This is a bug in the HotSpot server mode and there is a
Sun defect opened, reference Sun Bug Id 4724509:
There can be other reasons that the core is so large, like a native
memory leak, possibly with the DB drivers.
The Sun JDK™ 1.3.1 doesn't always produce a Java™ thread dump when a
crash occurs, especially when the crash is inside the HotSpot library.
Below, the pmap command output shows that the native heap has grown to
over 2 GB. Such
a large heap size is unusual for a Java application and can lead to memory
problems, as in a crash in the C runtime library (libc) function memory:
core 'core1' of 16996:
/opt/WebSphere/AppServer/java/jre/bin/../bin/sparc/native_threads/java
00010000 24K read/exec
/opt/WebSphere/AppServer/java/jre/bin/sparc/native_threads/java
00024000 8K read/write/exec
/opt/WebSphere/AppServer/java/jre/bin/sparc/native_threads/java
00026000 2088888K read/write/exec [ heap ]
A known bug in Java HotSpot Server mode might cause this large heap. In
rare circumstances, the compiler gets into an expansion loop and asks for
larger and larger amounts of heap memory on malloc (memory allocation)
calls within the loop that is expanding the heap size. When this happens,
a java.lang.OutOfMemoryError with a large "CompileThread0" request is
logged in an hs_err_pid log and a pmap of the core file generated shows a
heap size that is unusually large, frequently 2000 MB or more. |