What is the two-phase termination pattern?

The pattern separates "please stop" from "has stopped". The requester sets a volatile flag and interrupts the worker, so both a busy worker and a blocked one notice. The worker checks the flag at safe points, catches InterruptedException as a stop signal, and then runs its own shutdown: flush buffers, release resources, notify a latch.

void shutdown() {
    running = false;      // volatile
    worker.interrupt();   // wake it if it is blocked
}

Why both: the flag alone leaves a blocked thread asleep, and the interrupt alone can be missed if it arrives between a check and a blocking call. The termination is complete only when the worker says so, which is what the join or the latch is for.