APAR:         PQ54436

Version:      3.5.2, 3.5.3, 3.5.4, 3.5.5 and 4.0.2

OS/Platforms: All

CMVC defects: PQ54436

Date: Jan. 17, 2002

Abstract:
  Two WAS nodes in a multi-node domain with LTPA security enabled; disconnect one node, 
  fire up an admin gui at the other node, and the app-server on that HEALTHY node 
  crashes, and does not come back up correctly. 


Description: 
The problems seen with the system management topology were caused by several timing issues. 
With the second admin node down, the Solaris hardware will wait for up to 4.5 minutes, 
depending on the configuration, before it returns from creating a socket. This time was 
multiplied by the fact that the ORB does retries to compensate for the file system handles 
not being available. The retries have been removed to lessen the amount of time that the wait
on the socket creation. The extended wait caused the transaction that is used by the system 
management topology to fail. Without the retries the ORB returns the failure to connect 
before the transaction times out.

A second problem occurred during this time out because of a synchronization block that was 
waiting on the socket creation. This sync block was reduced to include only the connection 
table update. This allows a second connection to be established during the wait period. 
This caused the topology to not show any other servers that were running with the admin 
node that was up.


How to install the fix:
  - create an efix directory to store the zip file (e.g. c:\WebSphere\AppServer\efix)
  - copy PQ54436-355.zip to the efix directory

How to apply the fix to the Admin Server:
  - add the zip file to the beginning of the Admin Server's classpath
    in admin.config.  The zip file has to be in front of ujc.jar in the classpath e.g.
    com.ibm.ejs.sm.adminserver.classpath=C\:/WebSphere/AppServer/efix/PQ54436-355.zip;...
  - Stop and restart the Admin Server  

How to apply the fix to the Admin Console for Unix platforms:                       
  - edit adminclient.sh in .../WebSphere/AppServer/bin                      
  - look for the comment "rem setup the classpath" in adminclient.bat
  - add the zip file to the beginning of the classpath by   
    adding "WAS_CP=$WAS_HOME/efix/PQ54436-355.zip" (without the quotes) directly 
    under the comment line mentioned in the previous step.                            
  - change the line "WAS_CP=$WAS_HOME/lib/ibmwebas.jar" to                  
    "WAS_CP=$WAS_CP:$WAS_HOME/lib/ibmwebas.jar"                             
  - save adminclient.sh and start the console

How to apply the fix to the Admin Console for NT platform:
  - edit adminclient.bat file in .../WebSphere/AppServer/bin subdirectory
  - look for the comment "rem setup the classpath" in adminclient.bat
  - add the zip file to the beginning of the classpath by   
    adding "set WAS_CP=%WAS_HOME%\efix\PQ54436-355.zip" (without the quotes) directly 
    under the comment line mentioned in the previous step.                            
  - change the line "set WAS_CP=%WAS_HOME%\lib\ibmwebas.jar" to                  
    "set WAS_CP=%WAS_CP%;%WAS_HOME%\lib\ibmwebas.jar"                             
  - save adminclient.bat and start the console