nanobind: leaked ... when a DuckDB type is referenced in overload signature and numpy is imported

未关闭
  • 哪里不对？
#732 5 条评论 3 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
25/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
numpy, python
领域
databases

调研方向

Start with the reproduction script in the issue: it needs only typing.overload referencing duckdb.DuckDBPyConnection plus import numpy, and the leak appears at interpreter exit. Run it with uv run --script to confirm the nanobind warnings. The cause is a reference-counting problem in the nanobind binding code, so the work is tracing which type or function objects are not released, which needs familiarity with nanobind. Done means the script exits without the leak warnings.

由索引模型根据 Issue 内容生成。

描述

needs triage
What happens?

When the interpreter exits, duckdb 2.0.0 prints nanobind leak warnings if both of these are true:

  • a DuckDB type (e.g. DuckDBPyConnection) is referenced from a typing.overload signature, which typing keeps in its module-level overload registry until shutdown
  • numpy is imported (torch 2.6.0 triggers it the same way)
nanobind: leaked 27 instances!
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 "fetchone"
 - leaked function "fetch_df"
 - leaked function "__exit__"
 - leaked function "_result_as_arrow"
 - leaked function ""
 - leaked function "abort"
 - leaked function "append"
 - leaked function "interrupt"
 - leaked function "dtype"
 - leaked function "_register_values"
 - ... 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

with duckdb_nanobind_leak.py as follows

# /// script
# requires-python = ">=3.13"
# dependencies = ["duckdb==2.0.0", "numpy==2.5.2"]
# ///
"""duckdb 2.0.0 prints "nanobind: leaked ..." at interpreter exit.

Run: uv run --script duckdb_nanobind_leak.py

Triggered when a DuckDB type is referenced from a `typing.overload` signature (typing keeps overloads in a
module-level registry until shutdown) and numpy is imported. Without the overload, or without numpy, exit is clean.
torch (2.6.0) triggers it the same way as numpy.
"""

from typing import overload

import duckdb
import numpy


@overload
def connect(database: str) -> duckdb.DuckDBPyConnection: ...
@overload
def connect(database: None) -> None: ...
def connect(database):
    return None if database is None else duckdb.connect(database)

run uv run --script duckdb_nanobind_leak.py

OS:

x86_64

DuckDB Version:

2.0.0

DuckDB Client:

Python

Hardware:

No response

Full Name:

Bartosz Marcinkowski

Affiliation:

MIM Fertility

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。