Frequent JVM freeze/crash

Question

From time to time our system freezes (can be after days or hours).
We have been running on 1.5.0_04 and 1.5.0_05 both client and server (on x86 machine).
We have a lot of file IO and other JNI (custom) calls.
We've been trying many thing to get it going again...
One thing (killing of the external JNI related tasks) sometimes causes the JVM to crash (not every time - sometimes it actually gets going again - and sometimes it just keeps on freezing).
All crash logs say the same:
EXCEPTION_ACCESS_VIOLATION (0xc0000005)
Which isn't the most descriptive message :-(
The stack trace (when running in client mode) is as follows:
C 0x080c45f6 (differs from occurence to occourence)
V [jvm.dll+0x11d08a]
V [jvm.dll+0x11d22a]
V [jvm.dll+0x11cfbf]
C [MSVCRT.dll+0x2a3b0]
C [kernel32.dll+0xb50b]
Every thread is threadblocked, except "main" which is threadin_native.
The active thread is the VMThread.
Does anybody have a clue to what the problem could be, or how to debug the jvm.dll (at least just to find out what the calls referenced in the stacktrace are)?

Answer 1

Did you try JCopnsole to inspect the running JVM?
You need to start the server with
-Dcom.sun.management.jmxremote
-Dcom.sun.management.jmxremote.port=998
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=falseto be able to connect to a remote JVM.
With JConsole you can inspect all running threads if they are blocked and who they are waiting for. You might find the cause of your freezes there ...
If this doesn't help, use your application log (hope, your application uses some kind of logging like Java logging or Log4J ...) to inject more log messages at the suspicious classes/methods. This way you will find more infos in your log when the app freezes again.