Test Driven Development of a Horizontally Scaling REST-Service with Pyramids, Crate and SQLAlchemy

Author
admin
Filed under
Date
August 11, 2014

Click here to download CrateDB

Download Now »

This blog post is a recap of a tutorial held by Bernd Dorn at EuroPython2014. The resulting sample web application that is publicly available on Github. In this tutorial we could cover certain aspects in much greater detail, but that would blow up a blog post to an unreadable length. So it focuses on the sandboxed test driven development rather than explaining how to build a Pyramids app. This post should also document the sample application in a general holistic manner for those who could not attend the EuroPython session. In this post we will explain how to develop a simple web application in a sandboxed environment utilising integration tests and using the following tools:

To follow and try out the example in this post you will need to use a Mac OS or Linux operating system and have Python 3.4 and Java 8 installed. Basic Python skills and knowledge in web technologies should be present. Before we start, you’d probably like to checkout the sample app.

git clone https://github.com/dobe/ep2014tutorial
cd ep2014tutorial

Creating a Full-Stack Development Sandbox

One of the many things we like about Python is the ability to create sandboxed environments really easily. Although there are quite a few solutions out there, our weapon of choice is still (since the early days) the zc.buildout system, or short: Buildout.

“Buildout is a Python-based build system for creating, assembling and deploying applications from multiple parts, some of which may be non-Python-based.” (Source: buildout.org)

This is exactly what we want when we combine multiple components in a single full-stack environment. Essentially, there are 2 commands to be remembered to bootstrap and build our environment:

python3.4 bootstrap.py
bin/buildout

The bootstrap.py file is included in this project, but for your own projects you can obtain it via:

wget http://downloads.buildout.org/2/bootstrap.py

The buildout command will then install/build all requirements for the projects. Before we do so, let’s look at the configuration file for buildout: the buildout.cfg located at project root.

Buildout Parts

Each section in the buildout.cfg describes a part that needs to be built for the project. Although some of you might be familiar with buildout configurations I want to point out quickly what each section does.

[buildout]
extends = versions.cfg
develop = .
parts = test
        crate-cmd
        scripts
versions = versions
show-picked-versions = true

The first part of the configuration is always the [buildout] part which defines what parts are built, what versions of Python packages should be used and where the packages can be found. For our sample app we require the test suite, the crate command and the scripts section.

[scripts]
recipe = zc.recipe.egg:script
relative-paths = true
scripts = app
eggs = ep2014tutorial
initialization =
  import sys
  sys.argv[1:1] = ['${buildout:directory}/etc/app.ini']

The scripts section defines the startup script for the Pyramid app. It generates an executable called app in the bin directory of the project using the entry point specified by the console_scripts key in setup.py You can take a look at the bin/app script which is just an ordinary executable Python file to find out how the other options in this section are used. At the end of this post we will use the script to start the app and see it in action.

[test]
relative-paths = true
recipe = zc.recipe.testrunner
defaults = ['--auto-color']
eggs = ep2014tutorial[test]

The zc.recipe.testrunner recipe will do everything necessary to setup the test runner for us. We just specify the eggs to be used and the coloured output. Done. Note here, that we use the test extras of the ep2014tutorial egg so the crate[test] extras are loaded as dependencies (see Test Layer).

[crate]
recipe = hexagonit.recipe.download
url = https://cdn.crate.io/downloads/releases/crate-0.40.2.tar.gz
strip-top-level-dir = true

The [crate] directive is responsible for downloading the Crate binary and extract it into the parts folder. You may recognise, that this section is not listed of the parts option of the buildout section. This is because it is a dependency of the next section …

[crate-cmd]
recipe = collective.recipe.template
output = ${buildout:bin-directory}/crate
mode = 755
inline =
  #!/bin/bash
  ${crate:location}/bin/crate -Des.config='${crate-yml:output}' $@

To conveniently start the Crate binary that is downloaded somewhere in the parts folder of the project, the [crate-cmd] generates a crate script inside the bin directory, referencing the original location of the Crate executable and adding the path to the crate.yml as a permanent argument. However any other command line arguments may be passed.

[crate-yml]
recipe = collective.recipe.template
output = ${buildout:directory}/etc/crate.yml
inline =
  path:
    logs: ${buildout:directory}/var/logs
    data: ${buildout:directory}/var/crate
  cluster.name: tutorial

This section is also only referenced from the [crate-cmd] section. It will write the crate.yml configuration file that is required for Crate. If you haven’t run the buildout command yet, you should do it now. With a single command we are set up and ready to code in a sandboxed full-stack environment.

Setup Integration Tests

Now that our environment is ready we can start developing our application.

Relevant Files

src/ep2014tutorial/tests.py
src/ep2014tutorial/probe_status.rst
src/ep2014tutorial/views.rst
src/ep2014tutorial/testing/
src/ep2014tutorial/testing/__init__.py
bin/test

Runner, Layer, Suite

Integration tests consist of 3 components: the test runner, test layers and test suites. The test runner is responsible for parsing and evaluating tests. We use the Zope testrunner as described previously in this post. A test layer is the setup of an environment in which tests run in. The crate Python package does include its own test layer that we can use for our own tests. This requirement is specified in setup.py

setup(
    ...
    extras_require=dict(
        test=['crate[test]', 'webtest']
    ),
    ...
)

In the ep2014tutorial.testing module we can import and instantiate a reusable test layer instance.

from crate.testing.layer import CrateLayer
crate_layer = CrateLayer('crate',
                         crate_home=crate_path(),
                         port=44200)

A test suite is a collection of doctest files with their test parameters, e.g. what layer it should use. The suite is created in the tests.py file.

def test_suite():
    suites = (
        create_suite('probe_status.rst', layer=testing.crate_layer),
        create_suite('views.rst', layer=testing.crate_layer),
    )
    return unittest.TestSuite(suites)

The functions setUp and tearDown are called before and after running a test suite to create and destroy the test environment.

setup(
    ...
    extras_require=dict(
        test=['crate[test]', 'webtest']
    ),
    ...
)

In the ep2014tutorial.testing module we can import and instantiate a reusable test layer instance.

from crate.testing.layer import CrateLayer
crate_layer = CrateLayer('crate',
                         crate_home=crate_path(),
                         port=44200)

A test suite is a collection of doctest files with their test parameters, e.g. what layer it should use. The suite is created in the tests.py file.

def test_suite():
    suites = (
        create_suite('probe_status.rst', layer=testing.crate_layer),
        create_suite('views.rst', layer=testing.crate_layer),
    )
    return unittest.TestSuite(suites)

The functions setUp and tearDown are called before and after running a test suite to create and destroy the test environment.

def setUp():
    ...

The setUp method for example includes creating the test client, setting the global test variable and creating the table layout for the database using the CREATE TABLE statement in schema.sql (the schema matches with the model described later on).

def tearDown():
    ...

To clean the database when tests are finished the relevant tables are dropped in the tearDown method.

Writing Integration Tests

There are many ways of doing test driven development (TDD), some of which are very compelling. Our approach to TDD however is very practical. Tests can/should be written before actually starting to code the implementation. This is possible, because we know what the result of certain calls – response of a REST HTTP call for example – needs to look like.

The integration tests themselves work the same way as regular Python doctests, except they are written inside their own file using reStructuredText syntax, so they can be used to build documentation using Sphinx. Keep that in mind when you write tests!

There are 2 files in the test suite, one to test the probe_status endpoint and one to test the remaining REST endpoints of our app. The latter is named views.rst.

We can use the test client that we declared as a global test variable in the setUp() step of our environment to emulate requests to our application.

Our application stores tweets and exposes the latest ones in JSON format on a given endpoint. It is also possible to limit the amount of returned tweets.

First example: There are no tweets in the database yet, so the result must be an empty array.

>>> pprint(app.get('/latest').json)
[]

Second example: Create a new Tweet instance and store it in the database (using SQLAlchemy).

>>> from ep2014tutorial import model
>>> model.DBSession.add(model.Tweet(
...              id='1',
...              created_at=datetime(2014, 7, 23, 14, 0, 0),
...              text='tweet1',
...              source='test'
... ))
>>> refresh()

Calling the endpoint again, we must receive a JSON array with a single item.

>>> pprint(app.get('/latest').json)
[{'created_at': '2014-07-23T14:00:00', 'id': '1', 'text': 'tweet1'}]

There are two more tests in views.rst that you can look at. Listing them all would exceed the scope of this post. However, I hope you got the idea. Now we do have tests for our application without writing the implementation yet!

Programming the Service Endpoints

Relevant Files

src/ep2014tutorial/model.py
src/ep2014tutorial/server.py
src/ep2014tutorial/views.py
bin/app

In a recent blog post we wrote about how to use Crate with SQLAlchemy in Python applications. If you are new to SQLAlchemy you probably want to read his post first.

The Model

class Tweet(Base):

    __tablename__ = 'tweets'

    id = Column(String, primary_key=True)
    created_at = Column(DateTime)
    text = Column(String)
    source = Column(String)
    retweeted = Column(Boolean)
    user = Column(Object)

    def __json__(self, request):
        # provide the Tweet as a JSON serializable dict
        res = dict(id=self.id,
            text=self.text,
            created_at=self.created_at.isoformat()
        )
        userId = self.user and self.user.get('id')
        if userId:
            res['user'] = userId
        return res

Our application uses a single model Tweets which inherits from the SQLAlchemy Base.

The Pyramids app factory initializes the database session and binds the engine to the Base.

def app_factory(global_config, **settings):
    ...
    db_init(config)

def db_init(config):
    settings = config.get_settings()
    engine = engine_from_config(settings, 'sqlalchemy.')
    DBSession.configure(bind=engine)
    Base.metadata.bind = engine

The Views

Views and their routing live inside views.py.

To create a view we define a method latest_tweets and use the Pyramids decorator @view_config with the appropriate route name.

@view_config(route_name='latest')
def latest_tweets(request, request_method='GET'):
    return []

Additionally we have to add a route and a renderer for the view so we can return a simple object in the view code instead of a response object.

def includeme(config):
    config.add_route("latest", "/latest")
    config.add_renderer(None, JSON(indent=4))

The includeme() method is called within the app factory where ep2014tutorial.views is added to the configurator.

Now the logic for obtaining the limit parameter from the URL and the database access can be added. list() will resolve and execute the SQL statement built with the SQLAlchemy ORM and return a list of Tweet objects. In order to make this list JSON serialisable we added the __json__ method to the Tweets class.

@view_config(route_name='latest')
def latest_tweets(request, request_method='GET'):
    limit = int(request.GET.get('limit', 10))
    query = DBSession.query(Tweet).order_by(
        Tweet.created_at.desc(), Tweet.id.desc()
    ).limit(limit)
    return list(query)

There is a second view in the sample application, too. However, the process is exactly the same as with the latest_tweets view and it is up to you to try it out.

Wire up the application

We briefly looked at the app_factory method that is used by Pyramid already when we looked at how the model is connected with the database engine.

def app_factory(global_config, **settings):
    config = Configurator(settings=settings)
    config.include('ep2014tutorial.views')
    config.scan('ep2014tutorial.views')
    db_init(config)
    return config.make_wsgi_app()

he app factory returns a WSGI application with the base configuration defined in the etc/app.ini file and enriched with the routing and view from the ep2014tutorial.views module.

[app:main]
use = egg:ep2014tutorial
crate.hosts = localhost:4200
crate.echo = True
sqlalchemy.url = crate://localhost:4200

The same configuration is also used to start the test server for the test environment.

Now that everything is chained together we can run the bin/test command and see if the app works correctly and as intended.

Finally it is also time to start the application using the bin/app command and the Crate instance using the bin/crate command.

Building a Cluster

To simulate the deployment of the application you can checkout the same Github example on 2 different machines on the same network and start both the app and the crate instance.

However, if you don’t have a second machine, you can checkout the project twice and run the app and Crate on different ports, e.g. the app on 8080 and 8081 and Crate on 4200 and 4201. You simply need to change the app.ini file to:

[app:main]
use = egg:ep2014tutorial
crate.hosts = localhost:4201
crate.echo = True
sqlalchemy.url = crate://localhost:4201

[server:main]
use = egg:waitress
port = 8081
host = 127.0.0.1

If you want you can even put an NGINX load balancer in front using both instances as upstream servers.

For more detailed information how to run and deploy an application with Crate you should check out our other blog post Using Crate Date in a Multi Node Setup.

Conclusion

It is fairly easy to bootstrap and build your application and all components needed in a sandboxed environment using the buildout build system. It’s even easily extendible. Once you’ve written the boilerplate code – such as test layers – you can start writing (integration) tests to test your code. The level of detail is up to you! For applications where the logic inside views is relatively simple, it is helpful to write integration tests before touching the actual implementation. Last but not least, deploying your app is as simple as developing locally, using the same methods and flows.

We’d love to hear how you develop your Crate projects. Let us know!

Back to topAll Blog