strand 在 boost asio 中的优势是什么？

Question

据我了解，学习 boost asio 并找出一个名为“strand”的类。 如果只有一个 io_service 关联到特定 strand 并按 strand 发布句柄。

示例（来自here）

boost::shared_ptr< boost::asio::io_service > io_service( 
    new boost::asio::io_service
);
boost::shared_ptr< boost::asio::io_service::work > work(
    new boost::asio::io_service::work( *io_service )
);
boost::asio::io_service::strand strand( *io_service );

boost::thread_group worker_threads;
for( int x = 0; x < 2; ++x )
{
    worker_threads.create_thread( boost::bind( &WorkerThread, io_service ) );
}

boost::this_thread::sleep( boost::posix_time::milliseconds( 1000 ) );

strand.post( boost::bind( &PrintNum, 1 ) );
strand.post( boost::bind( &PrintNum, 2 ) );
strand.post( boost::bind( &PrintNum, 3 ) );
strand.post( boost::bind( &PrintNum, 4 ) );
strand.post( boost::bind( &PrintNum, 5 ) );

然后 strand 将为我们序列化处理程序执行。但是这样做有什么好处呢？如果我们希望任务变为序列化？

Answer 1

想象一个系统，其中单个io_service 管理数百个网络连接的套接字。为了能够并行化工作负载，系统维护了一个调用 io_service::run 的工作线程池。

现在这种系统中的大部分操作都可以并行运行。但有些必须被序列化。例如，您可能不希望在同一个套接字上同时发生多个写操作。然后，您将使用每个套接字的一个链来同步写入：不同套接字上的写入仍然可以同时发生，而对相同套接字的写入将被序列化。工作线程不必关心同步或不同的套接字，它们只需抓住io_service::run 提供的任何东西。

有人可能会问：为什么我们不能只使用互斥锁来代替同步呢？ strand 的优点是，如果 strand 已经在工作，则不会首先安排工作线程。使用互斥体，工作线程会收到回调，然后会阻塞锁定尝试，阻止线程做任何有用的工作，直到互斥体可用。

Answer 2

我知道它太旧了，但希望它能帮助新用户通过示例来理解。读取代码中的cmets

#define BOOST_DATE_TIME_NO_LIB
#define BOOST_REGEX_NO_LIB

#include <boost/asio.hpp>
#include <boost/shared_ptr.hpp>
#include <boost/thread.hpp>
#include <boost/thread/mutex.hpp>
#include <boost/bind.hpp>
#include <iostream>

boost::mutex global_stream_lock;

void WorkerThread(boost::shared_ptr<boost::asio::io_service> iosvc, int counter) {
    global_stream_lock.lock();
    std::cout << "Thread " << std::this_thread::get_id() << ", " << counter << " Start.\n";
    global_stream_lock.unlock();

    iosvc->run();

    global_stream_lock.lock();
    std::cout << "Thread " << counter << " End.\n";
    global_stream_lock.unlock();
}

void async_send_handler(int number) {
    std::cout << "Number: " << number << ", threadID: " << std::this_thread::get_id() << std::endl;
}

int main(void) {
    boost::shared_ptr<boost::asio::io_service> io_svc(
        new boost::asio::io_service
    );

    boost::shared_ptr<boost::asio::io_service::work> worker(
        new boost::asio::io_service::work(*io_svc)
    );

    boost::asio::io_service::strand strand(*io_svc);

    global_stream_lock.lock();
    std::cout << "The program will exit once all work has finished.\n";
    global_stream_lock.unlock();

    boost::thread_group threads;
    for( int i = 1; i <= 5; i++ )
        threads.create_thread(boost::bind(&WorkerThread, io_svc, i));

    boost::this_thread::sleep(boost::posix_time::milliseconds(500));

    // Imagine you are invoking async_send on tcp or udp socket several times
    // and you want the handlers of this async_send call to be invoked sequentially

    // This code is almost equal to calling handlers of socket.async_send.
    // The handlers are invoked concurently and the order might be arbitrary
    io_svc->post(boost::bind(&async_send_handler, 1));
    io_svc->post(boost::bind(&async_send_handler, 2));
    io_svc->post(boost::bind(&async_send_handler, 3));
    io_svc->post(boost::bind(&async_send_handler, 4));
    io_svc->post(boost::bind(&async_send_handler, 5));

    // This code will do what you exactly want;
    // It will execute the handlers sequentially in that order
    strand.post(boost::bind(&async_send_handler, 1));
    strand.post(boost::bind(&async_send_handler, 2));
    strand.post(boost::bind(&async_send_handler, 3));
    strand.post(boost::bind(&async_send_handler, 4));
    strand.post(boost::bind(&async_send_handler, 5));

    worker.reset();

    threads.join_all();

    return 0;
}