boost::thread resource temporarily not available

Question

I have a very similar problem to this. Unfortunately, I am not allowed to comment on it so please excuse me for opening up another topic for this. My code is running a two-stage calculation iteratively which in principle looks like this:

while(!finishing_condition_met)
{
boost::thread_group executionGrp1;
        for(int w = 0; w < numThreads; w++)
        {
            boost::thread * curThread = new boost::thread(&Class::operation1, this, boost::ref(argument1), ...);
            executionGrp1.add_thread(curThread);
        }
        executionGrp1.join_all();

boost::thread_group executionGrp2;
        for(int w = 0; w < numThreads; w++)
        {
            boost::thread * curThread = new boost::thread(&Class::operation2, this, boost::ref(argument1), ...);
            executionGrp2.add_thread(curThread);
        }
        executionGrp2.join_all();
update_finished_criterion();
}

Since numThreads is significantly smaller than what the kernel would allow (it is set to hardware concurrency which is 56 on the current machine), I was surprised see this error. Does join_all() not take care of the finished threads? The thread_pool-approach suggested in the other post seems interesting but I am not exactly sure how to adapt it such that I can rerun everything within the loop multiple times while still waiting for the first stage to finish before starting the second stage.

Any suggestions are welcome! Thanks in advance.

EDIT: This is how I can cause this error in a minimalistic fashion. AFAIK, this is the standard way to implement parallel sections. Am I missing something?

#include "boost/thread.hpp"
#include "boost/chrono.hpp"
#include <iostream>
#include <algorithm>
#include <ctime>

using namespace std;

int numThreads = boost::thread::hardware_concurrency();

void wait(int seconds) {
    boost::this_thread::sleep_for(boost::chrono::milliseconds(seconds));
    return;
}

int subthread(int i) {
    wait(i/numThreads);
    return 1;
}

void threads(int nT) {

    boost::thread_group exeGrp;
    for (int i=0;i<nT;i++) {
        boost::thread * curThread = new boost::thread(&subthread, i);
        exeGrp.add_thread(curThread);
    }
    exeGrp.join_all();

}


int main() {

    for (int a=0;a<numThreads;a++) {
        cout << "Starting " << numThreads << " threads [" << a << "/" << numThreads << "]" << endl;
        threads(numThreads);
    }
    cout << "done" << endl;
}

Output when running code