1. Jesper Nøhr
  2. django-piston


Clone wiki

django-piston / Documentation

Piston Documentation

Getting Started

Getting started with Piston is easy. Your API code will look and behave just like any other Django application. It will have an URL mapping and handlers defining resources.

To get started, it is recommended that you place your API code in a separate folder, e.g. 'api'.

Your application layout could look like this:


Then, define a "namespace" where your API will live in your top-level urls.py, like so:

urlpatterns = patterns('',
   # all my other url mappings
   (r'^api/', include('mysite.api.urls')),

This will include the API's urls.py for anything beginning with 'api/'.

Next up we'll look at how we can create resources and how to map URLs to them.


A "Resource" is an entity mapping some kind of resource in code. This could be a blog post, a forum or even something completely arbitrary.

Let's start out by creating a simple handler in handlers.py:

from piston.handler import BaseHandler
from myapp.models import Blogpost

class BlogpostHandler(BaseHandler):
   allowed_methods = ('GET',)
   model = Blogpost   

   def read(self, request, post_slug):

Piston lets you map resource to models, and by doing so, it will do a lot of the heavy lifting for you.

A resource can be just a class, but usually you would want to define at least 1 of 4 methods:

read is called on GET requests, and should never modify data (idempotent.)

create is called on POST, and creates new objects, and should return them (or rc.CREATED.)

update is called on PUT, and should update an existing product and return them (or rc.ALL_OK.)

delete is called on DELETE, and should delete an existing object. Should not return anything, just rc.DELETED.

In addition to these, you may define any other methods you want. You can use these by including their names in the fields directive, and by doing so, the function will be called with a single argument: The instance of the model. It can then return anything, and the return value will be used as the value for that key.

NB: These "resource methods" should be decorated with the @classmethod decorator, as they will not always receive an instance of itself. For example, if you have a UserHandler defined, and you return a User from another handler, you will not receive an instance of that handler, but rather the UserHandler.

Since a single handler can be responsible for both single- and multiple-object data sets, you can differentiate between them in the read() method like so:

from piston.handler import BaseHandler
from myapp.models import Blogpost

class BlogpostHandler(BaseHandler):
   allowed_methods = ('GET',)
   model = Blogpost   

    def read(self, request, blogpost_id=None):
        Returns a single post if `blogpost_id` is given,
        otherwise a subset.

        base = Blogpost.objects
        if blogpost_id:
            return base.get(pk=blogpost_id)
            return base.all() # Or base.filter(...)


Emitters are what spews out the data, and are the things responsible for speaking YAML, JSON, XML, Pickle and Django. They currently reside in emitters.py as XMLEmitter, JSONEmitter, YAMLEmitter, PickleEmitter and DjangoEmitter.

Writing your own emitters is easy, all you have to do is create a class that subclasses Emitter and has a render method. The render method will receive 1 argument, 'request', which is a copy of the request object, which is useful if you need to look at request.GET (like defining callbacks, like the JSON emitter does.)

To get the data to serialize/render, you can call self.construct() which always returns a dictionary. From there, you can do whatever you want with the data and return it (as a unicode string.)

NB: New in 23ebc37c78e8 : Emitters can now be registered with the Emitter.register function, and can be removed (in case you want to remove a built-in emitter) via the Emitter.unregister function.

The built-in emitters are registered like so:

class JSONEmitter(Emitter):

Emitter.register('json', JSONEmitter, 'application/json; charset=utf-8')

If you write your own emitters, you can import Emitter and call 'register' on it to put your emitter into action. You can also overwrite built-in, or existing emitters, by using the same name (the first argument.)

This makes it very easy to add support for extended formats, like protocol buffers or CSV.

Emitters are accessed via the ?format GET argument, e.g. '/api/blogposts/?format=yaml', but since 23ebc37c78e8 , it is now possible to access them via a special keyword argument in your URL mapping. This keyword is called 'emitter_format' (to not clash with your own 'format' keyword), and can be used like so:

urlpatterns = patterns('',
   url(r'^blogposts(\.(?P<emitter_format>.+))$', ...),

Now a request for /blogposts.json will use the JSON emitter, etc.

Additionally, you may specify the format in your URL mapping, via the keyword arguments shortcut:

urlpatterns = patterns('',
   url(r'^blogposts$', resource_here, { 'emitter_format': 'json' }),

Mapping URLs

URL mappings in Piston work just like they do in Django. Lets map our BlogpostHandler:

In urls.py:

from django.conf.urls.defaults import *
from piston.resource import Resource
from mysite.myapp.api.handlers import BlogpostHandler

blogpost_handler = Resource(BlogpostHandler)

urlpatterns = patterns('',
   url(r'^blogpost/(?P<post_slug>[^/]+)/', blogpost_handler),
   url(r'^blogposts/', blogpost_handler),

Now any request coming in to /api/blogpost/some-slug-here/ or /api/blogposts/ will map to BlogpostHandler, with the two different data sets being differentiated in the handler itself. Note that a single handler can be used both for single-object and multiple-object resources.

Anonymous Resources

Resources can also be "anonymous". What does this mean? This is a special type of resource you can instantiate, and it will be used for requests that aren't authorized (via OAuth, Basic or any authentication handler.)

For example, if we look at our BlogpostHandler from earlier, it might be interesting to offer anonymous access to posts, although we don't want to allow anonymous users to create/update/delete posts. Also, we don't want to expose all the fields authorized users see.

This can be done by creating another handler, inheriting AnonymousBaseHandler (instead of BaseHandler.) This also takes care of the heavy lifting for you.

Like so:

from piston.handler import AnonymousBaseHandler, BaseHandler

class AnonymousBlogpostHandler(AnonymousBaseHandler):
   model = Blogpost
   fields = ('title', 'content')

class BlogpostHandler(BaseHandler):
   anonymous = AnonymousBlogpostHandler
   # same stuff as before 

You don't need a "proxy handler" subclassing BaseHandler to use anonymous handlers, you can just point directly at an anonymous resource as well.

Working with Models

Piston allows you to tie to a model, but does not require it. The benefit you get from doing so, will become obvious when you work with it:

  • If you don't override read/create/update/delete it provides sensible defaults (if the method is allowed by allowed_methods of course.)
  • You don't have to specify fields or exclude (but you still can, they aren't mutually exclusive!)
  • By using a model in a handler, Piston will remember your fields/exclude directives and use them in other handlers who return objects of that type (unless overridden.)

As we've seen earlier, tying to a model is as simple as setting the model class variable on a handler.

Also see: Why does Piston use fields from previous handlers

Configuring Handlers

Handlers can be configured with 4 different variables.


The model to tie to. See Working with Models.


A list of fields to include or exclude. Accepts nested listing, and follows foreign keys and manytomany fields. Also accepts compiled regular expressions. E.g.:

import re

class FooHandler(BaseHandler):
   fields = ('title', 'content', ('author', ('username', 'first_name')))
   exclude = ('id', re.compile('^private_'))

If User can access posts via a Many2many/ForeignKey fields then:

class UserHandler(BaseHandler):
      model = User
      fields = ('name', ('posts', ('title', 'date')))

will show the title and date from a users posts.

To use the default handler for a nested resource specify an empty list of fields:

class PostHandler(BaseHandler):
      model = Post
      exclude = ('date',)

class UserHandler(BaseHandler):
      model = User
      fields = ('name', ('posts', ()))

This UserHandler shows all fields for all posts for a user excluding the date.

Neither fields, nor exclude are required, and either one can be used by itself.


A pointer to an alternate anonymous resource. See Anonymous Resources


Piston supports pluggable authentication through a simple interface. Resources can be initialized to use any authentication handler that implements the interface. The default is to use the NoAuthentication handler. Adding to the Blogpost example, you could require Basic Authentication as follows:

from django.conf.urls.defaults import *
from piston.resource import Resource
from piston.authentication import HttpBasicAuthentication
from mysite.myapp.api.handlers import BlogpostHandler

auth = HttpBasicAuthentication(realm="Django Piston Example")
blogpost_handler = Resource(BlogpostHandler, authentication=auth)
urlpatterns = patterns('',
   url(r'^blogpost/(?P<post_slug>[^/]+)/', blogpost_handler),
   url(r'^blogposts/', blogpost_handler),

Piston comes with 2 built-in authentication mechanisms, namely piston.authentication.HttpBasicAuthentication and piston.authentication.OAuthAuthentication. The Basic auth handler is very simple, and you should use this for reference if you want to roll your own.

Note: that using piston.authentication.HttpBasicAuthentication with apache and mod_wsgi requires you to add the WSGIPassAuthorization On directive to the server or vhost config, otherwise django-piston cannot read the authentication data from HTTP_AUTHORIZATION in request.META. See: http://code.google.com/p/modwsgi/wiki/ConfigurationDirectives#WSGIPassAuthorization.

An authentication handler is a class, which must have 2 methods: is_authenticated and challenge.

is_authenticated will receive exactly 1 argument, a copy of the request object Django receives. This object will hold all the information you will need to authenticate a user, e.g. request.META.get('HTTP_AUTHENTICATION').

Upon successful authentication, this function must set request.user to the correct django.contrib.auth.models.User object. This allows for subsequent handlers to identify who is logged in.

It must return either True or False, indicating whether the user was logged in.

For cases where authentication fails, is where challenge comes in.

challenge will receive no arguments, and must return a HttpResponse containing the proper challenge instructions. For Basic auth, it will return an empty response, with the header WWW-Authenticate set, and status code 401. This will tell the receiving end that they need to supply us with authentication.

For anonymous handlers, there is a special class, NoAuthentication in piston.authentication that always returns True for is_authenticated.


OAuth is the preferred means of authorization, because it distinguishes between "consumers", i.e. the approved application on your end which is using the API. Piston knows and respects this, and makes good use of it, for example when you use the Dave Lee decorator, it will limit on a per-consumer basis, keeping services operational even if one service has been throttled.

Form Validation

Django has an excellent built-in form validation facility, and Piston can make good use of this.

You can decorate your actions with a @validate decorator, which receives 1 required argument, and one optional. The first argument is the form it will use for validation, and the second argument is the place to look for data. For the create action, this is 'POST' (default), and for update, it's 'PUT'.

For example:

from django import forms
from piston.utils import validate
from mysite.myapp.models import Blogpost

class BlogpostForm(forms.ModelForm):
   class Meta:
      model = Blogpost


def create(...

Or with a normal form:

from django import forms
from piston.utils import validate

class DataForm(forms.Form):
   data = forms.CharField(max_length=128)
   is_private = forms.BooleanField(default=True, required=False)


@validate(DataForm, 'PUT')
def update(....

If data sent to an action that is decorated with a @validate action does not pass the forms is_clean method, Piston will return an error to the client, and will not execute the action. If the validation passes, then the form object is attached to the request object. Thus you can get to the form (and thus the cleaned_data) via request.form as in this example:

@validate(EchoForm, 'GET')
def read(self, request):
    return {'msg': request.form.cleaned_data['msg']}

Helpers, utils & @decorators

For your convenience, there's a set of helpers and utilities you can use. One of those is rc from piston.utils. It contains a set of standard returns that you can return from your actions to indicate a certain situation to the client.

Since 26293e3884f4 , these return a fresh instance of HttpResponse, so you can use something like this:

resp = rc.CREATED
resp.write("Everything went fine!")
return resp

resp = rc.CREATED
resp.write("This will not have the previous 'fine' text in it.")
return resp

This change is backwards compatible, as it overrides __getattr__ to return a new instance rather than a singleton.

rc.ALL_OK200 OKEverything went well.
rc.CREATED201 CreatedObject was created.
rc.DELETED204 (Empty body, as per RFC2616)Object was deleted.
rc.BAD_REQUEST400 Bad RequestRequest was malformed/not understood.
rc.FORBIDDEN401 ForbiddenPermission denied.
rc.NOT_FOUND404 Not FoundResource not found.
rc.DUPLICATE_ENTRY409 Conflict/DuplicateObject already exists.
rc.NOT_HERE410 GoneObject does not exist.
rc.NOT_IMPLEMENTED501 Not ImplementedAction not available.
rc.THROTTLED503 ThrottledRequest was throttled.


Sometimes you may not want people to call a certain action many times in a short period of time. Piston allows you to throttle requests on a global basis, effectively denying them access until the throttle has expired.

Piston will respect OAuth (if used) and limit on a per-consumer basis. If OAuth is not used, Piston will resort to the logged in user, and for anonymous requests, it will fall back to the clients IP address.

Throttling can be enabled via the special Dave Lee decorator. It takes 2 required arguments, and an optional third argument.

The first argument is the number of requests allowed to be made within a certain amount of seconds. The second argument is the number of seconds. The third argument is optional, and should be a string, which will be appended to the cache key, effectively allowing you to do special throttling for a single action, or group several actions together. If omitted, the throttle will be global.

For example:

@throttle(5, 10*60)
def create(...

This will throttle if the client calls 'create' more than 5 times within 10 minutes.

You can do grouping like so:

@throttle(5, 10*60, 'user_writes')
def create(...

@throttle(5, 10*60, 'user_writes')
def update(...

Generating Documentation

Chances are, if you intend to publicly expose your API, that you want to supply documentation. Writing documentation is a tedious process, and even more so if you change things in your code.

Luckily, Piston can do a lot of the heavy lifting for you here as well.

In piston.doc there is a set of methods, allowing you to easily generate documentation using standard Django views and templates.

The function generate_doc returns a HandlerDocumentation instance, which has a few methods:

  • .model (get_model) returns the name of the handler,
  • .doc (get_doc) returns the docstring for the given handler.
  • .get_methods returns a list of methods available. The optional keyword argument include_defaults (False by default) will also include the fallback methods, if you haven't overloaded them. This may be useful if you want to use these, and still include them in your documentation.

get_methods yields a set of HandlerMethod's which are more interesting:

  • .signature (get_signature) will return the methods signature, stripping the first two arguments, which are always 'self' and 'request'. The client will not specify these two, so they are not interesting. Takes an optional argument, parse_optional (default True), which turns keyword arguments defaulting to None into "<optional>".
  • .doc (get_doc) returns the docstring for an action, so you should keep your handler/action specific documentation there.
  • .iter_args() will yield a 2-tuple with the argument name, and the default argument (or None.) If the default argument is None, the default argument will be 'None' (string). This will allow you to distinguish whether there is a default argument (even if it's None), or if it's empty.

For example:

from piston.handler import BaseHandler
from piston.doc import generate_doc

class BlogpostHandler(BaseHandler):
   model = Blogpost

   def read(self, request, post_slug=None):
     Reads all blogposts, or a specific blogpost if
     `post_slug` is supplied.

    def resource_uri():
        return ('api_blogpost_handler', ['id'])

doc = generate_doc(BlogpostHandler)

print doc.name # -> 'BlogpostHandler'
print doc.model # -> <class 'Blogpost'>
print doc.resource_uri_template # -> '/api/post/{id}'

methods = doc.get_methods()

for method in methods:
   print method.name # -> 'read'
   print method.signature # -> 'read(post_slug=<optional>)'

   sig = ''

   for argn, argdef in method.iter_args():
      sig += argn

      if argdef:
         sig += "=%s" % argdef

      sig += ', '

   sig = sig.rstrip(",")

   print sig # -> 'read(repo_slug=None)'

Resource URIs

Each resource can have an URI. They can be accessed in the Handler via his .resource_uri() method.

Also read FAQ: What is a URI Template.


zerok wrote an initial testsuite for Piston, located in tests/. It uses zc.buildout to run the tests, and isolates an environment with Django, etc. The suite comes with two testrunners: tests/bin/test-1.0 and tests/bin/test-1.1 which run the tests against the respective version of Django and are made available after you're finished with the first two steps as described below.

Running the tests is very easy:

$ python bootstrap.py 
Creating directory './bin'.
Generated script './bin/buildout'.

$ ./bin/buildout -v
Develop: 'tests/..'
Getting distribution for 'djangorecipe'.
Got djangorecipe 0.17.3.
Getting distribution for 'zc.recipe.egg'.
Got zc.recipe.egg 1.2.2.
Uninstalling django-1.0.
Installing django-1.0.
django: Downloading Django from: http://www.djangoproject.com/download/1.0.2/tarball/
Generated script './bin/django-1.0'.
Generated script './bin/test-1.0'.

$ ./bin/test-1.0
Creating test database...
Ran 6 tests in 0.283s

Destroying test database...

When running buildout make sure to pass it the -v option. There is currently a small problem with djangorecipe, which is used to create the testscripts etc., that causes the script to hang unless you use the "-v" option.

If you'd like to contribute, more tests are always welcome. There is coverage for many of the basic operations, but not 100%.

Receiving data

Piston, being layered on HTTP, works well with post-data (form data), but also works well with more expressive formats such as JSON and YAML.

This allows you to receive structured data easily, rather than just key-value pairs. Piston will attempt to deserialize incoming non-form data via a set of "loaders", depending on the Content-type specified by the client.

For example, if we send JSON to a handler giving the content-type "application/json", Piston will do 2 things:

  1. Place the deserialized data in request.data, and
  2. Set request.content_type to application/json. For form data, this will always be None.

You can use it like so (from testapp/handlers.py):

    def create(self, request):
        if request.content_type:
            data = request.data
            em = self.model(title=data['title'], content=data['content'])
            for comment in data['comments']:
                Comment(parent=em, content=comment['content']).save()
            return rc.CREATED
            super(ExpressiveTestModel, self).create(request)

If we send the following JSON structure into that, it will handle it appropriately:

{"content": "test", "comments": [{"content": "test1"}, {"content": "test2"}], "title": "test"}

It should be noted that sending anything that deserializes to this handler will also work, so you can send equally formatted YAML or XML, and the handler won't care.

If your handler doesn't accept post data (maybe it requires more verbose data), there's an easy way to require a specific type of data, via the utils.require_mime decorator.

This decorator takes a list of types it requires, and you can use the shorthand too, like 'yaml', 'json', etc.

There's also a shortcut for requiring 'json', 'yaml', 'xml' and 'pickle' all in one, called 'require_extended'.


class SomeHandler(BaseHandler):
   @require_mime('json', 'yaml')
   def create(...

   def update(...


Since b0a1571ff61a , Piston supports streaming its output to the client. This is disabled per default, for one reason:

  • Django's support for streaming breaks with ConditionalGetMiddleware and CommonMiddleware.

To get around this, Piston ships with two "proxy middleware classes" that won't execute during a streaming scenario, and hence won't look at (and exhaust) the data before sending it to the client. Without these, Django will look at the contents (to figure out E-Tags and Content-Length), and by doing so, the next peek it takes, will result in nothing.

In piston.middleware there are two classes you can effectively replace these with.

In settings.py:

   # ...
   # ...

Remove any mentions of ConditionalGetMiddleware and CommonMiddleware, or it won't work. If you have any other middleware that looks at the content prior to streaming, you can wrap those in the conditional middleware proxy too:

from piston.middleware import compat_middleware_factory

class MyMiddleware(...):

MyMiddlewareCompatProxy = compat_middleware_factory(MyMiddleware)

And then install MyMiddlewareCompatProxy instead.

Configuration variables

Piston is configurable in a couple of ways, which allows more granular control of some areas without editing the code.

settings.PISTON_EMAIL_ERRORSIf (when) Piston crashes, it will email the administrators a backtrace (like the Django one you see during DEBUG = True)
settings.PISTON_DISPLAY_ERRORSUpon crashing, will display a small backtrace to the client, including the method signature expected.
settings.PISTON_STREAM_OUTPUTWhen enabled, Piston will instruct Django to stream the output to the client, but please read streaming before enabling it.