Fix race condition in async task reschedule test (elastic#137405) (elastic#137449)

When rescheduling a task, the task may begin running before the test is
able to take more actions. This means reseting the barriers in the auto
reschedule test may occur after the second run of the task has begun,
which results in a broken barrier.

This commit fixes the race condition by moving the reset of each barrier
to occur after the await for the relevant barrier. This ensures that
each barrier is ready to be awaited on again by the end of the run. Due
to there being two barriers, the second barrier may be reset after the
test codes second barrier1.await, but that is ok.

closes elastic#131150

Author: rjernst

server/src/test/java/org/elasticsearch/common/util/concurrent/AbstractAsyncTaskTests.java

@@ -62,13 +62,17 @@ protected void runInternal() {
                     logger.error("barrier1 interrupted", e);
                     fail("interrupted");
                 }
+                barrier1.reset();
+
                 count.incrementAndGet();
                 try {
                     barrier2.await();
                 } catch (Exception e) {
                     logger.error("barrier2 interrupted", e);
                     fail("interrupted");
                 }
+                barrier2.reset();
+
                 if (shouldRunThrowException) {
                     throw new RuntimeException("foo");
                 }
@@ -83,8 +87,8 @@ protected void runInternal() {
         assertTrue(task.isScheduled());
         barrier2.await();
         assertEquals(1, count.get());
-        barrier1.reset();
-        barrier2.reset();
+
+        // wait for the rescheduled task to begin running
         barrier1.await();
         assertTrue(task.isScheduled());
         task.close();