[rules-users] The effect of not using shadow facts
trandinh tho
tdtho_cs at yahoo.com
Mon Jul 23 22:09:04 EDT 2007
Hello,
I got same problem to Chris's. Java classes are generated from XML schema (using XMLBean). And the problem occurs when I try to assert objects of those classes into working memory. The exception is:
java.lang.VerifyError: class b2BContext.impl.UserImplShadowProxy overrides final method hashCode.()I
at java.lang.ClassLoader.defineClass1(Native Method)
at java.lang.ClassLoader.defineClass(ClassLoader.java:620)
at java.lang.ClassLoader.defineClass(ClassLoader.java:465)
at org.drools.rule.MapBackedClassLoader.fastFindClass(MapBackedClassLoader.java:40)
at org.drools.rule.MapBackedClassLoader.loadClass(MapBackedClassLoader.java:59)
at java.lang.ClassLoader.loadClass(ClassLoader.java:251)
at org.drools.reteoo.Rete$ObjectTypeConf.<init>(Rete.java:352)
at org.drools.reteoo.Rete.assertObject(Rete.java:152)
at org.drools.reteoo.ReteooRuleBase.assertObject(ReteooRuleBase.java:190)
at org.drools.reteoo.ReteooWorkingMemory.doInsert(ReteooWorkingMemory.java:70)
at org.drools.common.AbstractWorkingMemory.insert(AbstractWorkingMemory.java:772)
at org.drools.common.AbstractWorkingMemory.insert(AbstractWorkingMemory.java:584)
at org.drools.jsr94.rules.StatefulRuleSessionImpl.addObject(StatefulRuleSessionImpl.java:162)
at org.drools.jsr94.rules.StatefulRuleSessionImpl.addObjects(StatefulRuleSessionImpl.java:185)
I am new to JBoss Rule, so may you show me how to disable Shadow. Please help.
Thanks,
Anthony.
Edson Tirelli <tirelli at post.com> wrote:
Chris,
I know why it happens, but I don't know what to do. Basically, we need to do a shallow clone of any collections asserted to the working memory to ensure integrity. So, what I try to do is:
* Check if the collection is cloneable. If it is, use clone method.
* Else, check if the collection has a default no-arg constructor. If so, create a new instance and use addAll() method to add all previous elements.
* Otherwise, use objenesis to instantiate object without calling the constructor.
What is happening in your case, as you don't have a default constructor and apparently is not cloneable, it is falling to the 3rd alternative above, and as you are extending a java.util.Collection class, it is raising the NPE because it is not executing the class constructor.
Not sure about how to handle such scenario since we don't get such exception until it is too late to rollback.
[]s
Edson
2007/7/20, Chris West <crayzfishr at gmail.com>: Edson,
It appears that revision #13637 of drools breaks the ability for me to use one of my existing classes. The attached eclipse project illustrates the problem. This test works on 4.0.0MR3, but not revision #13637. The exception is:
org.drools.spi.ConsequenceException: org.drools.RuntimeDroolsException: Error creating shadow fact for object: NamedList(Hello List): [1, 2, 3]
at org.drools.common.DefaultAgenda.fireActivation(DefaultAgenda.java :549)
at org.drools.common.DefaultAgenda.fireNextItem(DefaultAgenda.java:509)
at org.drools.common.AbstractWorkingMemory.fireAllRules(AbstractWorkingMemory.java:430)
at org.drools.common.AbstractWorkingMemory.fireAllRules (AbstractWorkingMemory.java:392)
at com.sample.DroolsTest.main(DroolsTest.java:29)
Caused by: org.drools.RuntimeDroolsException: Error creating shadow fact for object: NamedList(Hello List): [1, 2, 3]
at org.drools.reteoo.Rete$ObjectTypeConf.getShadow(Rete.java:458)
at org.drools.reteoo.Rete.assertObject(Rete.java:157)
at org.drools.reteoo.ReteooRuleBase.assertObject(ReteooRuleBase.java:190)
at org.drools.reteoo.ReteooWorkingMemory.doInsert (ReteooWorkingMemory.java:70)
at org.drools.common.AbstractWorkingMemory.insert(AbstractWorkingMemory.java:848)
at org.drools.common.AbstractWorkingMemory.insert(AbstractWorkingMemory.java:822)
at org.drools.base.DefaultKnowledgeHelper.insert (DefaultKnowledgeHelper.java:60)
at org.drools.base.DefaultKnowledgeHelper.insert(DefaultKnowledgeHelper.java:54)
at com.sample.Rule_Insert_named_list_0.consequence(Rule_Insert_named_list_0.java:7)
at com.sample.Rule_Insert_named_list_0ConsequenceInvoker.evaluate(Rule_Insert_named_list_0ConsequenceInvoker.java:19)
at org.drools.common.DefaultAgenda.fireActivation(DefaultAgenda.java:545)
... 4 more
Caused by: java.lang.NullPointerException
at java.util.ArrayList.ensureCapacity(ArrayList.java:163)
at java.util.ArrayList.addAll(ArrayList.java:475)
at com.sample.NamedListShadowProxy.updateProxy(Unknown Source)
at com.sample.NamedListShadowProxy.setShadowedObject(Unknown Source)
at org.drools.reteoo.Rete$ObjectTypeConf.getShadow(Rete.java:456)
... 14 more
The insert works if you modify my NamedList class to have a no arg constructor, but the class misbehaves in my rule set (as if shadow is not working properly).
Please take a look.
Thanks,
-Chris West
On 7/19/07, Edson Tirelli <tirelli at post.com> wrote:
Ouch!
Is all that trouble a result of using JDK proxies in drools? If it is, I think it is the case of us developing a whole set of unit and integration tests for this specific scenario, since none of our tests are triggering errors...
Thanks and please keep me posted of your progress or any problems you find.
[]s
Edson
2007/7/19, Chris West < crayzfishr at gmail.com >:Edson,
Thanks for incorporating this fix. The good news is that it fixes that problem.
The bad news for me is that I'm now experiencing a different problem (where my rules are not firing). I'll look into my new problem a little deeper.
Thanks again.
-Chris West
On 7/19/07, Edson Tirelli < tirelli at post.com> wrote:
Chris,
Right on the spot. I changed other references, but this one passed unnoticed. The correct is:
Class cls = null;
if ( object instanceof ShadowProxy ) {
cls = ((ShadowProxy)object).getShadowedObject().getClass();
} else {
cls = object.getClass();
}
I made a text search this time and found no other occurence of this problem.
I commited the fix in revision #13637. Take a look and let me know if you still has problems.
Thank you a lot,
Edson
2007/7/19, Chris West < crayzfishr at gmail.com>:Edson,
I think I've discovered the problem. In the file Rete.java, in the method "assertObject", there is a check for shadow proxy like below:
Class cls = object.getClass();
if ( object instanceof ShadowProxy ) {
cls = cls.getSuperclass();
}
If the class being proxied was final, and your new logic chose an interface of that class to build a proxy from, then the superclass is Object.class.
This leads to an incorrect selection of cachedNodes further down in the method.
I've traced this through the debugger with my object types, and it does show that a node for a SortieStatus is being given an object of type LaunchRecoveryStatusShadowProxy, which is not compatible.
Perhaps theres a different way to determine the type of object such that type LaunchRecoveryStatusShadowProxy will return LaunchRecoveryStatus rather than Object.
Please take a look and let me know if I need to provide more info.
Thanks,
-Chris West
On 7/18/07, Edson Tirelli < tirelli at post.com> wrote:
Chris,
What seems to be happening us that your SortieStatus interface has a state attribute. Drools is trying to read this attribute value and cast it to LaunchRecoveryStatusShadowProxy what is causing the problems...
Best way to solve would be to have the code so I can debug. Is it possible to isolate it and send me?
[]s
Edson
2007/7/18, Chris West < crayzfishr at gmail.com>:Edson,
It is certainly possible to create a JDK proxy with only some of the interfaces that are present on the delegate object that you are proxying, but in my case, my proxies have all the interfaces of the underlying object.
The top two lines of the call stack I sent shows the following:
Exception in thread "main" java.lang.ClassCastException: ascc.status.FlightOpsStatusBoard$LaunchRecoveryStatusShadowProxy
at org.drools.base.ascc.status.AirPlanStatusBoard$SortieStatus$getState.getValue(Unknown Source)
What's strange here is that the ClassCastException seems to be caused by casting an object of type SortieStatus to type LaunchRecoveryStatusShadowProxy, if I'm reading that right. The types SortieStatus and LaunchRecoveryStatus are both interfaces in my code, and they never appear on the same fact object (so no SortieStatus will ever be a LaunchRecoveryStatus and vice-versa). So I'm wondering why the cast is occuring, since it is not possible to work.
The unfortunate part is I cannot see into the class where the cast is occurring, as it is a generated class created by drools.
-Chris West
On 7/18/07, Edson Tirelli <tirelli at post.com> wrote:
Chris,
For the solution to work, it is important that a superclass or interface matches all the ObjectTypes in your rulebase that your final class (proxy) matches... I guess that is the case with JDK proxies, isn't it?
[]s
Edson
2007/7/18, Chris West <crayzfishr at gmail.com >: Edson,
I downloaded and built the latest from the trunk of the repository. I applied this new build toward my test case, and it seemed to fix the problem. However, when I applied it to my real project, it still exhibits the problem. If I discover more information about the problem I'll let you know.
Thanks,
Chris West
On 7/17/07, Edson Tirelli < tirelli at post.com> wrote:
Chris,
I found and developed an intermediate solution that shall work for your proxies.
If it is not possible to create a shadow fact for a class that is asserted (because the class is final or whatever), the engine goes up in the class hierarchy, looking for a class or interface for which is possible to create the proxy, but that still matches all ObjectTypes available in the rule base matched by the original class. The analysis is a bit complex, specially because new rules with new object types can be dynamically added to the rule base, but I believe the solution will work for JDK proxies and the most common proxy frameworks out there, that usually don't proxy multiple unrelated interfaces at once.
So, I ask you please to get latest snapshot from the repository and try it out for your use case and report back to the list the results, since seems there are a few other people using similar things.
Thanks,
Edson
2007/7/17, Chris West < crayzfishr at gmail.com>: Is that still true if the equals() and hashcode() methods are only based on the identity fields of the object (which cannot change)?
-Chris West
On 7/17/07, Mark Proctor <mproctor at codehaus.org> wrote: you only need to use modifyRetract if the object is inserted. The reason for this is if you change field values on your facts we will not be able to remove them from our various internal hashmaps; thus the need to remove first prior to any changes, then make the changes and then insert it again. We can't allow users to just call update() as we have no idea what the old values where, thus we cannot find the objects in our hashmaps.
Mark
Chris West wrote: Mark,
Using modifyRetract and modifyInsert seems to fix the problem (at least in my test case I finally created). I'll try this on my real code.
My only concern here is that it puts the burden on the rule author to know whether things are being shadowed or not. For shadowing that is explicitly turned off this is ok. But for implicit non-shadowing based on a class being final, this is not at all obvious to the rule auther.
Is there any way to have this hidden such that I can still call "update" but have it use "modifyRetract" and "modifyInsert" instead?
Also, I'm curious why I have to call modifyRetract before I start modifing the object, since the engine does not know about my modifications anyway until I call update or modifyInsert? By the way, I was unable to use the block setter approach in the rule consequence due to not having set methods for modifying my objects.
Thanks,
-Chris West
On 7/17/07, Mark Proctor <mproctor at codehaus.org > wrote: If you do not have shadow facts you cannot use the update() method, it will leave the working memory corrupted. Instead you must manage this yourself, before you change any values on the object you must call modifyRetract() and after you hvae finished your changes ot hte object call modifyInsert() - luckily if you are doing this in the consequence you can use the MVEL modify keyword combined with the block setter and it does this for you:
modify ( person ) { age += 1, location = "london" }
Mark
Chris West wrote: Hello,
With prior versions of JBoss Rules (3.0.5) I have been using JDK generated dynamic proxies as facts, and they have been working fine. However, after upgrading to JBoss Rules 4.0.0MR3, I cannot seem to get the dynamic proxies to work as facts. It seems that even though a rule fires that changes a field on the proxy, a second rule that should not be activated after the update still fires.
According to the JDK javadoc documentation, dynamic proxies are created as final. My assumption is that JBoss Rules is not creating Shadow facts for these since they are final. After reading the JIRA at http://jira.jboss.com/jira/browse/JBRULES-960, I now am questioning what the effect of not using shadow facts is on the engine. The relevant part of that is:
"The problem is that SpringAOP is generating a proxy whose methods equals() and hashCode() are "final". As drools must either override these methods in the shadow proxy or not shadow the fact at all, I'm disabling shadow proxy generation for this use case.
It is really important to note that if you are asserting SpringAOP proxies as facts into the working memory, you will not be able to change any field value whose field is constrained in rules or you may incur in a memory leak and non-deterministic behavior by the rules engine. Unfortunately there is nothing we can do about, since when SpringAOP makes the methods equals and hashcode final, we can't override them anymore and as so, we can't shadow them."
[ Show » ]
Edson Tirelli [02/Jul/07 03:29 PM] The problem is that SpringAOP is generating a proxy whose methods equals() and hashCode() are "final". As drools must either override these methods in the shadow proxy or not shadow the fact at all, I'm disabling shadow proxy generation for this use case. It is really important to note that if you are asserting SpringAOP proxies as facts into the working memory, you will not be able to change any field value whose field is constrained in rules or you may incur in a memory leak and non-deterministic behavior by the rules engine. Unfortunately there is nothing we can do about, since when SpringAOP makes the methods equals and hashcode final, we can't override them anymore and as so, we can't shadow them.
Although I'm not using SpringAOP, I believe my facts are not being shadowed.
Is it true that not using shadow facts may lead to non-deterministic behavior? Prior to shadow facts, the engine seemed to handle it. Any chance of reverting back to the old style of truth maintenance in the case of not using shadow facts.
I apologize if I'm not on the right track here. My only test case for my problem is the entire application right now, so I cannot offer it for discussion. Any advice would be greatly appreciated.
Thanks,
-Chris West
---------------------------------
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
---------------------------------
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
Software Engineer - JBoss Rules Core Developer
Office: +55 11 3529-6000
Mobile: +55 11 9287-5646
JBoss, a division of Red Hat @ www.jboss.com
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
Software Engineer - JBoss Rules Core Developer
Office: +55 11 3529-6000
Mobile: +55 11 9287-5646
JBoss, a division of Red Hat @ www.jboss.com
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
Software Engineer - JBoss Rules Core Developer
Office: +55 11 3529-6000
Mobile: +55 11 9287-5646
JBoss, a division of Red Hat @ www.jboss.com
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
Software Engineer - JBoss Rules Core Developer
Office: +55 11 3529-6000
Mobile: +55 11 9287-5646
JBoss, a division of Red Hat @ www.jboss.com
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
Software Engineer - JBoss Rules Core Developer
Office: +55 11 3529-6000
Mobile: +55 11 9287-5646
JBoss, a division of Red Hat @ www.jboss.com
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
_______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
Software Engineer - JBoss Rules Core Developer
Office: +55 11 3529-6000
Mobile: +55 11 9287-5646
JBoss, a division of Red Hat @ www.jboss.com _______________________________________________
rules-users mailing list
rules-users at lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
---------------------------------
Be a better Globetrotter. Get better travel answers from someone who knows.
Yahoo! Answers - Check it out.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/rules-users/attachments/20070723/1003ed1d/attachment.html
More information about the rules-users
mailing list