Middleware

Middleware is a framework of hooks into Django’s request/response processing. It’s a light, low-level “plugin” system for globally altering Django’s input or output.

Each middleware component is responsible for doing some specific function. For example, Django includes a middleware component, AuthenticationMiddleware, that associates users with requests using sessions.

This document explains how middleware works, how you activate middleware, and how to write your own middleware. Django ships with some built-in middleware you can use right out of the box. They’re documented in the built-in middleware reference.

Writing your own middleware

A middleware factory is a callable that takes a get_response callable and returns a middleware. A middleware is a callable that takes a request and returns a response, just like a view.

A middleware can be written as a function that looks like this:

def simple_middleware(get_response):
    # One-time configuration and initialization.

    def middleware(request):
        # Code to be executed for each request before
        # the view (and later middleware) are called.

        response = get_response(request)

        # Code to be executed for each request/response after
        # the view is called.

        return response

    return middleware

Or it can be written as a class whose instances are callable, like this:

class SimpleMiddleware:
    def __init__(self, get_response):
        self.get_response = get_response
        # One-time configuration and initialization.

    def __call__(self, request):
        # Code to be executed for each request before
        # the view (and later middleware) are called.

        response = self.get_response(request)

        # Code to be executed for each request/response after
        # the view is called.

        return response

The get_response callable provided by Django might be the actual view (if this is the last listed middleware) or it might be the next middleware in the chain. The current middleware doesn’t need to know or care what exactly it is, just that it represents whatever comes next.

The above is a slight simplification – the get_response callable for the last middleware in the chain won’t be the actual view but rather a wrapper method from the handler which takes care of applying view middleware, calling the view with appropriate URL arguments, and applying template-response and exception middleware.

Middleware can either support only synchronous Python (the default), only asynchronous Python, or both. See Asynchronous support for details of how to advertise what you support, and know what kind of request you are getting.

Middleware can live anywhere on your Python path.

__init__(get_response)

Middleware factories must accept a get_response argument. You can also initialize some global state for the middleware. Keep in mind a couple of caveats:

  • Django initializes your middleware with only the get_response argument, so you can’t define __init__() as requiring any other arguments.
  • Unlike the __call__() method which is called once per request, __init__() is called only once, when the Web server starts.

Marking middleware as unused

It’s sometimes useful to determine at startup time whether a piece of middleware should be used. In these cases, your middleware’s __init__() method may raise MiddlewareNotUsed. Django will then remove that middleware from the middleware process and log a debug message to the django.request logger when DEBUG is True.

Activating middleware

To activate a middleware component, add it to the MIDDLEWARE list in your Django settings.

In MIDDLEWARE, each middleware component is represented by a string: the full Python path to the middleware factory’s class or function name. For example, here’s the default value created by django-admin startproject:

MIDDLEWARE = [
    'django.middleware.security.SecurityMiddleware',
    'django.contrib.sessions.middleware.SessionMiddleware',
    'django.middleware.common.CommonMiddleware',
    'django.middleware.csrf.CsrfViewMiddleware',
    'django.contrib.auth.middleware.AuthenticationMiddleware',
    'django.contrib.messages.middleware.MessageMiddleware',
    'django.middleware.clickjacking.XFrameOptionsMiddleware',
]

A Django installation doesn’t require any middleware — MIDDLEWARE can be empty, if you’d like — but it’s strongly suggested that you at least use CommonMiddleware.

The order in MIDDLEWARE matters because a middleware can depend on other middleware. For instance, AuthenticationMiddleware stores the authenticated user in the session; therefore, it must run after SessionMiddleware. See Middleware ordering for some common hints about ordering of Django middleware classes.

Middleware order and layering

During the request phase, before calling the view, Django applies middleware in the order it’s defined in MIDDLEWARE, top-down.

You can think of it like an onion: each middleware class is a “layer” that wraps the view, which is in the core of the onion. If the request passes through all the layers of the onion (each one calls get_response to pass the request in to the next layer), all the way to the view at the core, the response will then pass through every layer (in reverse order) on the way back out.

If one of the layers decides to short-circuit and return a response without ever calling its get_response, none of the layers of the onion inside that layer (including the view) will see the request or the response. The response will only return through the same layers that the request passed in through.

Other middleware hooks

Besides the basic request/response middleware pattern described earlier, you can add three other special methods to class-based middleware:

process_view()

process_view(request, view_func, view_args, view_kwargs)

request is an HttpRequest object. view_func is the Python function that Django is about to use. (It’s the actual function object, not the name of the function as a string.) view_args is a list of positional arguments that will be passed to the view, and view_kwargs is a dictionary of keyword arguments that will be passed to the view. Neither view_args nor view_kwargs include the first view argument (request).

process_view() is called just before Django calls the view.

It should return either None or an HttpResponse object. If it returns None, Django will continue processing this request, executing any other process_view() middleware and, then, the appropriate view. If it returns an HttpResponse object, Django won’t bother calling the appropriate view; it’ll apply response middleware to that HttpResponse and return the result.

Note

Accessing request.POST inside middleware before the view runs or in process_view() will prevent any view running after the middleware from being able to modify the upload handlers for the request, and should normally be avoided.

The CsrfViewMiddleware class can be considered an exception, as it provides the csrf_exempt() and csrf_protect() decorators which allow views to explicitly control at what point the CSRF validation should occur.

process_exception()

process_exception(request, exception)

request is an HttpRequest object. exception is an Exception object raised by the view function.

Django calls process_exception() when a view raises an exception. process_exception() should return either None or an HttpResponse object. If it returns an HttpResponse object, the template response and response middleware will be applied and the resulting response returned to the browser. Otherwise, default exception handling kicks in.

Again, middleware are run in reverse order during the response phase, which includes process_exception. If an exception middleware returns a response, the process_exception methods of the middleware classes above that middleware won’t be called at all.

process_template_response()

process_template_response(request, response)

request is an HttpRequest object. response is the TemplateResponse object (or equivalent) returned by a Django view or by a middleware.

process_template_response() is called just after the view has finished executing, if the response instance has a render() method, indicating that it is a TemplateResponse or equivalent.

It must return a response object that implements a render method. It could alter the given response by changing response.template_name and response.context_data, or it could create and return a brand-new TemplateResponse or equivalent.

You don’t need to explicitly render responses – responses will be automatically rendered once all template response middleware has been called.

Middleware are run in reverse order during the response phase, which includes process_template_response().

Dealing with streaming responses

Unlike HttpResponse, StreamingHttpResponse does not have a content attribute. As a result, middleware can no longer assume that all responses will have a content attribute. If they need access to the content, they must test for streaming responses and adjust their behavior accordingly:

if response.streaming:
    response.streaming_content = wrap_streaming_content(response.streaming_content)
else:
    response.content = alter_content(response.content)

Note

streaming_content should be assumed to be too large to hold in memory. Response middleware may wrap it in a new generator, but must not consume it. Wrapping is typically implemented as follows:

def wrap_streaming_content(content):
    for chunk in content:
        yield alter_content(chunk)

Exception handling

Django automatically converts exceptions raised by the view or by middleware into an appropriate HTTP response with an error status code. Certain exceptions are converted to 4xx status codes, while an unknown exception is converted to a 500 status code.

This conversion takes place before and after each middleware (you can think of it as the thin film in between each layer of the onion), so that every middleware can always rely on getting some kind of HTTP response back from calling its get_response callable. Middleware don’t need to worry about wrapping their call to get_response in a try/except and handling an exception that might have been raised by a later middleware or the view. Even if the very next middleware in the chain raises an Http404 exception, for example, your middleware won’t see that exception; instead it will get an HttpResponse object with a status_code of 404.

You can set DEBUG_PROPAGATE_EXCEPTIONS to True to skip this conversion and propagate exceptions upwards.

Asynchronous support

New in Django 3.1.

Middleware can support any combination of synchronous and asynchronous requests. Django will adapt requests to fit the middleware’s requirements if it cannot support both, but at a performance penalty.

By default, Django assumes that your middleware is capable of handling only synchronous requests. To change these assumptions, set the following attributes on your middleware factory function or class:

  • sync_capable is a boolean indicating if the middleware can handle synchronous requests. Defaults to True.
  • async_capable is a boolean indicating if the middleware can handle asynchronous requests. Defaults to False.

If your middleware has both sync_capable = True and async_capable = True, then Django will pass it the request without converting it. In this case, you can work out if your middleware will receive async requests by checking if the get_response object you are passed is a coroutine function, using asyncio.iscoroutinefunction.

The django.utils.decorators module contains sync_only_middleware(), async_only_middleware(), and sync_and_async_middleware() decorators that allow you to apply these flags to middleware factory functions.

The returned callable must match the sync or async nature of the get_response method. If you have an asynchronous get_response, you must return a coroutine function (async def).

process_view, process_template_response and process_exception methods, if they are provided, should also be adapted to match the sync/async mode. However, Django will individually adapt them as required if you do not, at an additional performance penalty.

Here’s an example of how to create a middleware function that supports both:

import asyncio
from django.utils.decorators import sync_and_async_middleware

@sync_and_async_middleware
def simple_middleware(get_response):
    # One-time configuration and initialization goes here.
    if asyncio.iscoroutinefunction(get_response):
        async def middleware(request):
            # Do something here!
            response = await get_response(request)
            return response

    else:
        def middleware(request):
            # Do something here!
            response = get_response(request)
            return response

    return middleware

Note

If you declare a hybrid middleware that supports both synchronous and asynchronous calls, the kind of call you get may not match the underlying view. Django will optimize the middleware call stack to have as few sync/async transitions as possible.

Thus, even if you are wrapping an async view, you may be called in sync mode if there is other, synchronous middleware between you and the view.

Upgrading pre-Django 1.10-style middleware

class django.utils.deprecation.MiddlewareMixin

Django provides django.utils.deprecation.MiddlewareMixin to ease creating middleware classes that are compatible with both MIDDLEWARE and the old MIDDLEWARE_CLASSES, and support synchronous and asynchronous requests. All middleware classes included with Django are compatible with both settings.

The mixin provides an __init__() method that requires a get_response argument and stores it in self.get_response.

The __call__() method:

  1. Calls self.process_request(request) (if defined).
  2. Calls self.get_response(request) to get the response from later middleware and the view.
  3. Calls self.process_response(request, response) (if defined).
  4. Returns the response.

If used with MIDDLEWARE_CLASSES, the __call__() method will never be used; Django calls process_request() and process_response() directly.

In most cases, inheriting from this mixin will be sufficient to make an old-style middleware compatible with the new system with sufficient backwards-compatibility. The new short-circuiting semantics will be harmless or even beneficial to the existing middleware. In a few cases, a middleware class may need some changes to adjust to the new semantics.

These are the behavioral differences between using MIDDLEWARE and MIDDLEWARE_CLASSES:

  1. Under MIDDLEWARE_CLASSES, every middleware will always have its process_response method called, even if an earlier middleware short-circuited by returning a response from its process_request method. Under MIDDLEWARE, middleware behaves more like an onion: the layers that a response goes through on the way out are the same layers that saw the request on the way in. If a middleware short-circuits, only that middleware and the ones before it in MIDDLEWARE will see the response.
  2. Under MIDDLEWARE_CLASSES, process_exception is applied to exceptions raised from a middleware process_request method. Under MIDDLEWARE, process_exception applies only to exceptions raised from the view (or from the render method of a TemplateResponse). Exceptions raised from a middleware are converted to the appropriate HTTP response and then passed to the next middleware.
  3. Under MIDDLEWARE_CLASSES, if a process_response method raises an exception, the process_response methods of all earlier middleware are skipped and a 500 Internal Server Error HTTP response is always returned (even if the exception raised was e.g. an Http404). Under MIDDLEWARE, an exception raised from a middleware will immediately be converted to the appropriate HTTP response, and then the next middleware in line will see that response. Middleware are never skipped due to a middleware raising an exception.
Changed in Django 3.1:

Support for asynchronous requests was added to the MiddlewareMixin.