asyncio：如何在不强制完全暴露于 asyncio 的情况下组合同步代码

Question

It seems that python's asyncio is an all-or-nothing kind of deal. Specifically it seems to force full call stack to be async-aware.

Here is an example:

Let's assume that there is an algorithm implemented by function a() .
That algorithm calls function b() .

def a(b):
  for i in ...:
    # do smth...
    res = b()
    # do smth more ...

Now let's assume that function b() might be implemented differently in different environments and one might want to call a() inside an async event loop (with loop.call_soon() ) with b() calling some asyncio-aware code:

def b():
   await .... # this forces b() to be declared async.

Using existing event loop seems to be explicitly prohibited... So, are there methods to implement this functionality without forcing the full stack ( a() and b() ) to be explicitly defined as async coroutines?

Answer 1

Although taken at face value asyncio is an all-or-nothing deal, there are still two escape hatches that can help when dealing with legacy or heterogenous environment:

• Within asyncio, you can await loop.run_in_executor(callable, args...) to evaluate sync code in another thread, suspending the current coroutine until the result is ready.

• Outside asyncio, you can use asyncio.run_coroutine_threadsafe() to submit a coroutine to an event loop running in another thread. It immediately returns returns a concurrent.futures future which has a blocking result() method that blocks the current thread without adversely affecting the event loop. (The use of run_coroutine_threadsafe requires starting an event loop in a dedicated thread beforehand.)

With those two tools at your disposal, it is not that hard to mix asyncio and classic synchronous code.