Serve with ASGI¶

Django's ASGI handler hops to a worker thread twice per request, for its request_started and request_finished signals: about 1 ms on a small machine, often more than the view. djox's handler sends the same signals right on the event loop when that can't block it.

# asgi.py
import os

os.environ.setdefault("DJANGO_SETTINGS_MODULE", "mysite.settings")

from djox.asgi import get_asgi_application

application = get_asgi_application()  # was django.core.asgi.get_asgi_application()

Serve it as usual: uvicorn mysite.asgi:application, granian, daphne...

What changes¶

Nothing you can see, but speed:

• The same signals, receivers and moments. request_started before the view, request_finished when the response is closed.
• On the event loop when it is safe: every receiver is one Django (or djox) adds, and the request ran no sync code. An async view that uses djox (or no database) is in that case.
• On a thread, like Django, otherwise: a sync view or sync middleware (its connections must be closed on its thread), a receiver of your own, a cache that closes connections (memcached, Redis), a file response.

An async view on djox, on the cloud VM djox is tested on:

Good to know¶

• Everything else is Django's ASGIHandler: reading the body, middleware, views, sending the response, a client that disconnects early.
• It doesn't need "djox" in INSTALLED_APPS, nor djox's ENGINE, but only a view whose queries run on djox avoids the threads.