django/docs/topics/async.txt

====================
Asynchronous support
====================

.. currentmodule:: asgiref.sync

Django has support for writing asynchronous ("async") views, along with an
entirely async-enabled request stack if you are running under
:doc:`ASGI </howto/deployment/asgi/index>`. Async views will still work under
WSGI, but with performance penalties, and without the ability to have efficient
long-running requests.

We're still working on async support for the ORM and other parts of Django.
You can expect to see this in future releases. For now, you can use the
:func:`sync_to_async` adapter to interact with the sync parts of Django.
There is also a whole range of async-native Python libraries that you can
integrate with.

.. versionchanged:: 3.1

    Support for async views was added.

Async views
===========

.. versionadded:: 3.1

Any view can be declared async by making the callable part of it return a
coroutine - commonly, this is done using ``async def``. For a function-based
view, this means declaring the whole view using ``async def``. For a
class-based view, this means making its ``__call__()`` method an ``async def``
(not its ``__init__()`` or ``as_view()``).

.. note::

    Django uses ``asyncio.iscoroutinefunction`` to test if your view is
    asynchronous or not. If you implement your own method of returning a
    coroutine, ensure you set the ``_is_coroutine`` attribute of the view
    to ``asyncio.coroutines._is_coroutine`` so this function returns ``True``.

Under a WSGI server, async views will run in their own, one-off event loop.
This means you can use async features, like parallel async HTTP requests,
without any issues, but you will not get the benefits of an async stack.

The main benefits are the ability to service hundreds of connections without
using Python threads. This allows you to use slow streaming, long-polling, and
other exciting response types.

If you want to use these, you will need to deploy Django using
:doc:`ASGI </howto/deployment/asgi/index>` instead.

.. warning::

    You will only get the benefits of a fully-asynchronous request stack if you
    have *no synchronous middleware* loaded into your site. If there is a piece
    of synchronous middleware, then Django must use a thread per request to
    safely emulate a synchronous environment for it.

    Middleware can be built to support :ref:`both sync and async
    <async-middleware>` contexts. Some of Django's middleware is built like
    this, but not all. To see what middleware Django has to adapt, you can turn
    on debug logging for the ``django.request`` logger and look for log
    messages about *"Synchronous middleware ... adapted"*.

In both ASGI and WSGI mode, you can still safely use asynchronous support to
run code in parallel rather than serially. This is especially handy when
dealing with external APIs or data stores.

If you want to call a part of Django that is still synchronous, like the ORM,
you will need to wrap it in a :func:`sync_to_async` call. For example::

    from asgiref.sync import sync_to_async

    results = sync_to_async(Blog.objects.get)(pk=123)

You may find it easier to move any ORM code into its own function and call that
entire function using :func:`sync_to_async`. For example::

    from asgiref.sync import sync_to_async

    @sync_to_async
    def get_blog(pk):
        return Blog.objects.select_related('author').get(pk=pk)

If you accidentally try to call a part of Django that is still synchronous-only
from an async view, you will trigger Django's
:ref:`asynchronous safety protection <async-safety>` to protect your data from
corruption.

Performance
-----------

When running in a mode that does not match the view (e.g. an async view under
WSGI, or a traditional sync view under ASGI), Django must emulate the other
call style to allow your code to run. This context-switch causes a small
performance penalty of around a millisecond.

This is also true of middleware. Django will attempt to minimize the number of
context-switches between sync and async. If you have an ASGI server, but all
your middleware and views are synchronous, it will switch just once, before it
enters the middleware stack.

However, if you put synchronous middleware between an ASGI server and an
asynchronous view, it will have to switch into sync mode for the middleware and
then back to async mode for the view. Django will also hold the sync thread
open for middleware exception propagation. This may not be noticeable at first,
but adding this penalty of one thread per request can remove any async
performance advantage.

You should do your own performance testing to see what effect ASGI versus WSGI
has on your code. In some cases, there may be a performance increase even for
a purely synchronous codebase under ASGI because the request-handling code is
still all running asynchronously. In general you will only want to enable ASGI
mode if you have asynchronous code in your project.

.. _async-safety:

Async safety
============

.. envvar:: DJANGO_ALLOW_ASYNC_UNSAFE

Certain key parts of Django are not able to operate safely in an async
environment, as they have global state that is not coroutine-aware. These parts
of Django are classified as "async-unsafe", and are protected from execution in
an async environment. The ORM is the main example, but there are other parts
that are also protected in this way.

If you try to run any of these parts from a thread where there is a *running
event loop*, you will get a
:exc:`~django.core.exceptions.SynchronousOnlyOperation` error. Note that you
don't have to be inside an async function directly to have this error occur. If
you have called a sync function directly from an async function,
without using :func:`sync_to_async` or similar, then it can also occur. This is
because your code is still running in a thread with an active event loop, even
though it may not be declared as async code.

If you encounter this error, you should fix your code to not call the offending
code from an async context. Instead, write your code that talks to async-unsafe
functions in its own, sync function, and call that using
:func:`asgiref.sync.sync_to_async` (or any other way of running sync code in
its own thread).

You may still be forced to run sync code from an async context. For example,
if the requirement is forced on you by an external environment, such as in a
Jupyter_ notebook. If you are sure there is no chance of the code being run
concurrently, and you *absolutely* need to run this sync code from an async
context, then you can disable the warning by setting the
:envvar:`DJANGO_ALLOW_ASYNC_UNSAFE` environment variable to any value.

.. warning::

    If you enable this option and there is concurrent access to the
    async-unsafe parts of Django, you may suffer data loss or corruption. Be
    very careful and do not use this in production environments.

If you need to do this from within Python, do that with ``os.environ``::

    import os

    os.environ["DJANGO_ALLOW_ASYNC_UNSAFE"] = "true"

.. _Jupyter: https://jupyter.org/

Async adapter functions
=======================

It is necessary to adapt the calling style when calling sync code from an async
context, or vice-versa. For this there are two adapter functions, from the
``asgiref.sync`` module: :func:`async_to_sync` and :func:`sync_to_async`. They
are used to transition between the calling styles while preserving
compatibility.

These adapter functions are widely used in Django. The `asgiref`_ package
itself is part of the Django project, and it is automatically installed as a
dependency when you install Django with ``pip``.

.. _asgiref: https://pypi.org/project/asgiref/

``async_to_sync()``
-------------------

.. function:: async_to_sync(async_function, force_new_loop=False)

Takes an async function and returns a sync function that wraps it. Can be used
as either a direct wrapper or a decorator::

    from asgiref.sync import async_to_sync

    async def get_data(...):
        ...

    sync_get_data = async_to_sync(get_data)

    @async_to_sync
    async def get_other_data(...):
        ...

The async function is run in the event loop for the current thread, if one is
present. If there is no current event loop, a new event loop is spun up
specifically for the single async invocation and shut down again once it
completes. In either situation, the async function will execute on a different
thread to the calling code.

Threadlocals and contextvars values are preserved across the boundary in both
directions.

:func:`async_to_sync` is essentially a more powerful version of the
:py:func:`asyncio.run` function in Python's standard library. As well
as ensuring threadlocals work, it also enables the ``thread_sensitive`` mode of
:func:`sync_to_async` when that wrapper is used below it.

``sync_to_async()``
-------------------

.. function:: sync_to_async(sync_function, thread_sensitive=False)

Takes a sync function and returns an async function that wraps it. Can be used
as either a direct wrapper or a decorator::

    from asgiref.sync import sync_to_async

    async_function = sync_to_async(sync_function)
    async_function = sync_to_async(sensitive_sync_function, thread_sensitive=True)

    @sync_to_async
    def sync_function(...):
        ...

Threadlocals and contextvars values are preserved across the boundary in both
directions.

Sync functions tend to be written assuming they all run in the main
thread, so :func:`sync_to_async` has two threading modes:

* ``thread_sensitive=False`` (the default): the sync function will run in a
  brand new thread which is then closed once the invocation completes.

* ``thread_sensitive=True``: the sync function will run in the same thread as
  all other ``thread_sensitive`` functions. This will be the main thread, if
  the main thread is synchronous and you are using the :func:`async_to_sync`
  wrapper.

Thread-sensitive mode is quite special, and does a lot of work to run all
functions in the same thread. Note, though, that it *relies on usage of*
:func:`async_to_sync` *above it in the stack* to correctly run things on the
main thread. If you use ``asyncio.run()`` or similar, it will fall back to
running thread-sensitive functions in a single, shared thread, but this will
not be the main thread.

The reason this is needed in Django is that many libraries, specifically
database adapters, require that they are accessed in the same thread that they
were created in. Also a lot of existing Django code assumes it all runs in the
same thread, e.g. middleware adding things to a request for later use in views.

Rather than introduce potential compatibility issues with this code, we instead
opted to add this mode so that all existing Django sync code runs in the same
thread and thus is fully compatible with async mode. Note that sync code will
always be in a *different* thread to any async code that is calling it, so you
should avoid passing raw database handles or other thread-sensitive references
around.