[GR-60018] Revert recent ContextPauseTest#testPauseManyThreads change…

… in order to prevent hangs. PullRequest: graal/19412
oracle · Nov 25, 2024 · d79ab18 · d79ab18
2 parents ad42202 + 5652358
commit d79ab18
Showing 1 changed file with 9 additions and 9 deletions.
diff --git a/...nstrumentation.test/src/com/oracle/truffle/api/instrumentation/test/ContextPauseTest.java b/...nstrumentation.test/src/com/oracle/truffle/api/instrumentation/test/ContextPauseTest.java
@@ -373,7 +373,7 @@ public synchronized void waitUntilCount(long count) {
 
     @Test
     public void testPauseManyThreads() throws ExecutionException, InterruptedException, IOException {
-        int nThreads = Runtime.getRuntime().availableProcessors() + 2;
+        int nThreads = Runtime.getRuntime().availableProcessors() + 1;
         ExecutorService executorService = threadPool(nThreads, vthreads);
         ByteArrayOutputStream outputStream = new ByteArrayOutputStream();
         Engine.Builder engineBuilder = Engine.newBuilder().allowExperimentalOptions(true).option("engine.SynchronousThreadLocalActionMaxWait", "1").out(outputStream).err(outputStream);
@@ -462,7 +462,7 @@ public void onReturnExceptional(EventContext c, VirtualFrame frame, Throwable ex
             Assert.assertFalse(pauseFuture1.isCancelled());
             // The one thread executing context1 is now paused
             /*
-             * Execute the same code in context2 in availableProcessors + 1 (nThreads - 1) threads.
+             * Execute the same code in context2 in availableProcessors (nThreads - 1) threads.
              */
             for (int i = 1; i < nThreads; i++) {
                 executorService.submit(() -> {
@@ -484,13 +484,13 @@ public void onReturnExceptional(EventContext c, VirtualFrame frame, Throwable ex
              * the synchronous pause thread local action may not reach the synchronization point in
              * case virtual threads are used and is automatically cancelled. The reason for not
              * reaching the synchronization point is that when availableProcessors of these threads
-             * are pinned and then the waiting for the start sync of the pause thread local action
-             * using LockSupport.parkNanos, the (availableProcessors + 1)-th thread cannot run
-             * because the parallelism of the default virtual thread scheduler is only
-             * availableProcessors and none of the "parked" threads, though waiting, can be
-             * compensated for due to the pinning. Also, the timeout for the pause thread local
-             * action sync is pretty small (one second) so the cancellation might occur even for
-             * platform threads in case some unusual slowdown.
+             * are pinned and waiting for the start sync of the pause thread local action using
+             * LockSupport.parkNanos (or already paused in case of the already paused context1
+             * thread), the (availableProcessors + 1)-th thread cannot run because the parallelism
+             * of the default virtual thread scheduler is only availableProcessors and none of the
+             * "parked" threads, though waiting, can be compensated for due to the pinning. Also,
+             * the timeout for the pause thread local action sync is pretty small (one second) so
+             * the cancellation might occur even for platform threads in case some unusual slowdown.
              */
             boolean pause2Cancelled = pauseFuture2.isCancelled();
             // If context2 pausing is cancelled, new thread in context2 should not be paused even