There are two types of automated tests you can add for Azure CLI: unit tests and integration tests.
For unit tests, we support unit tests written in the forms standard unittest.
For integration tests, we provide the ScenarioTest
and LiveScenarioTest
classes to support replayable tests via VCR.py.
Azure CLI translates user inputs into Azure Python SDK calls which communicate with Azure REST API. These HTTP communications are captured and recorded so the integration tests can be replayed in an automation environment without making actual HTTP calls. This ensures that the commands actually work against the service when they are recorded (and then can be re-run live to verify) and provides protection against regressions or breaking changes when they are played back.
The scenario tests are run in replayable mode in the Travis CI during pull request verification and branch merge. However, the tests will be run in live mode nightly in internal test infrastructure. See Test Policies for more details.
The rationale behind the nightly live test:
- The live scenario tests actually verify the end to end scenario.
- The live scenario tests ensure the credibility of the tested scenario.
- The test recording tends to go stale. The sample it captures will eventually deviate from the actual traffic samples.
- The tests in playback mode does not verify the request body and it doesn't ensure the correct requests sequence.
- The unhealthy set of live tests prevent the Azure CLI team from rebaselining tests rapidly.
- Neglecting the live tests will reduce the quality and the credibility of the test bed.
It is a requirement for the command owner to maintain their test in live mode.
- DO NOT USE hard-coded or otherwise persistent resources. This makes it difficult or impossible to re-record tests or run them live. In general, all resources needed to support the test should be created as part of the test and torn down after the test finishes. The use of
ResourceGroupPreparer
helps to facilitate this. - Tests MUST be included for all new command modules and any new commands to existing test modules. PRs will be rejected outright if they do not include tests.
- Name test methods in the following format:
test_<module>_<feature>
. - The scenario test must be able to run repeatedly in live mode. The feature owner is responsible of maintaining their scenario tests.
- 100% commands coverage of module: Scenerio tests MUST cover all commands in the module except the
wait
commands. Please useazdev cmdcov {module}
to check command coverage of module. If cannot add a test because of some reason, please add missing_command_test_coverage in the linter_exclusions.yml. - 100% examples coverage of command: Scenerio tests MUST coverage all examples of the command.
- 100% arguments coverage of command: Scenerio tests should coverage all arguments in the command. Please use
azdev cmdcov {module} --level argument
to check argument coverage. If cannot add a test because of some reason, please add missing_parameter_test_coverage in the linter_exclusions.yml. - 100% boundary values coverage of argument: Scenerio tests should coverage boundary values of argument, especially
''
(empty string),null
,0
andFalse
values. These values have different meanings, but will all resolve toFalse
in conditional expressions (if {variable}
) of Python. It's important to make sure they work as expected.
- Set up your development environment
- Select the API profile you want to test. (Optional)
- Select the cloud you want to test. (Optional) You may need to register a cloud first (for example, if targeting a staging or test environment).
- Log in your Azure account using
az login
command. - Select the subscription you want to use for recording tests. The replaying does not rely on the knowledge of the subscription and the credentials will be removed from the recordings.
After the test is executed, a recording file will be generated at recording/<api_profile_name>/<test_name>.yaml
. The recording file will be created no matter the test pass or not. The behavior makes it easy for you to find issues when a test fails. To re-record the test, you can either delete the existing recording and re-run the test, or simply re-run the test using the --live
flag (ex: azdev test example_test --live
.
It is a good practice to add recording file to the local git cache, which makes it easy to diff the different versions of recording to detect issues or changes.
Once the recording file is generated, execute the test again. This time the test will run in playback mode. The execution is offline, and will not act on the Azure subscription.
If the replay passes, you can commit the tests as well as recordings and submit them as part of your PR.
When a recording file is missing, the test framework will execute the test in live mode. You can also force tests to run live either by setting the environment variable AZURE_TEST_RUN_LIVE
or by using the --live
flag with the azdev test
command.
Also, you can author tests which are for live test only by deriving the test class from LiveScenarioTest
. Since these tests will not be run as part of the CI, you lose much of the automatic regression protection by doing this. However, for certain tests that cannot be re-recorded, cannot be replayed, or fail due to service issues, this is a viable approach.
Recorded tests are run on TravisCI as part of all pull request validations. Therefore it is important to keep the recordings up-to-date, and to ensure that they can be re-recorded.
Live tests run nightly in a separate system and are not tied to pull requests.
Here are some issues that may occur when authoring tests that you should be aware of.
- Non-deterministic results: If you find that a test will pass on some playbacks but fail on others, there are a couple possible things to check:
- check if your command makes use of concurrency.
- check your parameter aliasing (particularly if it complains that a required parameter is missing that you know is there) If your command makes use of concurrency, consider using unit tests, LiveScenarioTest, or, if practical, forcing the test to operate on a single thread for recording and playback.
- Paths: When including paths in your tests as parameter values, always wrap them in double quotes. While this isn't necessary when running from the command line (depending on your shell environment), it will likely cause issues with the test framework.
- Defaults: Ensure you don't have any defaults configured with
az configure
prior to running tests. Defaults can interfere with the expected test flow.
from azure.cli.testsdk import ScenarioTest
class StorageAccountTests(ScenarioTest):
def test_list_storage_account(self):
self.cmd('az storage account list')
Notes:
- When the test is run and no recording file is available, the test will be run in live mode. A recording file will be created at
recording/<test_method_name>.yaml
. - Wrap the command in the
self.cmd
method. It will assert that the exit code of the command is zero. - All the functions and classes you need for writing tests are included in the
azure.cli.testsdk
namespace. It is recommended not to import from a sub-namespace to avoid breaking changes.
class StorageAccountTests(ScenarioTest):
def test_list_storage_account(self):
accounts_list = self.cmd('az storage account list').get_output_in_json()
assert len(accounts_list) > 0
Notes:
- The return value of
self.cmd
is an instance of the classExecutionResult
. It has the exit code and stdout as its properties. get_output_in_json
deserializes the output to a JSON object.
Tip: Don't make any rigid assertions based on assumptions which may not stand in a live test environment.
from azure.cli.testsdk import ScenarioTest
class StorageAccountTests(ScenarioTest):
def test_list_storage_account(self):
self.cmd('az account list-locations', checks=[
self.check("[?name=='westus'].displayName | [0]", 'West US')
])
Notes:
- The first argument in the
check
method is a JMESPath query. JMESPath is a query language for JSON. - If a command returns JSON, multiple JMESPath based checks can be added to the checks list to validate the result.
- In addition to the
check
method, there are other checks likeis_empty
which validate the output isNone
. The check mechanism is extensible. Any callable accepting a singleExecutionResult
argument can act as a check: see checkers.py.
from azure.cli.testsdk import ScenarioTest, ResourceGroupPreparer
class StorageAccountTests(ScenarioTest):
@ResourceGroupPreparer()
def test_create_storage_account(self, resource_group):
self.cmd('az group show -n {rg}', checks=[
self.check('name', '{rg}'),
self.check('properties.provisioningState', 'Succeeded')
])
Notes:
- The preparers are executed before each test in the test class when
setUp
is executed. Any resources created in this way will be cleaned up after testing. Unless you specify that a preparer use an existing resource via its associated environment variable. See Test-Related Environment Variables. - The resource group name is injected into the test method as a parameter. By default
ResourceGroupPreparer
passes the value as theresource_group
parameter. The target parameter can be customized (see following samples). - The resource group will be deleted asynchronously for performance reason.
- The resource group will automatically be registered into the tests keyword arguments (
self.kwargs
) with the key default key ofrg
. This can then be directly plugged into the command string incmd
and into the verification step of thecheck
method. The test infrastructure will automatically replace the values.
class StorageAccountTests(ScenarioTest):
@ResourceGroupPreparer(parameter_name='group_name', parameter_name_for_location='group_location')
def test_create_storage_account(self, group_name, group_location):
self.kwargs.update({
'loc': group_location
})
self.cmd('az group show -n {rg}', checks=[
self.check('name', '{rg}'),
self.check('location', '{loc}'),
self.check('properties.provisioningState', 'Succeeded')
])
Notes:
- In addition to the name, the location of the resource group can be also injected into the test method.
- Both parameters' names can be customized.
- You can add the location to the test's kwargs using the
self.kwargs.update
method. This allows you to take advantage of the automatic kwarg replacement in your command strings and checks.
class StorageAccountTests(ScenarioTest):
@ResourceGroupPreparer(parameter_name_for_location='location')
def test_create_storage_account(self, resource_group, location):
self.kwargs.update({
'name': self.create_random_name(prefix='cli', length=24),
'loc': location,
'sku': 'Standard_LRS',
'kind': 'Storage'
})
self.cmd('az storage account create -n {name} -g {rg} --sku {sku} -l {loc}')
self.cmd('az storage account show -n {name} -g {rg}', checks=[
self.check('name', '{name}'),
self.check('location', '{loc}'),
self.check('sku.name', '{sku}'),
self.check('kind', '{kind}')
])
Note:
One of the most important features of ScenarioTest
is name management.
For the tests to be able to run in a live environment and avoid name collision
a strong name randomization is required.
On the other hand, for the tests to be recorded and replayed,
the naming mechanism must be repeatable during playback mode.
The self.create_random_name
method helps the test achieve both goals.
The method will create a random name during recording, and when it is called during playback, it returns a name (internally it is called moniker) based on the sequence of the name request. The order won't change once the test is written. Peek into the recording file, and you will find no random name. For example, note names like 'clitest.rg000001' in the sample recording below: they aren't the names of the resources which are actually created in Azure. They're replaced before the requests are recorded.
- request:
body: '{"location": "westus", "tags": {"use": "az-test"}}'
headers:
Accept: [application/json]
Accept-Encoding: ['gzip, deflate']
CommandName: [group create]
Connection: [keep-alive]
Content-Length: ['50']
Content-Type: [application/json; charset=utf-8]
User-Agent: [python/3.5.2 (Darwin-16.4.0-x86_64-i386-64bit) requests/2.9.1 msrest/0.4.6
msrest_azure/0.4.7 resourcemanagementclient/0.30.2 Azure-SDK-For-Python
AZURECLI/2.0.0+dev]
accept-language: [en-US]
method: PUT
uri: https://management.azure.com/subscriptions/00000000-0000-0000-0000-000000000000/resourcegroups/clitest.rg000001?api-version=2016-09-01
response:
body: {string: '{"id":"/subscriptions/00000000-0000-0000-0000-000000000000/resourceGroups/clitest.rg000001","name":"clitest.rg000001","location":"westus","tags":{"use":"az-test"},"properties":{"provisioningState":"Succeeded"}}'}
headers:
cache-control: [no-cache]
content-length: ['326']
content-type: [application/json; charset=utf-8]
date: ['Fri, 10 Mar 2017 17:59:58 GMT']
expires: ['-1']
pragma: [no-cache]
strict-transport-security: [max-age=31536000; includeSubDomains]
x-ms-ratelimit-remaining-subscription-writes: ['1199']
status: {code: 201, message: Created}
In short, always use self.create_random_name
to generate
the name of any Azure resource used in a test.
Also, make sure the correct length is given to the method
because different resources have different limitations on the name length.
The method will always try to create the longest name possible
to fully randomize the name.
from azure.cli.testsdk import ScenarioTest, ResourceGroupPreparer, StorageAccountPreparer
class StorageAccountTests(ScenarioTest):
@ResourceGroupPreparer()
@StorageAccountPreparer()
def test_list_storage_accounts(self, storage_account):
accounts = self.cmd('az storage account list').get_output_in_json()
search = [account for account in accounts if account['name'] == storage_account]
assert len(search) == 1
Note:
- Like
ResourceGroupPreparer
, you can useStorageAccountPreparer
to prepare a disposable storage account for the test. The account is deleted along with the resource group during test teardown. - Creation of a storage account requires a resource group. Therefore
ResourceGroupPrepare
must be placed aboveStorageAccountPreparer
, since preparers are designed to be executed from top to bottom. (The core preparer implementation is in theazure.cli.testsdk.scenario_tests.preparers.AbstractPreparer
.) - The preparers communicate among themselves by adding values to the
kwargs
of the decorated methods. Therefore theStorageAccountPreparer
uses the resource group created in the precedingResourceGroupPreparer
. - The
StorageAccountPreparer
can be further customized to modify the parameters of the created storage account:
@StorageAccountPreparer(sku='Standard_LRS', location='southcentralus', parameter_name='storage')
class StorageAccountTests(ScenarioTest):
@ResourceGroupPreparer()
@StorageAccountPreparer(parameter_name='account_1')
@StorageAccountPreparer(parameter_name='account_2')
def test_list_storage_accounts(self, account_1, account_2):
accounts_list = self.cmd('az storage account list').get_output_in_json()
assert len(accounts_list) >= 2
assert next(acc for acc in accounts_list if acc['name'] == account_1)
assert next(acc for acc in accounts_list if acc['name'] == account_2)
Note:
- Two storage account names will be assigned to different function parameters.
- The resource group name is not needed for the test so the function doesn't have to declare a parameter to accept the name. However it doesn't mean that the resource group is not created. Its name is in the keyword parameter dictionary for any other preparer to consume, but it is removed before the test function is actually invoked.
with self.assertRaisesRegexp(CLIError, "usage error: --vnet NAME --subnet NAME | --vnet ID --subnet NAME | --subnet ID"):
self.cmd('container create -g {rg} -n {container_group_name} --image nginx --vnet {vnet_name}')
The above syntax is the recommended way to test that a specific error occurs. You must pass the type of the error as well as a string used to match the error message. If the error is encountered, the text will be validated and, if matching, the command will be deemed a success (for testing purposes) and execution will continue. If the command does not yield the expected error, the test will fail.
There are a few environment variables that modify the behaviour of the CLI's test infrastructure.
As mentioned previously, AZURE_TEST_RUN_LIVE runs tests live, if it is set to a truthy value (e.g. "True"). It is preferable to use --live
However there are other environment variables that alter the behavior of resource preparers such as the ResourceGroupPreparer
, such as:
AZURE_CLI_TEST_DEV_RESOURCE_GROUP_NAME
andAZURE_CLI_TEST_DEV_RESOURCE_GROUP_LOCATION
: If set, the first environment variable specifies an existing resource group to be returned by theResourceGroupPreparer
. The specified resource group will not be deleted after the test. The second specifies the existing resource group's location, defaults to the location passed when callingResourceGroupPreparer
.AZURE_CLI_TEST_DEV_STORAGE_ACCOUNT_NAME
: Specifies an existing storage account. Associated with theStorageAccountPreparer
.AZURE_CLI_TEST_DEV_KEY_VAULT_NAME
: Specifies an existing key vault name. Associated with theKeyVaultPreparer
.AZURE_CLI_TEST_DEV_SP_NAME
andAZURE_CLI_TEST_DEV_SP_PASSWORD
: Specifies the name of an existing service principal and its password. Associated with theRoleBasedServicePrincipalPreparer
.AZURE_CLI_TEST_DEV_APP_NAME
andAZURE_CLI_TEST_DEV_APP_SECRET
: Specifies an existing managed app and its secret (secret defaults toNone
). Associated with theManagedApplicationPreparer
.
Setting these variables can reduce the time taken to run live tests new resource as the preparers don't have to create and delete new resources.
To learn more about these environment variables, please see the preparers' source code in
/src/azure-cli-testsdk/azure/cli/testsdk/preparers.py
.