The problem is due to how the ROOT.war is deployed. Instead of being deployed as a unique war with its own UCL, it is being associated as a nested deployment of the jbossweb-tomcat50.sar, and uses the sar UCL as the war URL:
2004-01-19 14:55:04,085 DEBUG [org.jboss.mx.loading.UnifiedClassLoader] Added url: file:/C:/cvs/JBoss3.2/jboss-3.2/build/output/jboss-3.2.4RC1/server/tomcat5/de ploy/jbossweb-tomcat50.sar/ROOT.war/, to ucl: org.jboss.mx.loading.UnifiedClassL [EMAIL PROTECTED] url=file:/C:/cvs/JBoss3.2/jboss-3.2/build/output/jboss-3.2.4RC1/s erver/tomcat5/deploy/jbossweb-tomcat50.sar/ ,addedOrder=5} When the shutdown occurs, all wars are undeployed, and then the jbossweb-tomcat50.sar. The stopService method of the jbossweb-tomcat50.sar is called after the ROOT.war has been undeployed and the jbossweb-tomcat50.sar UCL removed from its repository. A sar which is a deployer with a nested deployment is not going to shutdown cleanly. I'm still looking at why the NPE was not seen before, but this shutdown behavior is not new. xxxxxxxxxxxxxxxxxxxxxxxx Scott Stark Chief Technology Officer JBoss Group, LLC xxxxxxxxxxxxxxxxxxxxxxxx -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Tom Elrod Sent: Thursday, January 08, 2004 12:06 AM To: [EMAIL PROTECTED] Subject: Re: [JBoss-dev] Web integration (clustering) updates I am able to reproduce the problem, but after spending a good bit of time trying to find the bug, I am at a loss. Here is what I have found so far (Scott maybe you would be able to figure it out based on this info?). The LoaderRepository variable, repository, within UnifiedClassLoader is null when loadClassImpl() is called. Best I can tell, this probably happens because UnifiedClassLoader::setRepository() is called with a null value or UnifiedClassLoader::unregister() is called (am guessing it is the later). The MainDeployer does call the unregister() method on the UnifiedClassLoader, but not until after it calls stop(), which is where the NPE occurs (so has not gotten to the unregister() call yet). My best guess is that the problem stems from destroying one of the sub deployments (which happens before the NPE and will call unregister(), but not familiar enough with classloader/deployment stuff to know if would be using same UnifiedClassLoader instance). I did notice an exception in the log before the NPE (am attaching snippet of the log). I also checked the MainDeployer, UnifiedClassLoader, and UnifiedClassLoader3 for recent code changes. The MainDeployer had only one change related to using a HashSet for the waitingForClasses collection. The UnifiedClassLoader3 did not have any changes. The UnifiedClassLoader had a lot of changes, but don't seem related to this (but could be missing something). Remy Maucherat wrote: > Tom Elrod wrote: > >> So is this still an issue? If so, how do I reproduce (will look at >> it tonight)? > > > The value of looking into this is that it could possibly be a > regression. Otherwise, the workaround is good enough :) > > - Use the TC 5 SAR instead of the TC 4.1 SAR > - In the server.xml of the SAR, replace <Host name="localhost"> by > <Host name="localhost" autoDeploy="false" deployOnStartup="false"> > (that way, Tomcat will do a Class.forName during its shutdown, but > before AbstractWebContainer.stopService is called) > > R�my > ------------------------------------------------------- The SF.Net email is sponsored by EclipseCon 2004 Premiere Conference on Open Tools Development and Integration See the breadth of Eclipse activity. February 3-5 in Anaheim, CA. http://www.eclipsecon.org/osdn _______________________________________________ JBoss-Development mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-development
