[jbosscache-dev] Re: Session Clustering for SIP

[Brian Stansberry]

Jean Deruelle wrote:
> It seems that in the FieldBasedClusteredSession class, the 
> setJBossInternalAttribute method (which is called by setAttribute) put 
> the object in the cache and then add Obervers to check for potentials 
> changes in the attributes of the object and if so set the session as 
> dirty so that on transaction commit it is replicated,
> 
> So I was thinking to do the same at cache level, when the put is done 
> add the observers and jbosscache takes care of the changes in the object 
> and replicate if needed on transaction commit.  But it seems that the 
> Observers can only be added on an AOPized object, so I still wonder why 
> this is needed to add those observers in the FieldBasedClusteredSession 
> if the object is already AOPized and not let JbossCache handle the 
> replication process automatically...?
> 
> 

The only thing AOPized is the attribute value.  The session itself 
isn't. The Observer impl generates a callback to the session so it knows 
it's dirty and can replicate session metadata (e.g. access timestamps, 
an access counter, etc).

Note that in AS 5 I'm aiming to get rid of this Observer usage. If 
changing the location where pojos is stored can work (see 
http://www.jboss.com/index.html?module=bb&op=viewtopic&t=134386) then a 
simple JBC CacheListener can serve the same function.  Even if not, I 
may be able to get rid of it and just rely on some new stuff that 
ensures the timestamp is replicated at least once a minute.

-- 
Brian Stansberry
Lead, AS Clustering
JBoss, a division of Red Hat
brian.stansberry at redhat.com