asyncio.gather和asyncio.wait似乎有类似的用途：我有一堆我想要执行/等待的异步事物（不一定等到一个在下一个启动之前完成）。它们使用不同的语法，并且在一些细节上有所不同，但是对于我而言，拥有2个功能上具有如此巨大重叠的功能似乎非常不灵活。我错过了什么？
答案 0 :(得分：108)
虽然在一般情况下类似（“运行并获得许多任务的结果”），但每个函数都有一些针对其他情况的特定功能：
asyncio.gather() 返回Future实例，允许高级别的任务分组：
import asyncio
from pprint import pprint
import random
async def coro(tag):
print(">", tag)
await asyncio.sleep(random.uniform(1, 3))
print("<", tag)
return tag
loop = asyncio.get_event_loop()
group1 = asyncio.gather(*[coro("group 1.{}".format(i)) for i in range(1, 6)])
group2 = asyncio.gather(*[coro("group 2.{}".format(i)) for i in range(1, 4)])
group3 = asyncio.gather(*[coro("group 3.{}".format(i)) for i in range(1, 10)])
all_groups = asyncio.gather(group1, group2, group3)
results = loop.run_until_complete(all_groups)
loop.close()
pprint(results)
可以通过致电group2.cancel()甚至all_groups.cancel()取消群组中的所有任务。另请参阅.gather(..., return_exceptions=True)，
asyncio.wait() 支持在第一个任务完成后或在指定的超时后等待停止，从而允许更低级别的操作精度：
import asyncio
import random
async def coro(tag):
print(">", tag)
await asyncio.sleep(random.uniform(0.5, 5))
print("<", tag)
return tag
loop = asyncio.get_event_loop()
tasks = [coro(i) for i in range(1, 11)]
print("Get first result:")
finished, unfinished = loop.run_until_complete(
asyncio.wait(tasks, return_when=asyncio.FIRST_COMPLETED))
for task in finished:
print(task.result())
print("unfinished:", len(unfinished))
print("Get more results in 2 seconds:")
finished2, unfinished2 = loop.run_until_complete(
asyncio.wait(unfinished, timeout=2))
for task in finished2:
print(task.result())
print("unfinished2:", len(unfinished2))
print("Get all other results:")
finished3, unfinished3 = loop.run_until_complete(asyncio.wait(unfinished2))
for task in finished3:
print(task.result())
loop.close()
答案 1 :(得分：21)
asyncio.wait的级别低于asyncio.gather。
顾名思义，asyncio.gather主要侧重于收集结果。它会等待一堆期货并以给定的顺序返回结果。
asyncio.wait等待期货。而不是直接给你结果，它提供完成和待处理的任务。你必须手工收集价值观。
此外，您可以指定等待所有期货结束，或者只指定wait的第一个期货。
答案 2 :(得分：9)
我还注意到你可以通过简单地指定列表在wait（）中提供一组协同程序：
result=loop.run_until_complete(asyncio.wait([
say('first hello', 2),
say('second hello', 1),
say('third hello', 4)
]))
而在gather（）中的分组只需指定多个协同程序即可完成：
result=loop.run_until_complete(asyncio.gather(
say('first hello', 2),
say('second hello', 1),
say('third hello', 4)
))
答案 3 :(得分：6)
当涉及到异常时，这两个函数的默认行为是一个很容易忽略的非常重要的区别。
我将使用此示例模拟协程，该协程有时会引发异常-
import asyncio
import random
async def a_flaky_tsk(i):
await asyncio.sleep(i) # bit of fuzz to simulate a real-world example
if i % 2 == 0:
print(i, "ok")
else:
print(i, "crashed!")
raise ValueError
coros = [a_flaky_tsk(i) for i in range(10)]
await asyncio.gather(*coros)输出-
0 ok
1 crashed!
Traceback (most recent call last):
File "/Users/dev/PycharmProjects/trading/xxx.py", line 20, in <module>
asyncio.run(main())
File "/Users/dev/.pyenv/versions/3.8.2/lib/python3.8/asyncio/runners.py", line 43, in run
return loop.run_until_complete(main)
File "/Users/dev/.pyenv/versions/3.8.2/lib/python3.8/asyncio/base_events.py", line 616, in run_until_complete
return future.result()
File "/Users/dev/PycharmProjects/trading/xxx.py", line 17, in main
await asyncio.gather(*coros)
File "/Users/dev/PycharmProjects/trading/xxx.py", line 12, in a_flaky_tsk
raise ValueError
ValueError
如您所见，索引1之后的花魁从未执行。
但是await asyncio.wait(coros)继续执行任务，即使其中一些失败-
0 ok
1 crashed!
2 ok
3 crashed!
4 ok
5 crashed!
6 ok
7 crashed!
8 ok
9 crashed!
Task exception was never retrieved
future: <Task finished name='Task-10' coro=<a_flaky_tsk() done, defined at /Users/dev/PycharmProjects/trading/xxx.py:6> exception=ValueError()>
Traceback (most recent call last):
File "/Users/dev/PycharmProjects/trading/xxx.py", line 12, in a_flaky_tsk
raise ValueError
ValueError
Task exception was never retrieved
future: <Task finished name='Task-8' coro=<a_flaky_tsk() done, defined at /Users/dev/PycharmProjects/trading/xxx.py:6> exception=ValueError()>
Traceback (most recent call last):
File "/Users/dev/PycharmProjects/trading/xxx.py", line 12, in a_flaky_tsk
raise ValueError
ValueError
Task exception was never retrieved
future: <Task finished name='Task-2' coro=<a_flaky_tsk() done, defined at /Users/dev/PycharmProjects/trading/xxx.py:6> exception=ValueError()>
Traceback (most recent call last):
File "/Users/dev/PycharmProjects/trading/xxx.py", line 12, in a_flaky_tsk
raise ValueError
ValueError
Task exception was never retrieved
future: <Task finished name='Task-9' coro=<a_flaky_tsk() done, defined at /Users/dev/PycharmProjects/trading/xxx.py:6> exception=ValueError()>
Traceback (most recent call last):
File "/Users/dev/PycharmProjects/trading/xxx.py", line 12, in a_flaky_tsk
raise ValueError
ValueError
Task exception was never retrieved
future: <Task finished name='Task-3' coro=<a_flaky_tsk() done, defined at /Users/dev/PycharmProjects/trading/xxx.py:6> exception=ValueError()>
Traceback (most recent call last):
File "/Users/dev/PycharmProjects/trading/xxx.py", line 12, in a_flaky_tsk
raise ValueError
ValueError
当然，可以通过使用-
更改两者的行为 asyncio.gather(..., return_exceptions=True)
或
asyncio.wait([...], return_when=asyncio.FIRST_EXCEPTION)
但这还不止于此！
注意：
Task exception was never retrieved
在上面的日志中。
asyncio.wait()不会重新引发子任务中的异常，除非您单独await。 （日志中的stacktrace只是消息，无法捕获！）
done, pending = await asyncio.wait(coros)
for tsk in done:
try:
await tsk
except Exception as e:
print("I caught:", repr(e))
输出-
0 ok
1 crashed!
2 ok
3 crashed!
4 ok
5 crashed!
6 ok
7 crashed!
8 ok
9 crashed!
I caught: ValueError()
I caught: ValueError()
I caught: ValueError()
I caught: ValueError()
I caught: ValueError()
另一方面，要捕获asyncio.gather()的异常，您必须-
results = await asyncio.gather(*coros, return_exceptions=True)
for result_or_exc in results:
if isinstance(result_or_exc, Exception):
print("I caught:", repr(result_or_exc))
（输出与以前相同）
答案 4 :(得分：2)
除了前面的所有答案外，我想谈一谈gather()和wait()在取消被取消的情况下的不同行为。
如果gather()被取消，所有已提交的等待中（尚未完成）也将被取消。
如果wait()任务被取消，它只会抛出一个CancelledError并且等待的任务保持不变。
简单的例子：
import asyncio
async def task(arg):
await asyncio.sleep(5)
return arg
async def cancel_waiting_task(work_task, waiting_task):
await asyncio.sleep(2)
waiting_task.cancel()
try:
await waiting_task
print("Waiting done")
except asyncio.CancelledError:
print("Waiting task cancelled")
try:
res = await work_task
print(f"Work result: {res}")
except asyncio.CancelledError:
print("Work task cancelled")
async def main():
work_task = asyncio.create_task(task("done"))
waiting = asyncio.create_task(asyncio.wait({work_task}))
await cancel_waiting_task(work_task, waiting)
work_task = asyncio.create_task(task("done"))
waiting = asyncio.gather(work_task)
await cancel_waiting_task(work_task, waiting)
asyncio.run(main())
输出：
asyncio.wait()
Waiting task cancelled
Work result: done
----------------
asyncio.gather()
Waiting task cancelled
Work task cancelled
有时必须结合使用wait()和gather()功能。例如，我们要等待至少完成一个 个任务，然后等待取消其余待处理的任务，如果waiting本身是取消，然后也取消所有待处理的任务。
作为真实示例，我们有一个断开连接事件和一个工作任务。我们想等待工作任务的结果，但是如果连接丢失，则取消它。或者，我们将提出几个并行请求，但是在至少完成一个响应后，取消所有其他请求。
可以通过以下方式完成：
import asyncio
from typing import Optional, Tuple, Set
async def wait_any(
tasks: Set[asyncio.Future], *, timeout: Optional[int] = None,
) -> Tuple[Set[asyncio.Future], Set[asyncio.Future]]:
tasks_to_cancel: Set[asyncio.Future] = set()
try:
done, tasks_to_cancel = await asyncio.wait(
tasks, timeout=timeout, return_when=asyncio.FIRST_COMPLETED
)
return done, tasks_to_cancel
except asyncio.CancelledError:
tasks_to_cancel = tasks
raise
finally:
for task in tasks_to_cancel:
task.cancel()
async def task():
await asyncio.sleep(5)
async def cancel_waiting_task(work_task, waiting_task):
await asyncio.sleep(2)
waiting_task.cancel()
try:
await waiting_task
print("Waiting done")
except asyncio.CancelledError:
print("Waiting task cancelled")
try:
res = await work_task
print(f"Work result: {res}")
except asyncio.CancelledError:
print("Work task cancelled")
async def check_tasks(waiting_task, working_task, waiting_conn_lost_task):
try:
await waiting_task
print("waiting is done")
except asyncio.CancelledError:
print("waiting is cancelled")
try:
await waiting_conn_lost_task
print("connection is lost")
except asyncio.CancelledError:
print("waiting connection lost is cancelled")
try:
await working_task
print("work is done")
except asyncio.CancelledError:
print("work is cancelled")
async def work_done_case():
working_task = asyncio.create_task(task())
connection_lost_event = asyncio.Event()
waiting_conn_lost_task = asyncio.create_task(connection_lost_event.wait())
waiting_task = asyncio.create_task(wait_any({working_task, waiting_conn_lost_task}))
await check_tasks(waiting_task, working_task, waiting_conn_lost_task)
async def conn_lost_case():
working_task = asyncio.create_task(task())
connection_lost_event = asyncio.Event()
waiting_conn_lost_task = asyncio.create_task(connection_lost_event.wait())
waiting_task = asyncio.create_task(wait_any({working_task, waiting_conn_lost_task}))
await asyncio.sleep(2)
connection_lost_event.set() # <---
await check_tasks(waiting_task, working_task, waiting_conn_lost_task)
async def cancel_waiting_case():
working_task = asyncio.create_task(task())
connection_lost_event = asyncio.Event()
waiting_conn_lost_task = asyncio.create_task(connection_lost_event.wait())
waiting_task = asyncio.create_task(wait_any({working_task, waiting_conn_lost_task}))
await asyncio.sleep(2)
waiting_task.cancel() # <---
await check_tasks(waiting_task, working_task, waiting_conn_lost_task)
async def main():
print("Work done")
print("-------------------")
await work_done_case()
print("\nConnection lost")
print("-------------------")
await conn_lost_case()
print("\nCancel waiting")
print("-------------------")
await cancel_waiting_case()
asyncio.run(main())
输出：
Work done
-------------------
waiting is done
waiting connection lost is cancelled
work is done
Connection lost
-------------------
waiting is done
connection is lost
work is cancelled
Cancel waiting
-------------------
waiting is cancelled
waiting connection lost is cancelled
work is cancelled