[omniORB] Destroying itself

[Wernke zur Borg]

Hi Attila,
 
if you call _remove_ref() right after the object activation or any time later but before deactivate_object(), the later call to deactivate_object() is sufficient for its deletion, as Clarke Brunt wrote.
 
If you call _remove_ref() only after deactivate_object() there is no problem if the object is still busy. The deletion will only happen after all references to the object have been removed, so there will be no clash or exception.
 
Regards, Wernke
 


________________________________

	From: Attila Czigány [mailto:atteeharp at gmail.com] 
	Sent: 10 May 2006 12:11
	To: Wernke zur Borg
	Cc: Luke Deller; omniorb-list
	Subject: Re: [omniORB] Destroying itself
	
	
	Hello,
	Thanks for the replies.
	
	

		I am working on a quite similar application, and I think the pattern is very common. In my case there are a number of methods on the processor object that the client can call. Therefore I have added a close() method so that the client can signal when the object is no longer needed. In the close() function I call poa->deactivate_object() but this does not deactivate the object straight away nor does it delete the servant!

	Not? deactivate_object() says to the POA, that the servant should be destroyed. It waits until the servant finishes its jobs, and then deactivates and deletes the servant. So are you sure about that in your app?
	Here is a dump of mine:
	app: Finish method started.
	omniORB: State root<1> (active) -> deactivating
	omniORB: Object is still busy -- etherealise later.
	app: Finish method finished
	omniORB: sendChunk: to giop:tcp:10.0.0.10:42682 24 bytes
	omniORB: POA(RootPOA) etherealising object root<1> (deactivating).
	 id: IDL:root/Generator:1.0
	omniORB: State root<1> (deactivating) -> etherealising
	omniORB: Removing root<1> (etherealising) from object table
	omniORB: Object table entry root<1> (dead) deleted.
	omniORB: RefCountServantBase has zero ref count -- deleted.
	app: destructor invoked, got dead
	
	So it is deleted right, deactivation decreases the ref_count, so the _remove_ref() should not be invoked, just after the activation.
	I just wanted to do this without an additional method. I just wanted to automate this process.
	(As it is in EJB3 for example) 


		To my knowledge there are two possibilities for servant deletion: 
		

		(a) Use a ServantActivator POA for the activation of servants - its method etherealize() will be called when the object is no longer in use. This pattern is well described in [Bolton, Pure Corba].


	Yes I thought something similar to this, is this complex very much? This would eliminate the additional close/remove method. If yes, maybe the efforts do not worth the time. 
	


		(b) Derive your servant from RefCountServantBase and call _remove_ref() after poa->deactivate_object(). I have made some tests and up to now the object gets properly destructed in all cases. Perhaps somebody can confirm that this is a safe way to proceed.

	This should not be used, because the _remove_ref() decreases immediatelly the ref_count, and the object is still busy, as we invoked this things from its method, and it hasnt returned. So if you use _remove_ref, the deletion will be faster than the deactivation and you will get exceptions, so the remove is not clear, and poa can deactivate nothing as the object is already deleted.
	Correct me if I am mistaken.
	
	
	I will check the ServantActivator POA, I hope it is not too complex:) Does anybody have some experiences with it?
	Thanks again, Attila
	

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.omniorb-support.com/pipermail/omniorb-list/attachments/20060510/f09ed124/attachment.htm