c++ - boost::asio io_service 线程池与 io_service::work

Question

我有一个 boost::asio 线程池，它有一个 asio::io_service和 asio::work :
boost::asio::io_service m_service;
boost::asio::io_service::work m_work;
在线程池析构函数中，我调用 m_service.stop()然后加入所有线程。但是，在进程关闭期间(在线程池析构函数中)间歇性地遇到此崩溃:

    #0  0x0000000000000000 in ?? ()
    #1  0x00007f4c8e59d160 in boost::asio::detail::task_io_service_operation::destroy (
        this=0x6d579f8)
        at /home/depot/Boost_1_61_0/boost/asio/detail/task_io_service_operation.hpp:43
    #2  0x00007f4c8e59d115 in boost::asio::detail::op_queue_access::destroy (o=0x6d579f8)
        at /home/depot/Boost_1_61_0/boost/asio/detail/op_queue.hpp:47
    #3  0x00007f4c8e59d044 in boost::asio::detail::op_queue::~op_queue (this=0x6d57a20)
        at /home/depot/Boost_1_61_0/boost/asio/detail/op_queue.hpp:81
    #4  0x00007f4c8e685d65 in boost::asio::detail::task_io_service::~task_io_service (
        this=0x6d57960)
        at /home/depot/Boost_1_61_0/boost/asio/detail/task_io_service.hpp:40
    #5  0x00007f4c8e685dc9 in boost::asio::detail::task_io_service::~task_io_service (
        this=0x6d57960)
        at /home/depot/Boost_1_61_0/boost/asio/detail/task_io_service.hpp:40
    #6  0x00007f4c8e5aa1ce in boost::asio::detail::service_registry::destroy (
        service=0x6d57960)
        at /home/depot/Boost_1_61_0/boost/asio/detail/impl/service_registry.ipp:101
    #7  0x00007f4c8e68537c in boost::asio::detail::service_registry::~service_registry (
        this=0x6d55020)
        at /home/depot/Boost_1_61_0/boost/asio/detail/impl/service_registry.ipp:45
    #8  0x00007f4c8e6852b0 in boost::asio::io_service::~io_service (
        this=0x7f4c8eb169c8 )
        at /home/depot/Boost_1_61_0/boost/asio/impl/io_service.ipp:53
    #9  0x00007f4c8e71d295 in crossplat::threadpool::~threadpool (
        this=0x7f4c8eb169b0 )
        at /home/depot/include/threadpool.h:87
    #10 0x00007f4c8aed6fe8 in __run_exit_handlers (status=0,
    ---Type  to continue, or q  to quit---
        listp=0x7f4c8b2605f8 , run_list_atexit=run_list_atexit@entry=true)
        at exit.c:82
    #11 0x00007f4c8aed7035 in __GI_exit (status=) at exit.c:104
    #12 0x00007f4c8aebd837 in __libc_start_main (main=0x1ce0980 ,
        argc=3, argv=0x7fffee2eda98, init=, fini=,
        rtld_fini=, stack_end=0x7fffee2eda88) at ../csu/libc-start.c:325
    #13 0x0000000001cdf159 in _start () at /usr/include/c++/v1/string:2027

asio 文档建议明确销毁 asio::work所有处理程序正常完成的对象。问题:

1. 是否需要调用io_service::stop()在显式破坏 asio::work 对象之后？
2. io_service::stop() 之间有什么区别？并摧毁io_service::work
3. 这个线程池析构函数看起来是否正确，它能帮助修复上述崩溃吗？我已将 m_work 更改为指针 std::unique_ptr<boost::asio::io_service::work> m_work; :
   ~threadpool()
{
m_work.reset(); // Allow io_service::run to exit
// join all threads
}

Answer 1

Is it required to call io_service::stop() after explicitly destructing the asio::work object?

没有。如果没有更多工作，服务将完成(轮询/运行将返回)。

What is the difference between io_service::stop() and destroying io_service::work

工作只是释放锁，允许服务完成。

io_service::stop() 强制停止事件循环，即使仍有待处理的工作。

Does this threadpool destructor look right, can it help fix the above crash.

是的，析构函数应该加入所有线程。您的代码没有显示它，所以我建议您验证一下。

想法

我注意到 threadpool 正在从退出处理程序中销毁。这对我来说意味着线程池可能具有静态存储持续时间。如果它还依赖于任何其他东西(比如，也许，一个静态的 io_service 实例？)你可能会遇到 SIOF : 静态初始化顺序惨败。