Re: [rules-users] fireUntilHalt and timing of rule activations

Thanks for the suggestions.  They all look like good ways to handle the situation I described.  However, they require modifying all of the rules to check for the latch object and its state, which I would prefer not to do and doesn't seem like would be necessary. It seems to me that this is something that Drools can handle internally without the rules having to be written this way.  Since the rules engine processes rules in a single thread, it's a concurrency issue.  fireUntilHalt should be blocked when a fact is inserted/updated/retracted, until all activations as a result of that change in working memory are completed. Thoughts? Norman ________________________________ From: Wolfgang Laun <wolfgang.laun(a)gmail.com&gt; To: Rules Users List <rules-users(a)lists.jboss.org&gt; Sent: Sun, October 3, 2010 10:51:08 PM Subject: Re: [rules-users] fireUntilHalt and timing of rule activations 2010/10/4 Greg Barton <greg_barton(a)yahoo.com&gt; > > If you don't have some way of associating the data with a particular Latch > it's easy to get overlap when processing datasets.  In general you need some > way to group the data together.  You can avoid a back reference to the Latch > by having a Set of some sort in the Latch to which you add all data in the > batch. Which burdens all inserts and retracts to be accompanied by correpsonding updates of the Set/Map. > >  If you use a Set backed by an IdentityHashMap the overhead is pretty > small, and rules look like this: > > rule "CountAs" >        dialect "java" >        salience -1 >        when >                l : Latch() >                a : A( this memberOf l.dataSet ) >        then >                retract(a); >                l.incACount(); >                System.out.println("Found an A in " + l); > end > > See attached project. > > THough I like the cleverness of using the "data in matched objects alters > rule properties" trick, you could have just as easily had a "Latch(value == > true)" conditional, and that would be more clear, It was meant to emphasize the role of Latch.value as an enabler for the rule in contrast to a regular constraint being part of the application logic. YMMV ;-) > > IMO.  I'm curious to see of the enabled trick would perform better, > though. Whichever way, there is a considerable burden associated with writing the rules and, possibly, inserts and retracts. I wonder what the benefits of having the session run all the time are as opposed to simply let it fire until it stops; then do the inserts and then fireUntilHalt again? If there is, I'd opt for the addition of an atomic insertAll(Object... objects) and then none of this hocus-pocus would be necessary. -W > > GreG > > --- On Sun, 10/3/10, Wolfgang Laun <wolfgang.laun(a)gmail.com&gt; wrote: > > From: Wolfgang Laun <wolfgang.laun(a)gmail.com&gt; > Subject: Re: [rules-users] fireUntilHalt and timing of rule activations > To: "Rules Users List" <rules-users(a)lists.jboss.org&gt; > Date: Sunday, October 3, 2010, 4:23 AM > > There is another way of associating a Latch object with rules, without > having to store a reference to a Latch in objects: > > rule "CountAs" > enabled ( $v ) > when >      Latch( $v : value ) >      X( ... ) > > then ... end > > At the beginning, insert Latch( false ), which blocks all rules with this > construction, or modify the Latch object to false before inserting more > facts. Then, insert the facts, and, at the end, modify Latch to true. > > >     Latch latch = new Latch( true ); >     FactHandle fh = kSession.insert( latch ); >     kSession.fireAllRules(); >     latch.setValue( false ); >     kSession.update( fh, latch ); > > Of course, you can use multiple Latch objects, adding a name field with a > specific value, so that a latch applies to a subset of rules only: > > > rule "CountAs" > > enabled ( $v ) > > when > >      Latch( name == "CountAs", $v : value ) >      ... > > But be aware that changes to Latch objects will retrigger rules that have > fired previously; so with this approach you'll have to make sure to retract > facts when they have been processed. > > > -W > > > 2010/10/3 Greg Barton <greg_barton(a)yahoo.com&gt; > > Nope, you're not missing anything.  What you need is a control object of > some sort thst's inserted after all of the "real" data is inserted. (See > attached project for an example.) Rules will look like this, if the control > object is called BatchLatch and data objects A: > > > > > rule "CountAs" > >         dialect "java" > >         salience -1 > >         when > >                 l : Latch() > >                 a : A( latch == l ) > >         then > >                 retract(a); > >                 l.incACount(); > >                 System.out.println("Found an A in " + bl); > > end > > > > Note that the A object being processed is tied back to the latch.  This is > so multiple latches can be processed simultaneously and their processing > won't be intermingled.  This is necessary because there's no guarantee that > two Latch objects aren't in working memory at once. (Though you could create > a rule that enforces this.) > > > > > GreG > > > > --- On Sat, 10/2/10, Norman C <rent_my_time(a)yahoo.com&gt; wrote: > > > > > From: Norman C <rent_my_time(a)yahoo.com&gt; > > > Subject: [rules-users] fireUntilHalt and timing of rule activations > > > To: rules-users(a)lists.jboss.org > > > Date: Saturday, October 2, 2010, 10:22 AM > > > Hi All, > > > > > > In my app, I have a separate thread calling fireUntilHalt() > > > continuously.  I > > > have quite a few rules, and I am using salience extensively > > > to control the order > > > > > > in which rules are executed.  What I have seen (by adding > > > an event listener) is > > > that as a new fact is inserted, various rules are > > > activated.  Often, the > > > fireUntilHalt will start executing fireNextItem in > > > DefaultAgenda before all of > > > the activations are complete.  So if the rule with the > > > highest salience > > > value hasn't been activated at this point, then the first > > > rule to be fired isn't > > > > > > the correct one. > > > > > > This can be worked around by waiting for insert to return > > > and then calling > > > fireAllRules().  But it seems like the session should > > > block fireUntilHalt from > > > trying to execute activated rules until all activations are > > > complete.  Or am I > > > missing something here? > > > > > > thanks, > > > Norman > > > > > > > > > > > > > > > > > > _______________________________________________ > > > rules-users mailing list > > > rules-users(a)lists.jboss.org > > > https://lists.jboss.org/mailman/listinfo/rules-users > > > > > > > > > > _______________________________________________ > > rules-users mailing list > > rules-users(a)lists.jboss.org > > https://lists.jboss.org/mailman/listinfo/rules-users > > > > > > -----Inline Attachment Follows----- > > _______________________________________________ > rules-users mailing list > rules-users(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/rules-users > > > > > _______________________________________________ > rules-users mailing list > rules-users(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/rules-users > _______________________________________________ rules-users mailing list rules-users(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/rules-users