[jboss-user] [JBoss Microcontainer Development] New message: "CallbackItem.changeCallback() very expensive due to context tracking"

Kabir Khan do-not-reply at jboss.com
Mon Feb 22 08:51:35 EST 2010


User development,

A new message was posted in the thread "CallbackItem.changeCallback() very expensive due to context tracking":

http://community.jboss.org/message/527596#527596

Author  : Kabir Khan
Profile : http://community.jboss.org/people/kabir.khan@jboss.com

Message:
--------------------------------------------------------------
I'm running a benchmark with 10 callbacks installed and 990 beans matching those callbacks, so I end up with 9,900 calls to SingleCallbackItem.changeCallback().
 
Relevant part of stacktrace along with number of calls and %cpu time taken
 
Thread [main] (Suspended) 
(19,800 calls, 46.9% cpu) AbstractKernelControllerContext(AbstractControllerContext).getContextTracker() line: 389  
(9,900 calls, 47.3% cpu) AbstractKernelControllerContext(AbstractControllerContext).getTarget(ControllerContext) line: 474 
(9,900 calls, 47.6% cpu) ClassSingleCallbackItem(OwnerCallbackItem<T,C>).getTarget(ControllerContext, boolean) line: 68 
ClassSingleCallbackItem(SingleCallbackItem<T>).changeCallback(ControllerContext, boolean) line: 62
 ClassSingleCallbackItem(AbstractCallbackItem<T>).changeCallback(Controller, ControllerContext, boolean) line: 80 
 ClassSingleCallbackItem(OwnerCallbackItem<T,C>).changeCallback(Controller, ControllerContext, boolean) line: 116 
 
What is really heavy inside ACC.getContextTracker() are:
A - 138,600 callls, 34% cpu
B - 19,800 calls 10.3% cpu
 
   public ContextTracker getContextTracker()
   {
      if (tracker == null || tracker == NOOP)
      {
         synchronized (this)
         {
            // since we got through, we must be the same caller
            if (tracker == NOOP)
               return null;
 
            // we waited, got through, but it's now changed
            if (tracker != null && tracker != NOOP)
               return tracker;
 
            tracker = NOOP; // mark that we're initializing
            ContextTracker ct = null;
            MetaData metaData = scopeInfo.getMetaData(); //B
            if (metaData != null)
            {
               ct = metaData.getMetaData(ContextTracker.class); //C
               if (ct == null)
               {
                  List<ScopeLevel> levels = CommonLevelsUtil.getSubLevels(DEFAULT_MINIMAL);
                  int instanceIndex = levels.indexOf(CommonLevels.INSTANCE);
                  for (int i = instanceIndex; i >= 0 && ct == null; i--)
                  {
                     MetaData md = metaData.getScopeMetaData(levels.get(i));  //A
                     if (md != null)
                        ct = md.getMetaData(ContextTracker.class);
                  }
               }
            }
            tracker = ct; // should we care if it's still null?
         }
      }
      return tracker;
   }

 
For A, maybe scopeInfo could keep a reference to the MetaData to avoid having to access the repository every time we call scopeInfo.getMetaData()?
 
For B I don't really understand exactly why this is being done? It looks like we check for ContextTracker in each metadata level at INSTANCE level and above, up to JVM. Wouldn't that be handled by the call to metaData.getMetaData() in C anyway?

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

To reply to this message visit the message page: http://community.jboss.org/message/527596#527596




More information about the jboss-user mailing list