The nightly automation runs showed since the middle of december failure
of the Intel platforms tests. It appears that restarting the cimom
causes a hotspot crash.

###@###.### 2002-01-07

A sample repository is located at:
   /net/micheler/files/logr_bonita0103

When the CLI tests are run, they get to a certain point and then an error is
shown stating WBEM server not found. However if you grep for the process you see
it running. When we stop and restart the process using init.wbem, we get the
following exception:
bonita# /etc/init.d/init.wbem start
Starting Solaris Management Console server version 2.1.0.
endpoint created: :898
Unable to start CIMOM (CIM_ERR_FAILED); some CIM-based services may not be
available.
CIM_ERR_FAILED:
com.sun.wbem.repository.LogException: recovery failed; nested exception is:
        java.lang.ClassCastException
CIM_ERR_FAILED:
com.sun.wbem.repository.LogException: recovery failed; nested exception is:
        java.lang.ClassCastException
Solaris Management Console server is ready.

If you stop and restart the CIMOM, you receive a hotspot error.


---------------------------------------------------------------------------

On initial investigation, it appears that the server is confused about whether
or not it is already running. And this only happens on intel platforms. So
the thinking was that this was a Java problem. But when Javasoft attempted
to look at it by modifying /usr/sadm/lib/smc/bin/smcwbemserver as follows
to run a debug version of the Java runtime :

  GCARG="-XX:+VerifyBeforeGC -XX:+VerifyAfterGC"
  JAVA="/net/curious-george/disk4/JDK/1.4/solx86/bin/java_g"

  ${JAVA} "${JAVAARGS[@]}" -Xmx128m ${GCARG} -Djava.security.policy=${VIPER_HOME}/policy/smcwbemserver.policy -Dviper.argv0=$0 -Dviper.home=${VIPER_HOME} -Dviper.rthome=${SMC_RTHOME} -Dviper.display=${X_DISPLAY} -Dpropdir=${WBSERLIB} ${JAVARMIOPTS} com.sun.management.viperimpl.server.ViperWbemServer "$@" |&

they couldn't get past this error:

  Starting Solaris Management Console server version 2.1.0.
  Solaris Management Console server exiting: no clean temporary directory.
  This could be caused by an abnormal termination of previous server by a
  different user. Check the permission of directory /var/run/smc+port/.

byennaco@East 2002-01-07

----------------------------------------------------------------------------

The above problem with running with java_g may be due to the fact that SMC's
native libs are not debug-built versions. I will attempt to generate these.

In the meantime, we learned that the original problem easily reproduceable
using the default client VM only (java -client). The problem was NOT reproduceable
with any of the following options:
    java -server
    java -client -Xint
    java -client -Xcomp
    java -client -Xcomp -Xbatch

The machine plankton.east (root passwd "noway!") is an intel machine running S9-build53,
and can be used to troubleshoot the problem.

"/etc/init.d/init.wbem stop" is the command to stop the server.
"/etc/init.d/init.wbem start" is the command to start the server.

When the HotSpot VM dumps, the SMC/WBEM server is in an intermittent state. To
clear this state, simply stop the server before attempting to start it again.

byennaco@East 2002-01-07

----------------------------------------------------------------------------