How to implement blocking read using POSIX threads

Question

I would like to implement a producer/consumer scenario that obeys interfaces that are roughly:

class Consumer {
private:
    vector<char> read(size_t n) {
        // If the internal buffer has `n` elements, then dequeue them
        // Otherwise wait for more data and try again
    }
public:
    void run() {
        read(10);
        read(4839);
        // etc
    }
    void feed(const vector<char> &more) {
        // Safely queue the data
        // Notify `read` that there is now more data
    }
};

In this case, feed and run will run on separate threads and read should be a blocking read (like recv and fread). Obviously, I will need some kind of mutual exclusion on my deque, and I will need some kind of notification system to inform read to try again.

I hear condition variables are the way to go, but all my multithreading experience lies with Windows and am having a hard time wrapping my head around them.

Thanks for any help!

(Yes, I know it's inefficient to return vectors. Let's not get into that.)

Answer 1

This code is not production ready. No error checking is done on the results of any library calls.

I have wrapped the lock/unlock of the mutex in LockThread so it is exception safe. But that's about it.

In addition if I was doing this seriously I would wrap the mutex and condition variables inside objects so they can cot be abused inside other methods of Consumer. But as long as you take note that the lock must be acquired before you use the condition variable (in any way) then this simple situation can stand as is.

Out of interest have you checked the boost threading library?

#include <iostream>
#include <vector>
#include <pthread.h>

class LockThread
{
    public:
    LockThread(pthread_mutex_t& m)
        :mutex(m)
    {
        pthread_mutex_lock(&mutex);
    }
    ~LockThread()
    {
        pthread_mutex_unlock(&mutex);
    }
    private:
        pthread_mutex_t& mutex;
};
class Consumer
{
    pthread_mutex_t     lock;
    pthread_cond_t      cond;
    std::vector<char>   unreadData;
    public:
    Consumer()
    {
        pthread_mutex_init(&lock,NULL);
        pthread_cond_init(&cond,NULL);
    }
    ~Consumer()
    {
        pthread_cond_destroy(&cond);
        pthread_mutex_destroy(&lock);
    }

    private:
        std::vector<char> read(size_t n)
        {
            LockThread  locker(lock);
            while (unreadData.size() < n)
            {
                // Must wait until we have n char.
                // This is a while loop because feed may not put enough in.

                // pthread_cond() releases the lock.
                // Thread will not be allowed to continue until
                // signal is called and this thread reacquires the lock.

                pthread_cond_wait(&cond,&lock);

                // Once released from the condition you will have re-aquired the lock.
                // Thus feed() must have exited and released the lock first.
            }

            /*
             * Not sure if this is exactly what you wanted.
             * But the data is copied out of the thread safe buffer
             * into something that can be returned.
             */
            std::vector<char>   result(n); // init result with size n
            std::copy(&unreadData[0],
                      &unreadData[n],
                      &result[0]);

            unreadData.erase(unreadData.begin(),
                             unreadData.begin() + n);
            return (result);
        }
public:
    void run()
    {
        read(10);
        read(4839);
        // etc
    }
    void feed(const std::vector<char> &more)
    {
        LockThread  locker(lock);

        // Once we acquire the lock we can safely modify the buffer.
        std::copy(more.begin(),more.end(),std::back_inserter(unreadData));

        // Only signal the thread if you have the lock
        // Otherwise race conditions happen.
        pthread_cond_signal(&cond);

        // destructor releases the lock and thus allows read thread to continue.
    }
};


int main()
{
    Consumer    c;
}

Answer 2

I tend to use what I call a "Syncronized Queue". I wrap the normal queue and use a Semaphore class for both locking and making read block just as you desire:

#ifndef SYNCQUEUE_20061005_H_
#define SYNCQUEUE_20061005_H_

#include <queue>
#include "Semaphore.h"

// similar, but slightly simpler interface to std::queue
// this queue implementation will serialize pushes and pops
// and block on a pop while empty (as apposed to throwing an exception)
// it also locks as neccessary on insertion and removal to avoid race 
// conditions

template <class T, class C = std::deque<T> > class SyncQueue {
protected:
    std::queue<T, C>    m_Queue;
    Semaphore           m_Semaphore;
    Mutex               m_Mutex;

public:
    typedef typename std::queue<T, C>::value_type value_type;
    typedef typename std::queue<T, C>::size_type size_type;

    explicit SyncQueue(const C& a = C()) : m_Queue(a), m_Semaphore(0) {}

    bool empty() const              { return m_Queue.empty(); }
    size_type size() const          { return m_Queue.size(); }

    void push(const value_type& x);
    value_type pop();
};

template <class T, class C>
void SyncQueue<T, C>::push(const SyncQueue<T, C>::value_type &x) {
    // atomically push item
    m_Mutex.lock(); 
    m_Queue.push(x); 
    m_Mutex.unlock(); 

    // let blocking semaphore know another item has arrived
    m_Semaphore.v();
}

template <class T, class C>
typename SyncQueue<T, C>::value_type SyncQueue<T, C>::pop() {
    // block until we have at least one item
    m_Semaphore.p();

    // atomically read and pop front item
    m_Mutex.lock();
    value_type ret = m_Queue.front();
    m_Queue.pop();
    m_Mutex.unlock();

    return ret;
}

#endif

You can implement semaphores and mutexes with the appropriate primitives in your threading implementation.

NOTE: this implementation is an example for single elements in a queue, but you could easily wrap this with a function which buffers results until N have been provided. something like this if it is a queue of chars:

std::vector<char> func(int size) {
    std::vector<char> result;
    while(result.size() != size) {
        result.push_back(my_sync_queue.pop());
    }
    return result;
}