initialize_globals()在边缘场景下的获取堆栈卡住问题

[yangrudan]

问题

python进程调用so库抢占GIL锁时, 获取堆栈发送命令后会卡住

[yangrudan]

过程

PROBING=1 python main.py, 非主线程看起来依赖gil锁的获取

这种情况通过另一个终端probing -t <pid> bt卡住, 请求无法返回

[yangrudan]

enhance: add timeout monitoring to command-line tools to prevent them… #37 超时机制效果

[reiase]

能提供下测试的python脚本吗

[yangrudan]

能提供下测试的python脚本吗

add.cpp

#include <pybind11/pybind11.h>
#include <thread> // 包含线程相关的头文件
#include <chrono> // 
#include <stdio.h>                
int add(int a, int b) {
    while(1) {
        std::this_thread::sleep_for(std::chrono::seconds(2)); // 每2秒暂停一次
    }
    //printf("aaaaaa\n");
    //std::this_thread::sleep_for(std::chrono::seconds(2)); // 每秒暂停一次
        // 可以在这里加入一些逻辑，比如打印日志等
        // std::cout << "Stuck in an infinite loop..." << std::endl;
    return a + b;
}

int my_add(int a, int b){
    return add(a, b);
}

int my_add_function(int a, int b){
    return my_add(a, b);
}

PYBIND11_MODULE(add, m) {
    m.def("add_aa", &my_add_function, "A function that adds two numbers");
}

main.py

import add
import time
import threading


# result = add.add_aa(3, 5)
# print(f"The result is: {result}")
def add_numbers_2():
    add_numbers_1()


def add_numbers_1():
    add_numbers()

def add_numbers():
    while True:
        time.sleep(10)
        result = add.add_aa(3, 5)
    #print(f"The result is: {result}")

if __name__ == "__main__":
    add_numbers_2()

setup.py

from setuptools import setup, Extension
from pybind11.setup_helpers import Pybind11Extension, build_ext

ext_modules = [
    Pybind11Extension(
        "add",
        ["add.cpp"],
        extra_compile_args=['-g', '-O0'],  # 添加-O0禁用优化，增强调试效果
        define_macros=[('DEBUG', '1')],  # 可选：定义DEBUG宏
    ),
]

setup(
    name="add",
    ext_modules=ext_modules,
    cmdclass={"build_ext": build_ext},
)

运行命令

PROBING=1 PROBING_PORT=9922 python main.py

[yangrudan]

_core模块调用initialize_globals()引入了GIL锁的依赖 @reiase