the default sanitizers might be considered too aggressive and breaks tests during playback, you can opt out of certain sanitizers using the remove_batch_sanitizers function in your respective conftest.py files. For example:

from devtools_testutils import remove_batch_sanitizers, test_proxy


@pytest.fixture(scope="session", autouse=True)
def add_sanitizers(test_proxy):
    ...
    #  Remove the following body key sanitizer: AZSDK3493: $..name
    remove_batch_sanitizers(["AZSDK3493"])

A list of sanitizers and their IDs can be found here. However, please be mindful when opting out of a sanitizer, and ensure that no sensitive data is being exposed.

Some sanitizers IDs that are often opted out of are:

  • AZSDK2003: Location - Header regex sanitizer
  • AZSDK3430: $..id - Body key sanitizer
  • AZSDK3493: $..name - Body key sanitizer

Functional vs. unit tests

The tests written above are functional tests: they generate HTTP traffic and send data to the service. Client operations that interact with the service should be tested by functional tests wherever possible.

Tests that don't make HTTP requests -- e.g. tests for model serialization or mocked service interactions for complex scenarios -- can be referred to as "unit tests". For unit tests, the best practice is to have a separate test class from the one containing functional tests. For example, the azure-data-tables package has client-side validation for the table name and properties of the entity; below is an example of how these could be tested:

import pytest
from azure.data.tables import TableServiceClient, EntityProperty, EdmType

class TestTablesUnitTest(object):

    def test_invalid_table_name(self):
        account_name = 'fake_account_name'
        account_key = 'fake_account_key1234567890'
        tsc = TableServiceClient(
            account_url='https://{}.table.core.windows.net/'.format(account_name),
            credential=account_key
        )

        invalid_table_name = "bad_table_name" # table name cannot have an '_' character

        with pytest.raises(ValueError):
            tsc.create_table(invalid_table_name)

    def test_entity_properties(self):
        ep = EntityProperty('abc', EdmType.STRING)
        ep = EntityProperty(b'abc', EdmType.BINARY)
        ep = EntityProperty(1.2345, EdmType.DOUBLE)

        with pytest.raises(ValueError):
            ep = EntityProperty(2 ** 75, EdmType.Int64) # Tables can only handle integers up to 2 ^ 63

Async tests need to be marked with a @pytest.mark.asyncio to be properly handled. For example:

import pytest
from azure.data.tables.aio import TableServiceClient

class TestTablesUnitTest(object):

    @pytest.mark.asyncio
    async def test_invalid_table_name(self):
        account_name = 'fake_account_name'
        account_key = 'fake_account_key1234567890'
        tsc = TableServiceClient(
            account_url='https://{}.table.core.windows.net/'.format(account_name),
            credential=account_key
        )

        invalid_table_name = "bad_table_name" # table name cannot have an '_' character

        with pytest.raises(ValueError):
            await tsc.create_table(invalid_table_name)

Further reading

For information about more advanced testing scenarios, refer to the advanced tests notes.