Subj : Application hangs in pthread_join
To   : comp.programming.threads
From : Jeff
Date : Fri Mar 18 2005 07:53 am

I have a multi-threaded application that occasionally hangs as it waits
in pthread_join. This is a HTTP load tester (siege) and it tends to
happen on GNU/Linux when the thread count exceeds 500. I suspect a
thread gets hosed in a socket read and the main thread can't join it.

Here is my thread creation and collection:

  /**
   * loop until my.cusers and create a corresponding thread...
   */
  for( x = 0; x < my.cusers && our.shutting_down != TRUE; x++ ){
    client[x].id    = x;
    client[x].bytes = 0;
    client[x].time  = 0.0;
    client[x].hits  = 0;
    client[x].code  = 0;
    client[x].fail  = 0;
    client[x].U     = urls;
    client[x].rand_r_SEED = pthread_rand_np(&randrseed);
    result = pthread_create(&(peer[x]), &scope_attr,
(void*)start_routine, &(client[x]));
    if(result == EAGAIN){
      my.verbose = FALSE;
      joe_fatal("system resources exhausted");
    }
    /**
     * lock and increment the thread counter
     */
    pthread_mutex_lock( &our.mutex_lock );
    if( our.shutting_down != TRUE ){
      our.total_clients ++;
      our.total_threads ++;
    }
    pthread_mutex_unlock(&our.mutex_lock);
    pthread_cond_signal(&our.exit_cond);
  } /* end of for pthread_create */

  /**
   * our.total_clients is the threads high water
   * mark; join all the threads that were spawned.
   */
  x = our.total_clients;
  while(x){
    pthread_join(peer[x--], &statusp);  <= hangs here!
  }


Any thoughs would be appreciated.

Jeff

.