Nested asyncio.run is parallel on first call but serial on second about nest_asyncio

[erdewit]

It's easier to see what happens if the script is rewritten as follows:

import asyncio
from datetime import datetime
import nest_asyncio

nest_asyncio.apply()

async def task(i):
    print(f"[{datetime.now()}] start running task {i}")
    asyncio.run(asyncio.sleep(1))
    print(f"[{datetime.now()}] task {i} ready")
    print(f"[{datetime.now()}] start running task {i + 5}")
    asyncio.run(asyncio.sleep(1))
    print(f"[{datetime.now()}] task {i + 5} ready")

async def parallel():
    tasks = [asyncio.create_task(task(i)) for i in range(5)]
    await asyncio.gather(*tasks, return_exceptions=True)


asyncio.run(parallel())

with the following output:

[2023-11-18 10:54:59.821213] start running task 0
[2023-11-18 10:54:59.821304] start running task 1
[2023-11-18 10:54:59.821341] start running task 2
[2023-11-18 10:54:59.821374] start running task 3
[2023-11-18 10:54:59.821400] start running task 4
[2023-11-18 10:55:00.822808] task 4 ready
[2023-11-18 10:55:00.822876] start running task 9
[2023-11-18 10:55:01.824596] task 9 ready
[2023-11-18 10:55:01.824738] task 3 ready
[2023-11-18 10:55:01.824762] start running task 8
[2023-11-18 10:55:02.826505] task 8 ready
[2023-11-18 10:55:02.826657] task 2 ready
[2023-11-18 10:55:02.826682] start running task 7
[2023-11-18 10:55:03.828482] task 7 ready
[2023-11-18 10:55:03.828613] task 1 ready
[2023-11-18 10:55:03.828636] start running task 6
[2023-11-18 10:55:04.829585] task 6 ready
[2023-11-18 10:55:04.829789] task 0 ready
[2023-11-18 10:55:04.829838] start running task 5
[2023-11-18 10:55:05.831698] task 5 ready

The first 5 tasks are scheduled at the same time and start executing at the same time. But the other 5 tasks only get scheduled when an earlier task has finished. Since task is synchronous, it will block other tasks from scheduling their next tasks. That's why the later tasks don't run in parallel here. They can be run in parallel with asyncio.create_task.