Aiopyramid uses a design similar to the uWSGI asyncio plugin. The asyncio event loop runs in a parent greenlet, while wsgi callables run in child greenlets. Because the callables are running in greenlets, it is possible to suspend a callable and switch to parent to run coroutines all on one event loop. Each task tracks which child greenlet it belongs to and switches back to the appropriate callable when it is done.

The greenlet model makes it possible to have any Python code wait for a coroutine even when that code is unaware of asyncio. The uWSGI asyncio plugin sets up the architecture by itself, but it is also possible to setup this architecture whenever we have a running asyncio event loop using spawn_greenlet().

For example, there may be times when a coroutine would need to call some function a that later calls a coroutine b. Since coroutines run in the parent greenlet (i.e. on the event loop) and the function a cannot yield from b because it is not a coroutine itself, the parent coroutine will need to set up the Aiopyramid architecture so that b can be synchronized with synchronize() and called like a normal function from inside a.

The following code demonstrates this usage without needing to setup a server.

>>> import asyncio
>>> from aiopyramid.helpers import synchronize, spawn_greenlet
>>> @synchronize
... @asyncio.coroutine
... def some_async_task():
...   print('I am a synchronized coroutine.')
...   yield from asyncio.sleep(0.2)
...   print('Synchronized task done.')
>>> def normal_function():
...   print('I am normal function that needs to call some_async_task')
...   some_async_task()
...   print('I (normal_function) called it, and it is done now like I expect.')
>>> @asyncio.coroutine
... def parent():
...   print('I am a traditional coroutine that needs to call the naive normal_function')
...   yield from spawn_greenlet(normal_function)
...   print('All is done.')
>>> loop = asyncio.get_event_loop()
>>> loop.run_until_complete(parent())
I am a traditional coroutine that needs to call the naive normal_function
I am normal function that needs to call some_async_task
I am a synchronized coroutine.
Synchronized task done.
I (normal_function) called it, and it is done now like I expect.
All is done.

Please feel free to use this in other asyncio projects that don’t use Pyramid because it’s awesome.

To avoid confusion, it is worth making explicit the fact that this approach is for incorporating code that is fast and non-blocking itself but needs to call a coroutine to do some io. Don’t try to use this to call long-running or blocking Python functions. Instead, use run_in_executor, which is what Aiopyramid does by default with view callables that don’t appear to be coroutines.


Aiopyramid was originally based on pyramid_asyncio, but I chose a different approach for the following reasons:

On the other hand Aiopyramid is designed to follow these principles:

  • Aiopyramid should extend Pyramid through existing Pyramid mechanisms where possible.
  • Asynchronous code should be wrapped so that existing callers can treat it as synchronous code.
  • Ultimately, no framework can guarantee that all io calls are non-blocking because it is always possible for a programmer to call out to some function that blocks (in other words, the programmer forgets to wrap long-running calls in run_in_executor). So, frameworks should leave the determination of what code is safe to the programmer and instead provide tools for programmers to make educated decisions about what Python libraries can be used on an asynchronous server. Following the Pyramid philosophy, frameworks should not get in the way.

The first principle is one of the reasons why I used view mappers rather than patching the router. View mappers are a mechanism already in place to handle how views are called. We don’t need to rewrite vast parts of Pyramid to run a view in the asyncio event loop. Yes, Pyramid is that awesome.

The second principle is what allows Aiopyramid to support existing extensions. The goal is to isolate asynchronous code from code that expects a synchronous response. Those methods that already exist in Pyramid should not be rewritten as coroutines because we don’t know who will try to call them as regular methods.

Most of the Pyramid framework does not run io blocking code. So, it is not actually necessary to change the framework itself. Instead we need tools for making application code asynchronous. It should be possible to run an existing simple url dispatch application asynchronously without modification. Blocking code will naturally end up being run in a separate thread via the run_in_executor method. This allows you to optimize only those highly concurrent views in your application or add in websocket support without needing to refactor all of the code.

It is easy to simulate a multithreaded server by increasing the number of threads available to the executor.

For example, include the following in your application’s constructor:

import asyncio
from concurrent.futures import ThreadPoolExecutor


It should be noted that Aiopyramid is not thread-safe by nature. You will need to ensure that in memory resources are not modified by multiple non-coroutine view callables. For most existing applications, this should not be a problem.