[jira] [Commented] (LUCENE-4987) Test framework may fail internally under J9 (some serious JVM exclusive-section issue).

["Dawid Weiss (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/LUCENE-4987?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13651882#comment-13651882 ] 

Dawid Weiss commented on LUCENE-4987:
-------------------------------------

I think I found the reason. J9 seems to be optimizing away the following code:
{code}
  @Override public synchronized void handleEvent(Object event)
      throws InvocationTargetException {
    super.handleEvent(event);
  }
{code}
this method is removed entirely as the stack trace from the 'wtf' exception shows:
{code}
java.lang.reflect.InvocationTargetException
	at sun.reflect.GeneratedMethodAccessor3.invoke(Unknown Source)
	at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:55)
	at java.lang.reflect.Method.invoke(Method.java:613)
	at com.google.common.eventbus.EventHandler.handleEvent(EventHandler.java:74)
	at com.google.common.eventbus.EventBus.dispatch(EventBus.java:314)
	at com.google.common.eventbus.EventBus.dispatchQueuedEvents(EventBus.java:296)
	at com.google.common.eventbus.EventBus.post(EventBus.java:267)
	at com.carrotsearch.ant.tasks.junit4.it.TestJ9SanityCheck$2.call(TestJ9SanityCheck.java:75)
	at com.carrotsearch.ant.tasks.junit4.it.TestJ9SanityCheck$2.call(TestJ9SanityCheck.java:1)
	at java.util.concurrent.FutureTask$Sync.innerRun(FutureTask.java:345)
	at java.util.concurrent.FutureTask.run(FutureTask.java:177)
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1121)
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:614)
	at java.lang.Thread.run(Thread.java:777)
Caused by: java.lang.RuntimeException: Wtf? two threads in a handler: Thread[pool-100-thread-1,5,TGRP-TestJ9SanityCheck] and Thread[pool-100-thread-2,5,TGRP-TestJ9SanityCheck]
	at com.carrotsearch.ant.tasks.junit4.it.TestJ9SanityCheck$1.onSlaveIdle(TestJ9SanityCheck.java:52)
	... 14 more
{code}

The eventhandler type is definitely SynchronizedEventHandler and on Hotspot you do get a synchronized indirection layer (a call under a synchronized monitor), whereas on J9 this becomes a full race condition.


                
> Test framework may fail internally under J9 (some serious JVM exclusive-section issue).
> ---------------------------------------------------------------------------------------
>
>                 Key: LUCENE-4987
>                 URL: https://issues.apache.org/jira/browse/LUCENE-4987
>             Project: Lucene - Core
>          Issue Type: Bug
>            Reporter: Dawid Weiss
>            Assignee: Dawid Weiss
>            Priority: Minor
>         Attachments: j9.zip
>
>
> This was reported by Shai. The runner failed with an exception:
> {code}
> [junit4:junit4] Caused by: java.util.NoSuchElementException
> [junit4:junit4]         at java.util.ArrayDeque.removeFirst(ArrayDeque.java:289)
> [junit4:junit4]         at java.util.ArrayDeque.pop(ArrayDeque.java:518)
> [junit4:junit4]         at com.carrotsearch.ant.tasks.junit4.JUnit4$1.onSlaveIdle(JUnit4.java:809)
> [junit4:junit4]         ... 17 more
> {code}
> The problem is that this is impossible because the code around JUnit4.java:809 looks like this:
> {code}
>      final Deque<String> stealingQueue = new ArrayDeque<String>(...);
>      aggregatedBus.register(new Object() {
>         @Subscribe
>         public void onSlaveIdle(SlaveIdle slave) {
>           if (stealingQueue.isEmpty()) {
>             ...
>           } else {
>             String suiteName = stealingQueue.pop();
>             ...
>           }
>         }
>       });
> {code}
> and the contract on Guava's EventBus states that:
> {code}
>  * <p>The EventBus guarantees that it will not call a handler method from
>  * multiple threads simultaneously, unless the method explicitly allows it by
>  * bearing the {@link AllowConcurrentEvents} annotation.  If this annotation is
>  * not present, handler methods need not worry about being reentrant, unless
>  * also called from outside the EventBus
> {code}
> I wrote a simple snippet of code that does it in a loop and indeed, two threads can appear in the critical section at once. This is not reproducible on Hotspot and only appears to be the problem on J9/1.7/Windows (J9 1.6 works fine).
> I'll provide a workaround in the runner (an explicit monitor seems to be working) but this is some serious J9 issue.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org