Just an update.
The good news is that I managed to reproduce this on an 8-way Intel box.
Sometimes it hangs due to the applications LinkedBlockingDeque and
sometimes it hangs in the executor's LinkedBlockingQueue.
The bad news is that as soon as I modified the LBQ/LBD and
ReentrantLock code to allow me to probe deeper, I can no longer
reproduce the problem :-( And all I did to the code was make everything
public so that I could dump the lock state when the hang occurred.
Very frustrating ...
David
David Holmes said the following on 07/18/09 12:28:
Sorry Ryan, misread the stack trace. The main thread is interacting with the
executors LBQ not the application one.
I'll see if I can use this test to reproduce the issue on Monday.
David Holmes
-----Original Message-----
From: [email protected]
[mailto:[email protected]]on Behalf Of David
Holmes
Sent: Saturday, 18 July 2009 12:15 PM
To: Ryan Betts; [email protected]
Cc: core-libs-dev
Subject: Re: [concurrency-interest] LinkedBlockingDeque deadlock? (Test
case)
Ryan,
Thanks very much for this.
When this "deadlocks" does top show whether the java process is actually
quiet or is it still consuming CPU? I notice that in the stack
dump the main
thread is in the process of unlocking the lock that the other threads are
waiting upon and hence the system is not deadlocked - unless that main
thread is actually "stuck" somewhere. Can you take a series of
stackdumps to
see if/how the main thread is executing?
Feel free to just send them to me rather than the mailing lists.
Thanks,
David Holmes
-----Original Message-----
From: [email protected]
[mailto:[email protected]]on Behalf Of Ryan
Betts
Sent: Saturday, 18 July 2009 2:48 AM
To: [email protected]
Cc: core-libs-dev
Subject: Re: [concurrency-interest] LinkedBlockingDeque deadlock? (Test
case)
Hello all,
I've been working with Ariel on the j.u.c.LinkedBlockingDeque deadlock
we observe. The attached test case reproduces the deadlock
intermittently, usually requiring several minutes. The test simply
prints periods until it deadlocks. Attempts to produce a simpler test
case using only ReentrantLock or without an ExecutorService were
unsuccessful.
I've attached jstack reports showing the deadlocked case and the dmesg
output from one of the servers that reproduces the defect. To date,
this deadlock only occurs on our 2 socket i7s.
We have not filed a defect with Sun for this issue. If that would be
helpful (or necessary), we would of course happily oblige.
$ java -version
java version "1.6.0"
OpenJDK Runtime Environment (build 1.6.0-b09)
OpenJDK 64-Bit Server VM (build 1.6.0-b09, mixed mode)
$ uname -a
Linux host3e 2.6.18-128.1.16.el5 #1 SMP Tue Jun 30 06:07:26 EDT 2009
x86_64 x86_64 x86_64 GNU/Linux
$ javac LBDLockPatternTest.java
$ java LBDLockPatternTest
Thank you,
*--Ryan.
_______________________________________________
Concurrency-interest mailing list
[email protected]
http://cs.oswego.edu/mailman/listinfo/concurrency-interest
