Guide¶

Installation¶

Assuming you are running Linux or OS X and are familiar with pip, installation is as easy as:

pip install cosmic

If you are not yet familiar with virtualenv, it is an indespensible tool for Python development. It lets you create isolated Python environments for every project you are working on. This means that different projects can depend on different versions of the same library.

If you would like to work on the bleeding edge of Cosmic development, you can clone the repo using git:

git clone https://github.com/cosmic-api/cosmic.py.git cosmic-py

Then to install the current version (ideally you want to do this in a virtualenv):

cd cosmic-py
python setup.py develop

What’s in an API?¶

A web API is:

An interface through which a server can share data and functionality with clients over the internet.
A component of the server architecture that glues the database and business logic to HTTP.

In the context of Cosmic, an API is represented, unsurprisingly, by an instance of the API class. What is interesting, however, is that this object is serializable to JSON. The JSON form of an API is the API spec.

You may find it strange that we say “serialize an API” when we could simply say “generate an API spec”. The reason we say this is to highlight the fact that an API is simply a Teleport datatype. The API object on the server and the API object on the client are instances of the same class, in fact, they are almost identical. The difference is that for every server endpoint, there is a hook into the server’s database or business logic, whereas each client endpoint replaces this with an HTTP call.

Let’s serialize a trivial API. Note that to_json() is a standard Teleport method:

>>> from cosmic.api import API
>>> trivia = API("trivia", homepage="http://example.com")
>>> API.to_json(trivia)
{
    u'name': 'trivia',
    u'homepage': 'http://example.com',
    u'actions': {u'map': {}, u'order': []},
    u'models': {u'map': {}, u'order': []}
}

Let’s take a look at what’s inside.

First, there is the basic metadata: the API name and homepage. The name of the API should be unique. Though this is not yet enforced by Cosmic, we plan on indexing Cosmic APIs on our website in which case it will become a requirement.

Then, the API spec contains descriptions of actions and models. These will be explained in detail in the next two sections. Here is the Teleport schema for the API type:

schema = Struct([
    required("name", String),
    optional("homepage", String),
    required("actions", OrderedMap(Struct([
        optional("accepts", Schema),
        optional("returns", Schema),
        optional("doc", String)
    ]))),
    required("models", OrderedMap(Struct([
        required("properties", OrderedMap(Struct([
            required(u"schema", Schema),
            required(u"required", Boolean),
            optional(u"doc", String)
        ]))),
        required("links", OrderedMap(Struct([
            required(u"schema", Schema),
            required(u"required", Boolean),
            optional(u"doc", String)
        ]))),
        required("query_fields", OrderedMap(Struct([
            required(u"schema", Schema),
            required(u"required", Boolean),
            optional(u"doc", String)
        ]))),
        required("methods", Struct([
            required("get_by_id", Boolean),
            required("get_list", Boolean),
            required("create", Boolean),
            required("update", Boolean),
            required("delete", Boolean),
        ])),
        required("list_metadata", OrderedMap(Struct([
            required(u"schema", Schema),
            required(u"required", Boolean),
            optional(u"doc", String)
        ])))
    ])))
])

Client and Server¶

In Cosmic, the same API class is used for the API server and the API client. In fact, the server and the client objects behave almost identically. After you run your server component, you can build the client in a single line of code

>>> myapi = API.load('http://localhost:5000/spec.json')

RPC via Actions¶

See also

ActionEndpoint for HTTP spec.

RPC stands for remote procedure call. It allows remote clients to call procedures (functions) in your code. These are commonly implemented as POST handlers on action-style URLs, such as POST /register_user. Cosmic goes along with this convention, listening to POST requests on /actions/<name> URLs.

So what’s in an action? Clearly, we need a name in order to generate the URL. But apart from the name, Cosmic also expects type definitions for the input and output values of the action. These definitions are used for serialization, validation and help with generating documentation. Here is the Teleport schema that describes an action:

Struct([
    optional("accepts", Schema),
    optional("returns", Schema),
    optional("doc", String)
])

Actions are registered with the action() decorator:

>>> from cosmic.types import Array, Integer
>>> @mathy.action(accepts=Array(Integer), returns=Integer)
... def add(numbers):
...     return sum(numbers)
...
>>>

The function used in the action is perfectly usable:

>>> add([1, 2, 3])
6

But now there is another way of accessing it:

>>> mathy.actions.sum([1, 2, 3])
6

And from the client, it is accessed identically:

>>> mathy = API.load('http://localhost:5000/spec.json')
>>> mathy.actions.add([1, 2, 3])
6

If you are not yet familiar with Teleport, you might be wondering what is the purpose of the name and order items in the actions object above. This is the way Teleport uses JSON to represent an ordered mapping. Both actions and models are contained in the Teleport’s OrderedMap type.

Both accepts and returns are optional. If no accepts schema is provided, the action will take no input data, and if the returns schema is not provided, the action will return nothing when it completes.

Normally, the action function is expected to take a single non-keyword argument. If your action needs to take multiple arguments, use the Teleport Struct type:

@mathy.action(accepts=Struct([
    required(u'numerator', Integer),
    required(u'denominator', Integer),
]), returns=Integer)
def divide(numerator, denominator):
    return numerator / denominator

This may be called remotely as:

>>> mathy = API.load('http://localhost:5000/spec.json')
>>> mathy.actions.divide(numerator=10, denominator=5)
2

REST via Models¶

Models are data type definitions attached to an API, they use Teleport schemas to describe their data.

Let’s take a look at the model object:

from cosmic.api import API
from cosmic.models import BaseModel

places = API('places')

@places.model
class Address(BaseModel):
    properties = [
        required(u"number", Integer),
        optional(u"street", String),
        optional(u"city", String)
    ]

As you can see, a model class should inherit from BaseModel and in order to register it with an API, you must use the model() decorator on it. Once a model has been registered with an API, it becomes accessible as part of the models namespace, for example places.models.Address.

Once registered with an API, a model becomes available in the models namespace. The beauty of this namespace is that it is identical on the client and server.

Models can be used to create REST-ful endpoints. A model roughly corresponds to a database table. If you want to give clients access to objects of the data type defined by the model, you also need to define a set of CRUD methods that Cosmic will turn into HTTP endpoints.

The links parameter describes relationships between models. A link from one model to another is similar to a foreign key in a relational database.

Links are defined similarly to properties:

places = API('places')

@places.model
class City(BaseModel):
    properties = [
        optional(u"name", String)
    ]

@places.model
class Address(BaseModel):
    properties = [
        required(u"number", Integer),
        required(u"street", String),
    ]
    links = [
        required(u"city", City)
    ]

These models are merely data type definitions, they do not have REST endpoints because they are not connected to any database.

If apart from defining a data type we also want to provide access to a collection of objects of this data type, there are 5 methods that Cosmic allows us to override. These methods correspond to 5 HTTP endpoints. Methods must be declared by adding their name to the methods property on the model class.

get_by_id¶

See also

GetByIdEndpoint for HTTP spec.

The simplest method to implement is get_by_id(). It takes a single parameter (an id is always a string) and returns a dict representing the object. If the object doesn’t exist, it must raise NotFound.

from cosmic.exceptions import NotFound

places = API('places')

@places.model
class City(BaseModel):
    methods = ["get_by_id", "create", "update", "delete", "get_list"]
    properties = [
        optional(u"name", String)
    ]

    @classmethod
    def get_by_id(cls, id):
        try:
            return cities[id]
        except KeyError:
            raise NotFound

cities = {
    "0": {"name": "Toronto"},
    "1": {"name": "San Francisco"},
}

As you can see, Cosmic doesn’t care what kind of database you use, as long as the method returns the right value. Now if we want to use this method, we can do, on the client or server:

>>> city = places.models.City.get_by_id("1")
{"name": "San Francisco"}

create¶

See also

CreateEndpoint for HTTP spec.

The create() method takes a patch (a model representation where every field is optional) and returns a tuple with the new id and representation:

@classmethod
def create(cls, patch):
    new_id = str(len(cities))
    cities[new_id] = patch
    return new_id, cities[new_id]

update¶

See also

UpdateEndpoint for HTTP spec.

The update() method takes an id and patch and either applies the patch, returning the new representation, or raises NotFound.

@classmethod
def update(cls, id, patch):
    if id not in cities:
        raise NotFound
    cities[id] = patch
    return cities[id]

delete¶

See also

DeleteEndpoint for HTTP spec.

The delete() method, upon deleting the object, returns nothing. It raises NotFound if the object does not exist:

@classmethod
def delete(cls, id):
    if id not in cities:
        raise NotFound
    del cities[id]

get_list¶

See also

GetListEndpoint for HTTP spec.

The get_list() method takes keyword arguments as specified by the query_fields model property. This schema is used to serialize them into a URL query string with the help of URLParams.

query_fields = [
    optional(u"country", String)
]

@classmethod
def get_list(cls, country=None):
    if country is None:
        return cities.items()
    elif country == "Canada":
        return [("0", cities["0"])]
    elif country == "USA":
        return [("1", cities["1"])]
    else:
        return []

The return value of this function is a (possibly empty) list of tuples where the first element is the object id and the second is the object representation.

You are free to invent your own pagination schemes using custom query fields.

Often it will be useful to return metadata along with the items, for example, the total count if the list is paginated, or a timestamp. You can specify this by including the list_metadata attribute.

list_metadata = [
    required(u"total_count", Integer)
]

@classmethod
def get_list(cls):
    metadata = {"total_count": len(cities)}
    return (cities.items(), metadata)

As you can see, when list_metadata is specified, the return value of get_list() is a tuple, where the first item is the list, and the second is a dict containing the metadata.

Serving¶

For development, run() is fine, but for production, you should use a WSGI server such as Gunicorn. In order to do this, use Server to expose the raw WSGI application.

from cosmic.api import API
from cosmic.http import Server
from cosmic.types import *

words = API('words')


@words.action(accepts=String, returns=String)
def pluralize(word):
    if word.endswith('y'):
        return word[:-1] + 'ies'
    else:
        return word + 's'

wsgi_app = Server(words).wsgi_app

Now you can run it in your favorite web server:

$ gunicorn -b 127.0.0.1:5001 words:wsgi_app

Authentication¶

Currently, Cosmic does not provide a standard authentication mechanism. It does provide powerful HTTP hooks which can be used to implement different authentication schemes.

On the server, you can use standard WSGI middleware, and you can subclass Server:

from flask import make_response
from cosmic.api import API
from cosmic.http import Server, error_response

planetarium = API("planetarium")

class CustomServer(Server):

    def view(self, endpoint, request, **url_args):
        if request.headers.get('Authorization', None) != 'secret':
            return error_response("Unauthorized", 401)
        return super(CustomServer, self).view(endpoint, request, **url_args)

wsgi_app = CustomServer(planetarium).wsgi_app

On the client, we can subclass ClientHook to add authentication info to each request:

from cosmic.api import API
from cosmic.http import ClientHook

planetarium = API.load('https://api.planetarium.com/spec.json')

class CustomClientHook(ClientHook):

    def build_request(self, endpoint, *args, **kwargs):
        request = super(ClientHook, self).build_request(endpoint, *args, **kwargs)
        request.headers["Authorization"] = "secret"
        return request

planetarium.client_hook = CustomClientHook(base_url="https://api.planetarium.com")

Storing Global Data¶

In every web application some data must be available globally during request processing, for example, the database connection or the currently authenticated user. Some frameworks, like Django, attach this data to the request object which gets passed around explicitly. Others, like Flask, store it in a thread-local object. Cosmic borrows the latter approach, offering you a simple dictionary-like class for this purpose: SafeGlobal.

from cosmic.globals import SafeGlobal

g = SafeGlobal()

Now we can use it to store the current user:

class CustomServer(Server):

    def view(self, endpoint, request, **url_args):
        secret = request.headers.get('Authorization', None)
        if secret == '12345':
            g['current_user'] = 'bob@example.com'
        elif secret == 'qwert':
            g['current_user'] = 'alice@example.com'
        else:
            return error_response("Unauthorized", 401)
        return super(CustomServer, self).view(endpoint, request, **url_args)

For testing, it may be necessary to call some functions with a predefined context, for example, call a function on behalf of Bob. For this, use the scope() method:

with g.scope({'current_user': 'bob@example.com'}):
    assert get_account_balance() == 100

Guide¶

Installation¶

What’s in an API?¶

Client and Server¶

RPC via Actions¶

REST via Models¶

get_by_id¶

create¶

update¶

delete¶

get_list¶

Serving¶

Authentication¶

Storing Global Data¶

Table Of Contents

Related Topics

This Page