I was working in KnowledgeAgent incremental change-set processing. Right now, the agent only adds to the kbase those rules that were modified. The bad news are that in order to know if a rule was modified or not, the agent must compile the resource.
Here you can find more info: http://ilesteban.wordpress.com/2010/03/25/knowledge-agent-incremental-change-set-processing-and-binary-diff/ <http://ilesteban.wordpress.com/2010/03/25/knowledge-agent-incremental-change-set-processing-and-binary-diff/>I'm planning to post a more detailed explanation of what happens inside the kagent when processing change-sets. Best, 2010/5/19 Luiz Eduardo Guida Valmont <[email protected]> > Hi all, > > I'm working on an application that has to maintain a knowledge base that'll > change (a lot) over time; meaning some rules might change, others will be > added and, finally, some will be deleted from the KB. Our tests show the > following un-expected (imho *) behaviors: > > - Whenever the contents of the KB change, all rules associated with it > are re-compiled.It doesn't matter whether one rule or all of them were > changed. > - If the time taken to re-compile all rules is greater than the > frequency according to which resource change notification / scanner > services > poll the KB, then the KB will be re-compiled more than once. I can't > measure > how many extra compilations happen due to time constraints; it takes more > than 10 minutes for the application to compile all rules and something like > 5 minutes to recompile the smallest possible version of it. > > So... my question is: is possible to fine tune this behavior so that drools > won't recompile everything from scratch? More than that, I'd like to have it > wait for itself to finish any compilation that might happening at the time. > Is it possible? > > My approach to using drools is the following: create a KnowledgeAgent, then > have it apply a change set that points to a directory that contains DRL > files. And this has to be done for five different KA's. I suppose this is > the correct approach. > > Thanks in advance for any feedback that you might have. > > Regards, > Luiz Eduardo > > _______________________________________________ > rules-users mailing list > [email protected] > https://lists.jboss.org/mailman/listinfo/rules-users > > -- XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX Esteban Aliverti
_______________________________________________ rules-users mailing list [email protected] https://lists.jboss.org/mailman/listinfo/rules-users
