[ Team LiB ] Previous Section Next Section

10.2 Race Conditions

A race condition occurs when two threads attempt to use the same resource at the same time. The following class demonstrates a simple race condition. Two threads simultaneously try to increment a counter. If each thread can complete the increment( ) method in its entirety without the other thread executing, then all is fine, and the counter monotonically increases. Otherwise, the thread context switcher has the opportunity to interrupt one thread in the middle of executing the increment( ) method and let the other thread run through this method. Note that the thread can actually be interrupted anywhere, not necessarily in the middle of the increment( ) method, but I've greatly increased the likelihood of an interruption in the increment( ) method by including a print statement there:

package tuning.threads;
   
public class ThreadRace
  implements Runnable
{
  //global counter
  static int num=0;
   
  public static void increment(  )
  {
    int n = num;
    //This next line gives the context switcher an ideal
    //place to switch context.
    System.out.print(num+" ");
    //And when it switches back, n will still be the old
    //value from the old thread.
    num = n + 1;
  }
   
  public static void main(String args[  ])
  {
    ThreadRace d1 = new ThreadRace(  );
    ThreadRace d2 = new ThreadRace(  );
   
    Thread d1Thread = new Thread(d1);
    Thread d2Thread = new Thread(d2);
   
    d1Thread.start(  );
    d2Thread.start(  );
  }
   
  public void run(  )
  {
    for (int i = 200; i >= 0 ; i--)
    {
      increment(  );
    }
  }
}

The output from executing this class on a single-processor test machine is:

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 
31 32 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 33 34 35 36 37 38 39 40 41 
42 43 44 45 46 47 48 49 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49

You see that after 16, the next number is 16 again, and after the first 32, the next number is 17, as the threads switch back and forth in the middle of the increment( ) method. On a multiprocessor machine, the situation is even more confused.

Synchronizing the increment( ) method ensures the correct behavior of a monotonically increasing counter, as this gives exactly the desired behavior: the method is forced to complete before another call to it from any thread can be started.

In this test, because the counter is static, the increment( ) method needs to be static for synchronization to work correctly. If the increment( ) method is not static, synchronizing it locks the monitor for each this object rather than for the class. In the example I used a different object in each thread. A non-static increment( ) method is synchronized separately on each this object, so the updates remain unsynchronized across the two threads.

It is not simply that the num variable is static (though it needs to be for this particular example to work). The critical point is that the monitor that locks the method must be the same monitor for the two threads; otherwise, each thread gains its own separate lock with no synchronization occurring. Generally, deciding what to synchronize can be quite subtle, and you need to keep in mind which monitor is going to be locked by any particular thread.

    Previous Section Next Section