When the barrier is constructed, the developer must specify the number of parties (threads) using the barrier. This number is used to trigger the barrier: the threads are all released when the number of threads waiting on the barrier is equal to the number of parties specified. There is also an option to specify an action--an object that implements the run() method. When the trigger occurs, the run() method on the barrierAction object is called prior to releasing the threads. This allows code that is not threadsafe to execute; generally, it calls the cleanup code for the previous phase and/or setup code for the next phase. The last thread that reaches the barrier--the triggering thread--is the thread that executes the action.
Each thread that calls the await() method gets back a unique return value. This value is related to the arrival order of the thread at the barrier. This value is needed for cases when the individual threads need to negotiate how to divide up work during the next phase of the process. The first thread to arrive is one less than the number of parties; the last thread to arrive will have a value of zero.
