AIORateLimiter: a RetryAfter halt is released early by any concurrent request finishing #5338

            except RetryAfter as exc:
                ...
                self._retry_after_event.clear()
                await asyncio.sleep(sleep)
            finally:
                # Allow other requests to be processed
                self._retry_after_event.set()

import asyncio, time
from telegram.error import RetryAfter
from telegram.ext import AIORateLimiter

async def main():
    rl = AIORateLimiter(overall_max_rate=0, group_max_rate=0, max_retries=1)

    async def flooded(*a, **kw):
        if not flooded.hit:
            flooded.hit = True
            raise RetryAfter(2)
        return True
    flooded.hit = False

    async def slow(*a, **kw):
        await asyncio.sleep(0.3)   # an ordinary in-flight request
        return True

    async def fast(*a, **kw):
        return True

    async def request(cb, cid):
        return await rl.process_request(callback=cb, args=(), kwargs={},
            endpoint="sendMessage", data={"chat_id": cid}, rate_limit_args=None)

    t0 = time.monotonic()
    b = asyncio.create_task(request(slow, 2))     # in flight first
    await asyncio.sleep(0.05)
    a = asyncio.create_task(request(flooded, 1))  # floods, halts the bot for 2s
    await asyncio.sleep(0.05)
    await b
    await request(fast, 3)                        # should wait out the halt
    print(f"third request sent {time.monotonic() - t0:.2f}s into a 2s halt")
    await a

asyncio.run(main())