In the previous post, we saw an example of asynchronous programming in Python using yield from.

In that example, we used our own custom event loop. We also mocked the network I/O operations run concurrently.

In this post, we will firstly rewrite the previous post’s example using asyncio. This will allow us to avoid writing a custom event loop, and use more standard coroutine interaction in an asynchronous program (that prescribed by asyncio).

Afterwards, we will adapt our rewritten script to use aiohttp, replacing our mocked network I/O operations with real HTTP GET requests.

## Replace custom code with asyncio

In the previous post, the code outside of main() looked like this

import time

start = time.time()

def start_network_io():
pass  # make low level call to OS to start network I/O

def is_network_io_complete(start):
# mock OS polling checking if network I/O is complete
# mocked so that network I/O completes after 3 seconds
return time.time() - start > 3

def get_network_io_response():
return 200

def network_io_coroutine():
start = time.time()
start_network_io()
while 1:
if is_network_io_complete(start):
break
yield  # hand control back to the event loop
# network I/O complete, response ready to be returned
return get_network_io_response()

def coroutine():
# our "easy" coroutine
# 1st line is blocking BUT DOES NOT BLOCK EVENT LOOP
# 1st line unblocks when network_io_coroutine returns
# This blocking then unblocking makes the code easy to write
response = yield from network_io_coroutine()
network_io_responses.append(response)

network_io_responses = []


To update the above to use asyncio, there are two things to change

• The two coroutines network_io_coroutine() and coroutine() should have the asyncio.coroutine decorator added to them (see page 543 of Fluent Python, 1st Edition - Luciano Ramalho for more details)
• The way to hand control back to the event loop in network_io_coroutine() is to use yield from asyncio.sleep(), not yield in an infinite while loop.

The updated code looks like this

import asyncio
import time

start = time.time()

def start_network_io():
pass

def get_network_io_response():
return 200

@asyncio.coroutine
def network_io_coroutine():
start_network_io()
yield from asyncio.sleep(3)
return get_network_io_response()

@asyncio.coroutine
def easy_coroutine():  # change name from coroutine to easy_coroutine
response = yield from network_io_coroutine()
network_io_responses.append(response)

network_io_responses = []


As a result, main() now looks a lot simpler

def main():
coro1 = easy_coroutine()
coro2 = easy_coroutine()
loop = asyncio.get_event_loop()
# "register" coroutines with event loop
loop.run_forever()
print('Network I/O responses', network_io_responses)
print(f'Script took {time.time() - start:.2f}s')


Unfortunately, the cost of this extra simplicity is quite prohibitive; the results are not displayed as anything after loop.run_forever() is never run.

Nonetheless, we can check the script works as it did before:

python -i /path/to/above/script.py,

wait three seconds or more, and hit CTRL+C; you should see a prompt as you are now inside a Python shell.

If you inspect the value of network_io_responses, you should see something like

>>> network_io_responses
[200, 200]


However, if you wait less than three seconds before hitting CTRL+C, neither of the two mocked network I/O operations will have returned a value yet

>>> network_io_responses
[]


Needless to say, there are a much cleaner ways to check our script works.

One solution is to run the event loop using run_until_complete() instead of run_forever().

When using run_forever(), we let the event loop know about our two coroutines with

loop.create_task(coro1)


create_task() returns a Task object wrapped around the coroutine given to it as an argument.

However, more importantly for us, it gets the event loop to schedule the coroutine given to it as an argument for execution.

When using run_until_complete(), we give the event loop one coroutine which it runs until completion.

But we have two coroutines we want to run in the event loop, coro1 and coro2.

The workaround is to create a third, “parent” coroutine which wraps around coro1 and coro2, and give this parent coroutine instead to run_until_complete().

The way to create the parent coroutine is to create an iterable containing coro1 and coro2, e.g. (coro1, coro2), and give the iterable to wait().

Making these changes, our script looks like

import asyncio
import time

start = time.time()

def start_network_io():
pass

def get_network_io_response():
return 200

@asyncio.coroutine
def network_io_coroutine():
start_network_io()
yield from asyncio.sleep(3)
return get_network_io_response()

@asyncio.coroutine
def easy_coroutine():
response = yield from network_io_coroutine()
network_io_responses.append(response)

network_io_responses = []

def main():
coro1 = easy_coroutine()
coro2 = easy_coroutine()
parent_coro = asyncio.wait((coro1, coro2))
loop = asyncio.get_event_loop()
loop.run_until_complete(parent_coro)
print('Network I/O responses', network_io_responses)
print(f'Script took {time.time() - start:.2f}s')

main()

Network I/O responses [200, 200]
Script took 3.00s


Adding a third coroutine is also easier now

def main():
coro1 = easy_coroutine()
coro2 = easy_coroutine()
coro3 = easy_coroutine()
parent_coro = asyncio.wait((coro1, coro2, coro3))
loop = asyncio.get_event_loop()
loop.run_until_complete(parent_coro)
print('Network I/O responses', network_io_responses)
print(f'Script took {time.time() - start:.2f}s')

main()

Network I/O responses [200, 200, 200]
Script took 3.00s


## Replace mock I/O operations with HTTP GET requests

We replace asyncio.sleep() in network_io_coroutine() with aiohttp.request()

import asyncio
import time
import aiohttp  # version 0.6.4

start = time.time()

@asyncio.coroutine
def network_io_coroutine():
r = yield from aiohttp.request('GET', 'https://pokeapi.co/api/v2/pokemon/1/')
return r.status

@asyncio.coroutine
def easy_coroutine():
response = yield from network_io_coroutine()
network_io_responses.append(response)

network_io_responses = []

def main():
coro1 = easy_coroutine()
coro2 = easy_coroutine()
coro3 = easy_coroutine()
parent_coro = asyncio.wait((coro1, coro2, coro3))
loop = asyncio.get_event_loop()
loop.run_until_complete(parent_coro)
print('Network I/O responses', network_io_responses)
print("Script took {:.2f}s".format(time.time() - start))  # Python version 3.4.0

main()

Network I/O responses [200, 200, 200]
Script took 0.39s


In theory, if we increase the number of requests, the running time should not change much

import asyncio
import time
import aiohttp

start = time.time()

@asyncio.coroutine
def network_io_coroutine(url):
r = yield from aiohttp.request('GET', url)
return r.status

@asyncio.coroutine
def easy_coroutine(url):
response = yield from network_io_coroutine(url)
network_io_responses.append(response)

network_io_responses = []

def main():
base_url = 'https://pokeapi.co/api/v2/pokemon/{}/'
coros = [easy_coroutine(base_url.format(i)) for i in range(1, 101)]
parent_coro = asyncio.wait(coros)
loop = asyncio.get_event_loop()
loop.run_until_complete(parent_coro)
expected_responses = [200] * 100
assert expected_responses == network_io_responses
print("Script took {:.2f}s".format(time.time() - start))

main()

Script took 1.43s


However, if we make the same number of requests synchronously,

import requests
import time

start = time.time()

network_io_responses = []

def main():
base_url = 'https://pokeapi.co/api/v2/pokemon/{}/'
for i in range(1, 101):
r = requests.get(base_url.format(i))
network_io_responses.append(r.status_code)
expected_responses = [200] * 100
assert expected_responses == network_io_responses
print("Script took {:.2f}s".format(time.time() - start))

main()

Script took 41.33s


Over 30 times slower!

From the examples in this post, we can see the benefits of concurrency via asynchronous programming when performing network I/O operations.

Disclaimer: In no way, shape, or form do I claim all the content in this post to be my own work / not copied, paraphrased, or derived in any other way from an external source.

To the best of my knowledge, all sources used are referenced. If you feel strongly about any of the content in this post from a plagarism, copyright, etc. point of view, please do not hesitate to get in touch to discuss and resolve the situation.

## References

Fluent Python, 1st Edition - Luciano Ramalho