抄自維基百科 :
生產者消費者問題(英語:Producer-consumer problem),也稱有限緩衝問題(英語:Bounded-buffer problem),是一個多線程同步問題的經典案例。該問題描述了共享固定大小緩衝區的兩個線程——即所謂的“生產者”和“消費者”——在實際運行時會發生的問題。生產者的主要作用是生成一定量的數據放到緩衝區中,然後重複此過程。與此同時,消費者也在緩衝區消耗這些數據。該問題的關鍵就是要保證生產者不會在緩衝區滿時加入數據,消費者也不會在緩衝區中空時消耗數據。
要解決該問題,就必須讓生產者在緩衝區滿時休眠(要麼乾脆就放棄數據),等到下次消費者消耗緩衝區中的數據的時候,生產者才能被喚醒,開始往緩衝區添加數據。同樣,也可以讓消費者在緩衝區空時進入休眠,等到生產者往緩衝區添加數據之後,再喚醒消費者。
本文用一個ItemRepository類表示產品倉庫,其中包含一個數組和兩個座標表示的環形隊列、一個std::mutex成員、用來保證每次只被一個線程讀寫操作 (爲了保證打印出來的消息是一行一行的,在它空閒的時候也借用的這個互斥量╮(╯▽╰)╭)、兩個std::condition_variable表示隊列不滿和不空的狀態,進而保證生產的時候不滿,消耗的時候不空。
#pragma once
#include <chrono>//std::chrono
#include <mutex>//std::mutex,std::unique_lock,std::lock_guard
#include <thread>//std::thread
#include <condition_variable>//std::condition_variable
#include <iostream>//std::cout,std::endl
#include <map>//std::map
namespace MyProducerToConsumer {
static const int gRepositorySize = 10;//total size of the repository
static const int gItemNum = 97;//number of products to produce
std::mutex produce_mtx, consume_mtx;//mutex for all the producer thread or consumer thread
std::map<std::thread::id, int> threadPerformance;//records of every thread's producing/consuming number
struct ItemRepository {//repository class
int m_ItemBuffer[gRepositorySize];//Repository itself (as a circular queue)
int m_ProducePos;//rear position of circular queue
int m_ConsumePos;//head position of circular queue
std::mutex m_mtx;//mutex for operating the repository
std::condition_variable m_RepoUnfull;//indicating that this repository is unfull(then producers can produce items)
std::condition_variable m_RepoUnempty;//indicating that this repository is unempty(then consumers can produce items)
}gItemRepo;
void ProduceItem(ItemRepository *ir, int item) {
std::unique_lock <std::mutex>ulk(ir->m_mtx);
while ((ir->m_ProducePos + 1) % gRepositorySize == ir->m_ConsumePos) {//full(spare one slot for indicating)
std::cout << "Reposity is full. Waiting for consumers..." << std::endl;
ir->m_RepoUnfull.wait(ulk);//unlocking ulk and waiting for unfull condition
}
//when unfull
ir->m_ItemBuffer[ir->m_ProducePos++] = item;//procude and shift
std::cout << "Item No." << item << " produced successfully by "
<<std::this_thread::get_id()<<"!" << std::endl;
threadPerformance[std::this_thread::get_id()]++;
if (ir->m_ProducePos == gRepositorySize)//loop
ir->m_ProducePos = 0;
ir->m_RepoUnempty.notify_all();//item produced, so it's unempty; notify all consumers
}
int ConsumeItem(ItemRepository *ir) {
std::unique_lock<std::mutex>ulk(ir->m_mtx);
while (ir->m_ConsumePos == ir->m_ProducePos) {//empty
std::cout << "Repository is empty.Waiting for producing..." << std::endl;
ir->m_RepoUnempty.wait(ulk);
}
int item = ir->m_ItemBuffer[ir->m_ConsumePos++];
std::cout << "Item No." << item << " consumed successfully by "
<<std::this_thread::get_id()<<"!" << std::endl;
threadPerformance[std::this_thread::get_id()]++;
if (ir->m_ConsumePos == gRepositorySize)
ir->m_ConsumePos = 0;
ir->m_RepoUnfull.notify_all();//item consumed, so it's unempty; notify all consumers
return item;
}
void ProducerThread() {
static int produced = 0;//static variable to indicate the number of produced items
while (1) {
std::this_thread::sleep_for(std::chrono::milliseconds(10));//sleep long enough in case it runs too fast for other threads to procude
std::lock_guard<std::mutex>lck(produce_mtx);//auto unlock when break
produced++;
if (produced > gItemNum)break;
gItemRepo.m_mtx.lock();
std::cout << "Producing item No." << produced << "..." << std::endl;
gItemRepo.m_mtx.unlock();
ProduceItem(&gItemRepo, produced);
}
gItemRepo.m_mtx.lock();
std::cout << "Producer thread " << std::this_thread::get_id()
<< " exited." << std::endl;
gItemRepo.m_mtx.unlock();
}
void ConsumerThread() {
static int consumed = 0;
while (1) {
std::this_thread::sleep_for(std::chrono::milliseconds(10));
std::lock_guard<std::mutex>lck(consume_mtx);
consumed++;
if (consumed > gItemNum)break;
gItemRepo.m_mtx.lock();
std::cout << "Consuming item available..." << std::endl;
gItemRepo.m_mtx.unlock();
ConsumeItem(&gItemRepo);
}
gItemRepo.m_mtx.lock();
std::cout << "Consumer thread " << std::this_thread::get_id()
<< " exited." << std::endl;
gItemRepo.m_mtx.unlock();
}
void InitItemRepository(ItemRepository* ir) {
ir->m_ConsumePos = 0;
ir->m_ProducePos = 0;
}
void Run() {
InitItemRepository(&gItemRepo);
std::thread thdConsume[11];
std::thread thdProduce[11];
for (auto& t : thdConsume)t = std::thread(ConsumerThread);
for (auto& t : thdProduce)t = std::thread(ProducerThread);
for (auto& t : thdConsume)t.join();
for (auto& t : thdProduce)t.join();
for (auto& iter : threadPerformance)cout << iter.first << ":" << iter.second << endl;
}
}