Local Unit Testing for Python

Unit testing allows you to check the quality of your code after you've written it, but you can also use unit testing to improve your development process as you go along. Instead of writing tests after you finish developing your application, consider writing the tests as you go. This helps you design small, maintainable, reusable units of code. It also makes it easier for you to test your code thoroughly and quickly.

When you do local unit testing, you run tests that stay inside your own development environment without involving remote components. App Engine provides testing utilities that use local implementations of datastore and other App Engine services. This means you can exercise your code's use of these services locally, without deploying your code to App Engine, by using service stubs.

A service stub is a method that simulates the behavior of the service. For example, the datastore service stub shown in Writing Datastore and Memcache Tests allows you to test your datastore code without making any requests to the real datastore. Any entity stored during a datastore unit test is held in memory, not in the datastore, and is deleted after the test run. You can run small, fast tests without any dependency on datastore itself.

This document describes how to write unit tests against several local App Engine services, then gives some information about setting up a testing framework.

  1. Introducing the Python testing utilities
  2. Writing Datastore and memcache tests
  3. Writing High Replication Datastore tests
  4. Writing mail tests
  5. Changing the default environment variables
  6. Setting up a testing framework

Introducing the Python testing utilities

An App Engine Python module called testbed makes service stubs available for unit testing. The testbed module was inspired by GAE Testbed, which was designed by JJ Geewax.

To write a test that uses testbed, you need to create and activate a Testbed instance, then declare the service stubs that you want to use. For an example, see Writing Datastore and Memcache Tests.

Service stubs are available for the following services:

  • App Identity init_app_identity_stub
  • Blobstore (use init_blobstore_stub)
  • Capability (use init_capability_stub)
  • Channel (use init_channel_stub)
  • Datastore (use init_datastore_v3_stub)
  • Files (use init_files_stub)
  • Images (only for dev_appserver; use init_images_stub)
  • LogService (use init_logservice_stub)
  • Mail (use init_mail_stub)
  • Memcache (use init_memcache_stub)
  • Task Queue (use init_taskqueue_stub)
  • URL fetch (use init_urlfetch_stub)
  • User service (use init_user_stub)
  • XMPP (use init_xmpp_stub)

To initialize all stubs at the same time, you can use init_all_stubs.

Writing Datastore and memcache tests

This section shows an example of how to write code that tests the use of the datastore and memcache services.

First, import the Python unittest module and the App Engine modules that are relevant to the services being tested—in this case memcache and ndb, which uses both the datastore and memcache. Also import the App Engine testbed module.

import unittest
from google.appengine.api import memcache
from google.appengine.ext import ndb
from google.appengine.ext import testbed

Then create a TestModel class. In this example, a function checks to see whether an entity is stored in memcache. If no entity is found, it checks for an entity in the datastore. (This may often be redundant in real life, since ndb uses memcache itself behind the curtains, but it's still an OK pattern for a test).

class TestModel(ndb.Model):
  """A model class used for testing."""
  number = ndb.IntegerProperty(default=42)
  text = ndb.StringProperty()

class TestEntityGroupRoot(ndb.Model):
  """Entity group root"""

def GetEntityViaMemcache(entity_key):
  """Get entity from memcache if available, from datastore if not."""
  entity = memcache.get(entity_key)
  if entity is not None:
    return entity
  key = ndb.Key(urlsafe=entity_key)
  entity = key.get()
  if entity is not None:
    memcache.set(entity_key, entity)
  return entity

Next, create a test case. No matter what services you are testing, the test case must create a Testbed instance and activate it. The test case must also initialize the relevant service stubs, in this case using init_datastore_v3_stub and init_memcache_stub. (The methods for initializing other App Engine service stubs are listed in Introducing the Python Testing Utilities.)

class DemoTestCase(unittest.TestCase):

  def setUp(self):
    # First, create an instance of the Testbed class.
    self.testbed = testbed.Testbed()
    # Then activate the testbed, which prepares the service stubs for use.
    # Next, declare which service stubs you want to use.

The init_datastore_v3_stub() method with no argument uses an in-memory datastore that is initially empty. If you want to test an existing datastore entity, include its pathname as an argument to init_datastore_v3_stub().

In addition to setUp(), include a tearDown() method that deactivates the testbed. This restores the original stubs so that tests do not interfere with each other.

def tearDown(self):

Then implement the tests.

def testInsertEntity(self):
    self.assertEqual(1, len(TestModel.query().fetch(2)))

Now you can use TestModel to write tests that use the datastore or memcache service stubs instead of using the real services.

For example, the method shown below creates two entities: the first entity uses the default value for the number attribute (42), and the second uses a nondefault value for number (17). The method then builds a query for TestModel entities, but only for those with the default value of number.

After retrieving all matching entities, the method tests that exactly one entity was found, and that the number attribute value of that entity is the default value.

def testFilterByNumber(self):
    root = TestEntityGroupRoot(id="root")
    TestModel(number=17, parent=root.key).put()
    query = TestModel.query(ancestor=root.key).filter(TestModel.number == 42)
    results = query.fetch(2)
    self.assertEqual(1, len(results))
    self.assertEqual(42, results[0].number)

As another example, the following method creates an entity and retrieves it using the GetEntityViaMemcache() function that we created above. The method then tests that an entity was returned, and that its number value is the same as for the previously created entity.

def testGetEntityViaMemcache(self):
    entity_key = TestModel(number=18).put().urlsafe()
    retrieved_entity = GetEntityViaMemcache(entity_key)
    self.assertNotEqual(None, retrieved_entity)
    self.assertEqual(18, retrieved_entity.number)

And finally, invoke unittest.main().

if __name__ == '__main__':

Writing High Replication Datastore tests

If your app uses the High Replication Datastore (HRD), you may want to write tests that verify your application's behavior in the face of eventual consistency. db.testbed exposes options that make this easy:

import unittest
from google.appengine.api import memcache
from google.appengine.ext import ndb
from google.appengine.ext import testbed
from google.appengine.datastore import datastore_stub_util

class DemoTestCase(unittest.TestCase):

  def setUp(self):
    # First, create an instance of the Testbed class.
    self.testbed = testbed.Testbed()
    # Then activate the testbed, which prepares the service stubs for use.
    # Create a consistency policy that will simulate the High Replication consistency model.
    self.policy = datastore_stub_util.PseudoRandomHRConsistencyPolicy(probability=0)
    # Initialize the datastore stub with this policy.
    # Initialize memcache stub too, since ndb also uses memcache

  def tearDown(self):

  def testEventuallyConsistentGlobalQueryResult(self):
    class TestModel(ndb.Model):

    user_key = ndb.Key('User', 'ryan')
    # Put two entities
    ndb.put_multi([TestModel(parent=user_key), TestModel(parent=user_key)])

    # Global query doesn't see the data.
    self.assertEqual(0, TestModel.query().count(3))
    # Ancestor query does see the data.
    self.assertEqual(2, TestModel.query(ancestor=user_key).count(3))

if __name__ == '__main__':

The PseudoRandomHRConsistencyPolicy class lets you control the likelihood of a write applying before each global (non-ancestor) query. By setting the probability to 0%, we are instructing the datastore stub to operate with the maximum amount of eventual consistency. Maximum eventual consistency means writes will commit but always fail to apply, so global (non-ancestor) queries will consistently fail to see changes. This is of course not representative of the amount of eventual consistency your application will see when running in production, but for testing purposes, it's very useful to be able to configure the local datastore to behave this way every time. If you use a non-zero probability, PseudoRandomHRConsistencyPolicy makes a deterministic sequence of consistency decisions so test outcomes are consistent:

class DemoTestCase(unittest.TestCase):

  # ...

  def testDeterministicOutcome(self):
    self.policy.SetProbability(.5)  # 50% chance to apply.
    self.policy.SetSeed(2) # Use the pseudo random sequence derived from seed=2.

    class TestModel(ndb.Model):


    self.assertEqual(0, TestModel.query().count(3))
    self.assertEqual(0, TestModel.query().count(3))
    # Will always be applied before the third query.
    self.assertEqual(1, TestModel.query().count(3))

The testing APIs are useful for verifying that your application behaves properly in the face of eventual consistency, but please keep in mind that the local High Replication read consistency model is an approximation of the production High Replication read consistency model, not an exact replica. In the local environment, performing a get() of an Entity that belongs to an entity group with an unapplied write will always make the results of the unapplied write visible to subsequent global queries. In production this is not the case.

Writing mail tests

You can use the mail service stub to test the mail service. Similar to other services supported by testbed, at first you initialize the stub, then invoke the code which uses the mail API, and finally test whether the correct messages were sent.

import unittest
from google.appengine.api import mail
from google.appengine.ext import testbed

class MailTestCase(unittest.TestCase):

  def setUp(self):
    self.testbed = testbed.Testbed()
    self.mail_stub = self.testbed.get_stub(testbed.MAIL_SERVICE_NAME)

  def tearDown(self):

  def testMailSent(self):
                   subject='This is a test',
                   body='This is a test e-mail')
    messages = self.mail_stub.get_sent_messages(to='')
    self.assertEqual(1, len(messages))
    self.assertEqual('', messages[0].to)


Changing the default environment variables

App Engine services often depend on environment variables. The activate() method of class testbed.Testbed uses default values for these, but you can set custom values based on your testing needs with the setup_env method of class testbed.Testbed.

For example, let's say you have a test that stores several entities in datastore, all of them linked to the same application ID. Now you want to run the same tests again, but using an application ID different from the one that is linked to the stored entities. To do this, pass the new value into self.setup_env() as app_id.

For example:

def setUp(self):
    self.testbed = testbed.Testbed()

Another frequent use for setup_env is to simulate a user being logged in, either with or without admin privileges, to check if your handlers operate properly in each case.

def setUp(self):
    self.testbed = testbed.Testbed()
    # other initialization if and as required

def simulate_login(self, user_email='', user_id='', is_admin=False):
        user_is_admin='1' if is_admin else '0',

Now, your test methods can call, for example, self.simulate_login() to simulate no user being logged in, self.simulate_login('', '123') to simulate a non-admin user being logged in, self.simulate_login('', '123', is_admin=True) to simulate an admin user being logged in.

Setting up a testing framework

The SDK's testing utilities are not tied to a specific framework. You can run your unit tests with any available App Engine testrunner, for example nose-gae or ferrisnose. You can also write a simple testrunner of your own, or use the one shown below.

The following scripts use Python's unittest module.

You can name the script anything you want. When you run it, provide the path to your Google Cloud SDK or Google App Engine SDK installation and the path to your test modules. The script will discover all tests in the path provided and will print results to the standard error stream.

import optparse
import sys
import unittest
import os

Run unit tests for App Engine apps.

SDK_PATH    Path to Google Cloud or Google App Engine SDK installation, usually ~/google_cloud_sdk
TEST_PATH   Path to package containing test modules"""

def main(sdk_path, test_path):
    # If the sdk path points to a google cloud sdk installation
    # then we should alter it to point to the GAE platform location.
    if os.path.exists(os.path.join(sdk_path, 'platform/google_appengine')):
      sys.path.insert(0, os.path.join(sdk_path, 'platform/google_appengine'))
      sys.path.insert(0, sdk_path)

    # Ensure that the google.appengine.* packages are available
    # in tests as well as all bundled third-party packages.
    import dev_appserver

    # Loading appengine_config from the current project ensures that any
    # changes to configuration there are available to all tests (e.g.
    # sys.path modifications, namespaces, etc.)
      import appengine_config
    except ImportError:
      print "Note: unable to import appengine_config."

    # Discover and run tests.
    suite = unittest.loader.TestLoader().discover(test_path)

if __name__ == '__main__':
    parser = optparse.OptionParser(USAGE)
    options, args = parser.parse_args()
    if len(args) != 2:
        print 'Error: Exactly 2 arguments required.'
    SDK_PATH = args[0]
    TEST_PATH = args[1]