C++ futures parallel processing

Question

I'm using std::futures to parallel process my algorithm. I split up the information into mutually exclusive pools and then perform the same operation on each pool in its own thread. The code looks like this:

class Processor
{
public:
    Processor(const std::string &strVal) : m_strVal(strVal)
    {
    }

    std::string GetVal() const {return m_strVal;}

    std::vector<std::string> Do()
    {
        // do some processing - this can throw an exception
    }

private:
    std::string m_strVal;
};

class ParallelAlgo
{
private:
    std::vector<std::string> m_vecMasterResults;

public:

    ProcessingFunction(const std::vector<std::string> &vecInfo)
    {
        // vecInfo holds mutually exclusive pools

        std::vector<std::future<std::vector<std::string> > > vecFutures(vecInfo.size());

        try
        {
            for (auto n = 0 ; n < vecInfo.size() ; n++)
            {
                vecFuture[n] = std::async(std::launch::async, &ParallelAlgo::WorkFunc, vecInfo[n].GetVal());
            }

            for (auto it = vecFutures.begin() ; it != vecFutures.end() ; ++it)
            {
                std::vector<std::string> RetVal = it->get();
                m_MasterResults.insert(m_MasterResults.begin(), RetVal.begin(), RetVal.end());
                vecFutures.erase(it);
            }
        }
        catch (exception &e)
        {
            for (auto it = vecFutures.begin() ; it != vecFuture.end() ; ++it)
            {
                // race condition?
                if (it->valid())
                {
                    it->wait_for(std::chrono::second(0));
                }
            }
        }
    }

    std::vector<std::string> ParallelAlgo::WorkFunc(const std::string &strVal)
    {
        Processor _Proccessor(strVal);
        return _Processor.Do();
    }
};

My question is how to handle the situation when an exception is thrown in Processor:Do()? Currently I catch the exception using a future, and then wait zero seconds for each future that hasn't finished; this is fine - these threads will simply terminate and the processing will not be completed. However, have I not introduced a race condition in the catch block. a future could finish between the calls to valid() and wait_for(), or is this not a concern as I'm not calling get() on these incomplete futures?

Answer 1

A call to valid only checks if there is a corresponding shared state, which will still be true for a finished thread until you call get on it.

There is no race condition there.