Hacktoberfest 2026: as issues que os mantenedores marcaram para outubro, abertas e boas para iniciantes. Ver issues do Hacktoberfest

asyncio: Windows socket reset during Proactor cleanup skips socket closure and transport detachment

Aberta
  • Algo errado?
#160,429 0 comentários 0 reações 0 responsáveis Ver no GitHub

Mantenedores costumam responder em até 1 dia

Ninguém assumiu esta issue ainda.

  • #160600 de @OsmnvAslan — fechado sem integrar

Avaliação

Dificuldade
3/5
Tempo estimado
1-2 dias
Facilidade para iniciantes
70/100
Tipo de issue
Bug
Clareza
Claramente especificada
Status de atividade
Ativa
Stack de tecnologia
python
Domínio
backend, operating-systems

Direção de pesquisa

Start by reading _ProactorBasePipeTransport._call_connection_lost() in Lib/asyncio/proactor_events.py and the related asyncio transport tests. Use the provided mock-based reproducer to check that a shutdown reset does not skip socket closure, server detachment, or completion marking, while exceptions from protocol.connection_lost() and unexpected shutdown failures remain visible. Done means those cleanup steps run for the expected reset and the regression is covered by a test.

Escrita pelo modelo de indexação a partir do texto da issue.

Descrição

stdlib topic-asyncio type-bug

Bug description

On Windows, _ProactorBasePipeTransport._call_connection_lost() can raise ConnectionResetError: [WinError 10054] from self._sock.shutdown(socket.SHUT_RDWR) while cleaning up a connection. The exception escapes into the event loop's exception handler before the subsequent socket closure, server detachment, and _called_connection_lost = True execute.

This was observed in an active application on CPython 3.14.5 (Windows), producing Exception in callback _ProactorBasePipeTransport._call_connection_lost() errors. The application continued running. The defect is the interrupted cleanup, beyond the unexpected error log: a socket can remain open until later finalization, and transport bookkeeping remains incomplete.

Deterministic reproducer

This standard-library-only example injects the observed socket-shutdown error and calls the unmodified CPython cleanup method. It demonstrates the skipped cleanup without depending on network timing; it does not reproduce the original TCP reset timing. Run it on Windows in a fresh interpreter without an application-installed asyncio patch.

from asyncio.proactor_events import _ProactorBasePipeTransport
from types import SimpleNamespace
from unittest.mock import Mock

reset = ConnectionResetError("connection reset during shutdown")
reset.winerror = 10054
sock = Mock()
sock.shutdown.side_effect = reset
server = Mock()
transport = SimpleNamespace(
    _called_connection_lost=False,
    _protocol=Mock(),
    _sock=sock,
    _server=server,
)

try:
    _ProactorBasePipeTransport._call_connection_lost(transport, None)
except ConnectionResetError:
    print("ConnectionResetError escaped cleanup")

print("socket.close calls:", sock.close.call_count)
print("server detach calls:", server._detach.call_count)
print("cleanup complete:", transport._called_connection_lost)

Observed output on CPython 3.14.5:

ConnectionResetError escaped cleanup
socket.close calls: 0
server detach calls: 0
cleanup complete: False

Expected behavior

An already-reset socket should still be closed and the transport detached and marked complete. Handle the expected reset around the socket shutdown operation, while ensuring remaining cleanup runs. Exceptions from protocol.connection_lost() and unexpected shutdown failures should remain observable; broadly suppressing exceptions around the whole callback would hide unrelated failures.

A potential improvement would prevent noisy error log messages by catching 'reasonable' shutdown failures and logging them via asyncio's debug logger (if debug mode is enabled for that proactor transport's event loop). Either way, shutdown errors should not cause cleanup to be skipped.

CPython versions tested on

CPython 3.14.5, Windows x86-64. Both the application traceback and the deterministic reproducer were observed on this version.

Source inspection, rather than execution on those interpreters, found the same method implementation in releases 3.14.6–3.14.9 and 3.15.0–3.15.1. See the 3.14.5 source and 3.15.1 source.

Related reports

  • The same shutdown traceback appears in a comment on #125981, a closed issue about SSL: APPLICATION_DATA_AFTER_CLOSE_NOTIFY. I could not find a dedicated issue tracking the interrupted Proactor socket cleanup.
  • #149388 concerned WinError 6 from PipeHandle.close() in this callback. That issue was fixed in PipeHandle; this report concerns WinError 10054 from the preceding socket shutdown() call.
Linked PRs
  • gh-160600
Linguagem predominante
Python
Estrelas
77.7k
Forks
38.3k
Merge médio
13h 43min
PRs com merge (30d)
704

Preparar o ambiente

Abrir no Codespaces

Inicia o contêiner de desenvolvimento do projeto no navegador, com a sua própria conta do GitHub.

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Mais de python/cpython

Todas as issues de python/cpython

Issues semelhantes

Mais issues de Python

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.