Leaked instances of DuckDBPyType since 2.0

未关闭
  • 哪里不对？
#736 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
48/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
python
领域
backend

调研方向

Start by running the pure Python reproducer with DuckDB 2.0, then compare it with the torch and typing.overload scenario. The issue points to DuckDBPyConnection annotations being retained during shutdown and reports leaks in nanobind bindings, but names no source file or test. Done means the reproducer exits without nanobind leak reports.

由索引模型根据 Issue 内容生成。

描述

needs triage
What happens?

Since 2.0, our runtimes end with what looks like leaked instances of DuckDBPyType.

nanobind: leaked 27 instances!
 - leaked instance 0x64c7922f1c50 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922f1c10 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e85b0 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e86d0 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e85d0 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8630 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8610 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8730 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922f1c70 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e87d0 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8670 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8690 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8810 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922f1c30 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8590 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8770 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e87b0 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e85f0 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e86f0 of type "_duckdb._sqltypes.DuckDBPyType"
 - leaked instance 0x64c7922e8710 of type "_duckdb._sqltypes.DuckDBPyType"
 - ... skipped remainder
nanobind: leaked 6 types!
 - leaked type "_duckdb._sqltypes.DuckDBPyType"
 - leaked type "_duckdb.DuckDBPyConnection"
 - leaked type "FunctionNullHandling"
 - leaked type "PythonUDFType"
 - leaked type "ExplainType"
 - leaked type "PythonExceptionHandling"
nanobind: leaked 84 functions!
 - leaked function "extract_statements"
 - leaked function "__exit__"
 - leaked function "fetchdf"
 - leaked function "_statement_count"
 - leaked function "install_extension"
 - leaked function "array_type"
 - leaked function ""
 - leaked function "interrupt"
 - leaked function "_own_sources"
 - leaked function "torch"
 - ... skipped remainder
nanobind: this is likely caused by a reference counting issue in the binding code.
See https://nanobind.readthedocs.io/en/latest/refleaks.html
To Reproduce

It seems this is due to interaction with other libs, torch in particular, and typing.overload.

This is a pure Python reproducer

import ctypes

import duckdb

ctypes.pythonapi.Py_IncRef(ctypes.py_object(duckdb.DuckDBPyConnection))

Here is a more realistic scenario.

$ cat lib.py
import typing

import duckdb


class Wrapper:
    def __init__(self, con: duckdb.DuckDBPyConnection): ...


# typing.overload stores the function in typing._overload_registry, which is not
# released at shutdown once torch has been imported.
@typing.overload
def f(x: int) -> None: ...

$ cat repro_torch.py
"""duckdb + torch + a module with a typing.overload and a DuckDBPyConnection annotation."""
import torch  # noqa: F401

import lib  # noqa: F401

Running python repro_torch.py provides the same error message.

OS:

Ubuntu 26

DuckDB Version:

2.0

DuckDB Client:

Python

Hardware:

No response

Full Name:

Jakob Stigenberg

Affiliation:

Qubos Systematic

Did you include all relevant configuration (e.g., CPU architecture, Linux distribution) to reproduce the issue?
  • Yes, I have
Did you include all code required to reproduce the issue?
  • Yes, I have
Did you include all relevant data sets for reproducing the issue?

Not applicable - the reproduction does not require a data set

主要语言
Python
星标
203
派生
124
平均合并
13 小时 36 分钟
30 天内合并 PR
61

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

duckdb/duckdb-python 的其他 Issue

查看 duckdb/duckdb-python 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。