Add asynchronous responses for use with an ASGI server

[florianvazelle]

This ticket follows the one I opened on ​django/daphne.

Initial Issue

Using Daphne as a server for a Django application seems to cause a malfunction in streaming responses.
When you pass an iterator into a response, the contents of the response are not retrieved chunk by chunk, but from a single block when the iterator has finished iterating.
I have a minimal project that can reproduce the behavior, at ​florianvazelle/minimal-daphne-stream.

Solution

After some research, in Daphne, we use Twisted with asyncioreactor. I managed to reproduce the bug in a minimal way: ​Minimal example of an "ASGI-like" streaming server with Twisted.
The streaming issue occurs when we call a blocking method in an iterator (io, sleep, request ...), which blocks the reactor. That's why the result is not done progressively.
The reactor does not get control of execution back until the end of the iteration.
To correct this behavior we need to add an asynchronous layer and use async / non-blocking alternatives.

Proposition

In my minimal project, the view will become:

import asyncio

from django.http.response import StreamingHttpResponse


async def iterable_content():
    for _ in range(5):
        await asyncio.sleep(1)
        print('Returning chunk')
        yield b'a' * 10000


def test_stream_view(request):
    return StreamingHttpResponse(iterable_content())

But django does not handle asynchronous generators.
Some works are already did in this tickets #32798 (StreamingHttpResponse Raises SynchronousOnlyOperation in ASGI Server) – Django, but considering the performance, we could manage it differently here.

I propose to add asynchronous responses, to be used in an async context, so with an ASGI server.
I make a PoC, in a very basic way: ​florianvazelle/django.