In a previous post I introduced and discussed about an open source library I authored called Servant. Servant is a framework for building RPC-style services with Python. Servant is to services as Django or Rails is to web applications…it’s framework you’d use when you need to build a service in Python.

I’ll go a bit deeper in this post and show what it’s like to actually author and use a service. Let’s first start by writing the service. To save some time we’ll take a look at the calculator_service which is in the examples directory in the Servant repository.

Writing the service


There are a few requirements to write a service. I’ll start with the boring parts which is mostly boilerplate to get your service setup as an install-able Python package. I also suggest looking at the directory structure for the demo calculator_service on Github:

At a bare minimum you’ll need a file and a directory with the correct name. In this example, the name is calculator_service.

From there you’ll also need a directory where your source code will live.


Every Servant service needs an entry-point to device the endpoints which will be exposed. This is pretty simple…create a file names which contains a class that subclasses servant.service.base.Service

Your service class will only need a few things:

  • name attribute
  • version attribute (use 1 when getting started)
  • action_map which is a dictionary mapping endpoint name to an Action class

That’s really it…of course there are more things you can do in your service class to handle configuration, etc. However, this is the minimum amount of work you’ll need to do in order to get started.

In our example you can see our calculator_service exposed three calculate endpoints…add, subtract and divide. These will do what you’d expect.


An Action is really the meat and potatoes for a particular endpoint. You can organize your actions however you like…one action per file, multiple actions per file, etc. Since we’re writing a pretty simple service and only have three actions, we’ll put them all in

Let’s walk through what it takes to implement a single action (or, an “endpoint” from the client’s perspective). Looking at AddAction we can see that there are two required inputs…number1 and number2. I liken the structure of a servant Action to a Django, SQLAlchemy or other type of model. You’ll need to subclass servant.service.actions.Action and usually define one or more inputs as class attributes which are servant field types.

In AddAction we ensure that the client passes us number1 and number2 by adding required=True to the field’s kwargs. This instructs servant to return an error in the case that the client doesn’t pass either of these. Out of the box we get some error checking for free which is quite nice. From there it’s also your responsibility as the developer to device the return type which the client will get back after a successful call. Here, we’ll return a single integer using an IntField field type which is named result.

Your entry point into an action is the run method. You can see above what it does…it’s simple one-liner. By the time your run method is being executed the basic error checking has completed and all of the required and optional fields will be available to you as class attributes. So, our run method simple sets the result using the two inputs we received from the client:

def run(self, **kwargs):
    self.result = self.number1 + self.number2

The run method doesn’t need to return anything…all return data is handled via the field attributes.

If you read the rest of the gist above you can see how the subtract and divide actions are implemented. Being a calculator, there really isn’t much to it. You’ll notice that the subtract action subclassed the add action and just overrides the run method. The divide action mostly just changes the names of the inputs and output.

Using the service

One neat feature of Servant is the ability to run your service as a locally installed Python library.

Let’s install both Servant and the calculator service into a new virtualenv:

$ cd path/to/servant
$ mkvirtualenv calculator_service
$ pip install -e .
$ cd examples/calculator_service
$ pip install -e .

Now that we have these installed, we can actually test our service!

import servant.client

client = servant.client.Client('calculator_service', version=1)

response = client.add(number1=10, number2=15)

if response.is_error():
    print response.errors, response.field_errors
    print response.result

When we run this, we get our expected result of 25.

Now, how do we run this on a separate host so that we’re actually a stand-alone service? Quite easy…first we’ll setup a wsgi server using uwsgi. We need two files…one is a 3-line python file which is the entry-point for the service. The other is a very short uwsgi config file. Servant has built-in support for running as a wsgi server so is quite short:

Now, fire up uwsgi: uwsgi uwsgi.ini

uwsgi is now ready to serve requests on port 8888…we can run the exact same client code as above. You’ll note that the client code above only depends on the servant library. Really, the only reference to the calculator_service is when we instantiate the Client class. There is only one change we need to make to the client code above to hit a remote system:

client.configure('http', host='', port=8888)

We’re configuring our client to connect via http to the given host and port. Once that single line is added we can run the client and get the exact same results.

I believe the power of being able to develop a service and run it locally like this and deploy it remotely with only a single line changing for the client is immensely powerful. Admittedly, when using Docker standing up a server is pretty easy, but being able to run service code without depending on a server is still quite valuable. Another big advantage is that it’s possible to actually pip install your service package with your client application and not depend on a remote system staying up. The advantage is by pip installing a service is that you’re able to deploy a service much faster and iterate simply by pip installing…when it’s time to migrate it’s just a matter of a minor change to your client code to point to the remote host.