I was able to reproduce this on Oracle Java 1.6.0_45.

I could trace the issue down to ScheduledThreadPoolExecutor not executing the DeleteTask submitted in [1].

Note that the executor is created using new ScheduledThreadPoolExecutor(0, threadFactory) in org.jboss.as.server.deployment.module.TempFileProviderService. 
Having seen several reports that ScheduledThreadPoolExecutor with corePoolSize=0 does not execute the tasks I tried ScheduledThreadPoolExecutor(1, threadFactory) and the issue stopped happening. I'll have to investigate further if this is a good solution also for newer Java versions.

[1] https://github.com/jbossas/jboss-vfs/blob/54cabe6f9001fbfe45d8a84b96f96cf215366e30/src/main/java/org/jboss/vfs/TempFileProvider.java#L97

ScheduledThreadPoolExecutor with corePoolSize 0 does not schedule the tasks not only on early versions of Java 7 (as stated in https://bugs.openjdk.java.net/browse/JDK-7091003 ) but also on Java 1.6 which we see happening here. 

There is this case in JDK-7091003:

final ScheduledExecutorService ex = Executors.newScheduledThreadPool(0);
ex.schedule(new Runnable() {
    @Override
    public void run() {
        System.out.println("hello");
    }
}, 1, TimeUnit.SECONDS);

ex.shutdown();
ex.awaitTermination(5, TimeUnit.SECONDS);

JDK-7091003 states that the test passes on JDK 1.6.0_26-b03 correctly outputting "hello". While I can confirm the output, the following needs to be added: If we change the test as follows:

final ScheduledExecutorService ex = Executors.newScheduledThreadPool(0);
ex.schedule(new Runnable() {
    @Override
    public void run() {
        System.out.println("hello");
    }
}, 1, TimeUnit.SECONDS);

System.out.println("scheduled");
Thread.sleep(5000);
System.out.println("slept");

ex.shutdown();
ex.awaitTermination(5, TimeUnit.SECONDS);

then the output on Java 6 (e.g. 1.6.0_26 and 1.6.0_45) is 

scheduled
slept
hello

This is clearly not what we expect. We expect "hello" before "slept" as is indeed the case on newest versions of Java 7 and Java 8.

The modified test case shows, that on Java 6, the task execution is actually triggered by ScheduledExecutorService.shutdown(). 

Using ScheduledThreadPoolExecutor with corePoolSize 1 still seems to be a valid fix on Oracle/Sun Java 6.

Note that all JDKs referred to in Comment#2 are from Oracle/Sun.

I have run the modified test on several more JDKs. Here is the summary:

Need a fix:
* Oracle/Sun 1.6.0_26, 1.6.0_45: the order is scheduled, slept, hello
* OpenJDK 1.6.0_24: the "hello" output is not coming at all. 
* IBM JDK 6.0-16.30: the order is scheduled, slept, hello

OK:
* Oracle 7, 8: the order is scheduled, hello, slept 
* OpenJDK 1.8.0_102: the order is scheduled, hello, slept 
* IBM JDK 7.1-2.10: the order is scheduled, hello, slept
* IBM JDK 8.0-0.0: the order is scheduled, hello, slept

Untested:
* OpenJDK 1.7 unable to install

Setting UpstreamNotAffected because newer EAP versions are not supported on Java 6.

The issue is fixed in EAP 6.4.12.CP.CR1 for both Oracle and OpenJDK. However, I am still able to reproduce it when using IBM 1.6 JDK.

Since the IBM Java issue is tracked in a separate BZ, setting this to verified.

Retroactively bulk-closing issues from released EAP 6.4 cummulative patches.

Retroactively bulk-closing issues from released EAP 6.4 cumulative patches.