java.exe crashes with EXCEPTION_ACCESS_VIOLATION on W2K

[Jens Becker]

Hello,

we currently have a very strange problem.

Our Java-RMI-server-application crashes with an
EXCEPTION_ACCESS_VIOLATION.
The access violation occurs in different modules (extract from
Dr.Watson-Log):

1. crash: ntdll - ZwDeviceIoControlFile
2. crash: javai - FindClassFromClass (FPO: [4,0,3])
3. crash: ntdll - RtlAllocateHeap
4. crash: ntdll - RtlSizeHeap
5. crash: javai - MakeClassSticky (FPO: [EBP 0x026ADA88] [2,0,4])
6. crash: javai - ObjAlloc (FPO: [EBP 0x00000004] [3,3,4])
7. crash: ntdll - ZwDeviceIoControlFile
8. crash: ntdll - RtlAllocateHeap
9. crash: ntdll - ZwWaitForSingleObject
10.crash: ntdll - RtlAllocateHeap
11.crash: ntdll - RtlAllocateHeap
12.crash: javai - MakeClassSticky (FPO: [1,0,3])

The effect cannot be reproduced.
There are no recognizable dependencies to special actions, time or
load.
The time between two crashes differ between 30 minutes and 3 days.

The application doesn't run the whole day, it starts in the morning
and stops
in the evening. Through the night it runs in a batch modus for e.g.
calculating purposes.

The server-application runs as a windows service (with srvany)
and uses a JNI-DLL (connection to SAP), but there's no JNI-call
at crash-time. A Oracle 9i database, Tivoli and UC4 is also
running on the same machine.
The problem exits on different machines (all clustered) with different
architectures, so nothing points to a hardware failure. We also have
some machines without any problems.

We don't have the problems on any machine while running on a NT 4.0
server,
only the clustered Windows 2000 servers are concerned.

We tried the following JRE's:
1.1.8 (SUN), 1.3.1 (SUN, 2 different builds) and the 1.3.1 coming
from the Oracle installation.

Thanks for any help or hint.

Regards,
Jens Becker

 

[Gordon Beaton]

Our Java-RMI-server-application crashes with an
EXCEPTION_ACCESS_VIOLATION.

The server-application runs as a windows service (with srvany)
and uses a JNI-DLL (connection to SAP), but there's no JNI-call
at crash-time.

You imply however that there is native code in the application...

Typical of errors involving pointers is that failures occur
erratically (or not at all), often in places without any apparent
connection to the actual error. If you have native code in your
application, it is not unlikely that you have corrupted the process'
memory in any number of ways, ultimately leading to this failure.

/gordon