Timer thread termination may not be detected and cause non-clean VM shutdown
David.Holmes at oracle.com
Tue Sep 13 10:18:35 UTC 2011
On 13/09/2011 5:51 PM, Jing LV wrote:
> I've meet a subtle issue with Timer. As we know, every Timer objects
> spawn a new thread (which is optionally a daemon thread). On VM
> shutdown, there is no way to ensure that the daemon thread terminates
That's exactly the point of daemon threads - they get blown away when the VM
terminates. If this is bad then don't use daemon threads.
> (cancel() and purge() only clear the queue). This can lead to non-clean
> VM shutdown (return non-zero exit code) and has caused problems in some
> applications especailly with attach API. Further, there is no way to
> determine when all tasks have completed after calling cancel().
Can you please give details of the problems that are encountered and under
> This problem was encouraged to occur when the test machine was busy. The
> original failure occurs when created& destroyed 15 processes with 1
> JVM's each (via JNI). It seems all versions of JDK has this issue.
> I was trying to make a simple testcase but no much progress - though
> logically I guess everyone can understand this easily. Current my
> workaround is to have some thread to join the Timer thread to wait its
> termination - but it seems have some side effect and cause problems.
Again please provide details. I'm unclear how join()ing a thread would have
problem causing side-effects. Though as a general rule you should never
join() a Thread you didn't create/define (otherwise how can you know when it
will terminate - the exception being when there is an API that causes the
thread to terminate.
> A quick idea to resolve the problem completely is to add a new API like
> "public void cancel(boolean wait)" to indicate if we will wait for the
> Timer threads to be terminated, or add some private methods to help the
> VM to do something to terminate the Timer.
Without knowing what problems are being encountered it is difficult to judge
whether such an API would be of assistance.
> Any suggestions/comments? Thanks.
More information about the core-libs-dev