memory from transport buffer not freed after connection_lost

• uvloop version: 0.17.0
• Python version: 3.11.4
• Platform: fedora 38
• Can you reproduce the bug with PYTHONASYNCIODEBUG in env?: yes, with PYTHONASYNCIODEBUG=1
• Does uvloop behave differently from vanilla asyncio? How?: yes, keeps resident memory

If I flood transport.write() and then close the connection (e.g. having the client telnet session stop) the memory stays resident.
Under stock asyncio, the memory immediately is released.

To reproduce,

1. start this program,
2. watch it in top
3. , telnet to port 8888,
4. watch memory usage spike.
5. close the telnet session,
6. see that memory usage stays high.
7. repeat this experiment with stock asyncio runner,
8. the memory is freed immediately. after telnet session is closed

This test script is a malloc bomb, so make sure you watch memory closely or you will consume all the memory on your machine

#!/usr/bin/env python3


import asyncio
import uvloop


class EchoServerProtocol(asyncio.Protocol):
    def __init__(self, on_connect, on_disconnect):
        self.on_connect = on_connect
        self.on_disconnect = on_disconnect

    def connection_lost(self, transport):
        print('Connection lost')
        self.on_disconnect()
        self.transport = None

    def connection_made(self, transport):
        self.on_connect(transport)
        peername = transport.get_extra_info('peername')
        print('Connection from {}'.format(peername))
        self.transport = transport

    def data_received(self, data):
        message = data.decode()
        print('Data received: {!r}'.format(message))

        print('Send: {!r}'.format(message))
        self.transport.write(data)

        print('Close the client socket')


async def send_data(transport):
    message = b'Hello World!' * 1000
    while True:
        transport.write(message)
        await asyncio.sleep(0)


task = None


def on_connect(transport):
    global task
    task = asyncio.create_task(send_data(transport))


def on_disconnect():
    global task
    task.cancel()


async def start():
    # Get a reference to the event loop as we plan to use
    # low-level APIs.
    loop = asyncio.get_running_loop()

    server = await loop.create_server(
        lambda: EchoServerProtocol(on_connect, on_disconnect),
        '127.0.0.1',
        8888,
    )

    async with server:
        await server.serve_forever()


if __name__ == '__main__':
    # with asyncio.Runner() as runner:
    with asyncio.Runner(loop_factory=uvloop.new_event_loop) as runner:
        runner.run(start())

首先使用 uvloop.new_event_loop 运行所提供的 Python 复现，先观察 telnet 会话关闭后的内存情况，然后将其与标准的 asyncio runner 进行比较。跟踪 transport buffer 和 connection_lost 路径；当断开连接后内存能够像在 asyncio 下那样被释放时，即视为完成。