Why do coroutines, when run with asyncio, even if they exit gracefully, remain in memory?

Question

Python

Good afternoon. Essence of the question:
a limited pool of coroutines is used, when filling it is expected using the following construction:
done, pending = await asyncio.wait(pool, return_when=asyncio.FIRST_COMPLETED)
Moreover, if you constantly display the number of all tasks in memory using asyncio.Task. all_tasks(loop), then it is constantly growing, i.e. I delete even completed tasks from the pool, but they hang in memory. When there are more than 940, I start getting "MemoryError. Fatal read error on socket transport..." exceptions. I suppose the problem is aiohttp.get mishandling, but specifically.. What is wrong with them or me? I am attaching the code. https://gist.github.com/anonymous/3b4b4e2357c75de0...

Answer 1

In general, the operation scheme of the event machine is such that completed tasks will be thrown out into the cold at the next opportunity, and will not necessarily be deleted immediately. A couple of scenarios were tested, when a large number of tasks spawn, the reactor starts to discard the excess more quickly. The memory error was obtained as a result of an oversight of a small but unpleasant bug.