Articles / Working with HTTP requests in ASGI

This article is part of an ongoing series exploring the emerging ASGI standard. To keep up to date with the full series, subscribe to the RSS feed.

In our last article we took a look at the basic structure of the ASGI interface. Now we’re going to go into some more detail on the message structure for HTTP requests, and take a look at how we can use some of the datastructures that the Starlette package provides in order to work with HTTP requests in ASGI.

The first thing that occurs in any ASGI application is that it is instantiated with a “scope” dictionary, that provides some initial information about the incoming request.

Here’s an example of how the scope dictionary might look for a simple HTTP request.

>>> scope = {
    "type": "http",
    "http_version": "1.1",
    "method": "GET",
    "scheme": "https",
    "path": "/",
    "query_string": b"search=red+blue&maximum_price=20",
    "headers": [
        (b"host", b""),
        (b"accept", b"application/json")
    "client": ("", 1453),
    "server": ("", 443)

The scope dictionary is pretty similar to WSGI’s “environ” dictionary. In fact, the ASGI specification documents how to map between the two.

environ = {
    "SCRIPT_NAME": "",
    "PATH_INFO": "/",
    "QUERY_STRING": "search=red+blue&maximum_price=20",
    "SERVER_NAME": "",
    "SERVER_PORT": 443,
    "REMOTE_HOST": "",
    "REMOTE_PORT": 1453,
    "HTTP_HOST": "",
    "HTTP_ACCEPT": "application/json",

The scope type

A fundamental difference between ASGI and WSGI, is that ASGI is a more general purpose messaging interface, whereas WSGI is strictly request/response.

The one element that must always be present in any ASGI scope is the “type” key, which is used to indicate the protocol type.

scope = {
    "type": "http",  # Deal with an incoming HTTP request.

Request URL

The full URL of an incoming request can be constructed based on the information contained in scheme, server, path, and query_string.

Since we don’t want to have to typically be working from the raw ASGI information, it’s useful for us to have some reusable tooling that we can use to abstract away some of the details for us.

The Starlette framework is deliberately designed as a zero-dependency library, so that it can be used as the basis for all kinds of ASGI applications or other higher-level frameworks.

One of the things that is provides is a set of basic datastructures for working with ASGI. For example:

>>> from starlette.datastructures import URL
>>> url = URL(scope=scope)
>>> url
>>> str(url)

The URL instance allows you to inspect the various components on the URL, in the same way as a the standard library’s urlparse:

>>> url.scheme

You can also modify components of the URL, and return a new instance:

>>> url.replace(hostname='')

Request headers

HTTP headers in ASGI are mandated to be a list of pairs of bytes, representing the header name and value. Because HTTP headers are case-insensitive, ASGI enforces that the byte-wise header representation must be strictly lower-cased.

HTTP headers can also contain multiple instances of the same header name, for example as used with Set-Cookie.

Starlette provides an immutable, case-insensitive, multi-dict for working with HTTP request headers from ASGI.

>>> from starlette.datastructures import Headers
>>> headers = Headers(scope=scope)
>>> headers
Headers({'host': '', 'accept': 'application/json'})

Instantiating a Headers datastructure is a cheap operation, and provides a more convenient interface for inspecting the request headers than iterating over the byte-wise pairs directly.

>>> headers['Accept']

For response headers, Starlette provides a MutableHeaders data structure, that is equivelent, but also also setting or deleting header values.

Request query params

One particular part of the URL that we need to deal with is the request query parameters, eg “?search=red+blue&maximum_price=20”.

To work with these, use the QueryParams class, which provides an immutable multi-dict implementation onto the parameters.

>>> from starlette.datastructures import QueryParams
>>> params = QueryParams(scope=scope)
>>> params
>>> params['search']
'red blue'

Request body

Most of the information about the incoming request is stored in the “scope”, and presented at the point the ASGI app is instantiated. However for the request body, that’s not possible.

In order to access the request body, we have to get a stream of messages from the “receive” channel. Here’s how we can obtain a single message in the stream:

message = await receive()

And an example of how an HTTP request body message is structured:

    'type': 'http.request.body',
    'body': b'{"example": "Some JSON data"}',
    'more_body': False

If you’re working with ASGI directly, here’s a pattern for how you’d consume the entire stream of the HTTP request body:

body = b''
more_body = True
while more_body:
    message = await receive()
    body += message.get('body', b'')
    more_body = message.get('more_body', False)

Starlette provides a lightweight way of wrapping both the “scope” and “receive” channel up in a request interface, that gives simpler ways to get access to the request body:

request = Request(scope, receive=receive)
body = await request.body()

Alternatively, to get a JSON-parsed representation:

request = Request(scope, receive=receive)
body = await request.json()

For large requests it’s possible that you may not want to consume the entire request body into memory at once.

From Python 3.6 onwards the asynchronous generator syntax is supported, which allows us to provide a nice simple API onto streaming the request body…

request = Request(scope, receive=receive)

async for chunk in
    ... # Do something with "chunk"

We can combine this with request parsing in order to handle form data gracefully.

When handling HTTP multipart requests, you’ll typically want to ensure that the request parsing is able to stream any file upload content into temporary files on disk, without having to load all the data into memory first.

request = Request(scope, receive=receive)

# Any upload files in the request body will be streamed into temporary files.
form = await request.form()


We’ve seen how the ASGI “scope” message is structured for HTTP requests, and how the message body is streamed over the “receive” channel.

We’ve also used some of the fundamental datastructures that Starlette provides, in order to work with ASGI at a slightly higher level.

Follow all our articles on the RSS feed.