Marina,
Operations over an existing rulebase (add/remove rules/packages) will lock existing stateful sessions in order to be executed making them thread safe. But having said that, for a JEE environment I would design it as in scenario #1.
I.e., requests are arriving and being handled by sessions created against a rulebase R1. When an update is made, the application will simply create a new rulebase (R2) and your singleton will now reference R2. Working memories have an internal reference to the rulebase used to create them, so existing sessions will work with the old rulebase until disposed, but any new request will create sessions against R2 (since your singleton now holds a reference for R2).
I like this solution because it provides, IMO, the safest and more consistent way of handling existing sessions, since you ensure a complete reasoning cycle (session) is made with an immutable set of rules, that can be easily tracked by whatever audit mechanisms you use. Only new requests will use the new rulebase, and are ensured to also complete the reasoning with immutable rules.
If you have long running sessions, though, the problem is a bit different and there is no other option IMO, except using solution #2.
Hope it helps,
Edson
2008/8/20 Marina
<ppine7@yahoo.com>
Hello,
I'm integrating DRools into a J2EE application and need to understand how it behaves in a multi-threaded distributed environment. I have many questions about thread safety but for now I'll ask about clustering only and post other questions separately.
So, as I understand, Drools per se is not cluster-aware. In my current design I have a singleton class, RuleEngineMAnager, that holds a reference to a RuleBase. Hence, there will be one instance of a rulebase per JVM per machine in a cluster. Rules can be added/modified from any node in the cluster, and the modifications have to me made visible in all rulebases.
The way I handle it now is by using a DB as the shared unit and setting up a flag when modifications are made in any JVM. Next time a request comes in for the rule engine to run all rule , the flag is checked and the RuleBase is updated.
So, my question is about how to handle the rulebase updates most efficiently.
I see two possibilities:
1. ditch the rulebase completely, build a new Package of rules , create a new rulebase and add the package to it
2. Use the same rulebase - just remove the old package of rules and load in the new one.
The issues here are:
-- what if there are operations going on in the rule engine, invoked from other threads, at the time I want to do either 1 or 2?
Say, in the scenario #2, the rule engine is running all rules in one thread. At the same time I'm asking it to drop the rule package from another thread and add the new one. Is it going to finish running all rules with the old package, return results, and only then do the update of the package?
And in the scenario #1, if the rule engine is running the rules, and I try to destroy the old rule base - are those operations sequential?
Any suggestions as to how to design this functionality in the most efficient and safe way?
thank you!
Marina Popova
_______________________________________________
rules-users mailing list
rules-users@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss, a division of Red Hat @
www.jboss.com