+++ This bug was initially created as a clone of Bug #1153852 +++

For a BRMS client application that integrates with Kie API and usage within the client is such that KieRepositoryScanner is registered with KieContainer (with module defined by "LATEST" GAV release id) to periodically check for newer release versions of the KieModule available in remote BRMS knowledge store (business central), download the newer version and automatically update the existing KieContainer used by the client application.

This works as expected for basic smaller Kie projects consisting of single DRL file containing one or two rules (modifications to LHS rule constraints in new versions of the KieModule were applied automatically when scanner retrieved the new version).

However, after adding several more rules and several functions to the DRL file, the KieRepositoryScanner dynamic update to the client's existing KieContainer sometimes results in the modified rules or the rules calling modified functions to not fire on subsequent KieSessions.

After some diggings, the issue seems to be narrowed down to be related to the functions.

If a rule makes a call to a function then any changes to either the rule or the function will cause the rule to not fire after the kieScanner picks up the modifications.

One other issue is that if you make a change that removes the call to the function from the rule, you get a NPE error with the stack trace:

Exception in thread "Timer-0" java.lang.NullPointerException
at org.drools.core.reteoo.ReteooBuilder.removeRule(ReteooBuilder.java:156)
at org.drools.core.reteoo.ReteooRuleBase.removeRule(ReteooRuleBase.java:1415)
at org.drools.core.reteoo.ReteooRuleBase.removeRule(ReteooRuleBase.java:1406)
at org.drools.compiler.compiler.PackageBuilder.preProcessRules(PackageBuilder.java:1146)
at org.drools.compiler.compiler.PackageBuilder.compileRules(PackageBuilder.java:1089)
at org.drools.compiler.compiler.PackageBuilder.compileAllRules(PackageBuilder.java:973)
at org.drools.compiler.compiler.CompositeKnowledgeBuilderImpl.buildRules(CompositeKnowledgeBuilderImpl.java:266)
at org.drools.compiler.compiler.CompositeKnowledgeBuilderImpl.buildPackages(CompositeKnowledgeBuilderImpl.java:98)
at org.drools.compiler.compiler.CompositeKnowledgeBuilderImpl.build(CompositeKnowledgeBuilderImpl.java:87)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.rebuildAll(KieContainerImpl.java:302)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.update(KieContainerImpl.java:178)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.updateDependencyToVersion(KieContainerImpl.java:114)
at org.kie.scanner.KieRepositoryScannerImpl.updateKieModule(KieRepositoryScannerImpl.java:231)
at org.kie.scanner.KieRepositoryScannerImpl.scanNow(KieRepositoryScannerImpl.java:218)
at org.kie.scanner.KieRepositoryScannerImpl$ScanTask.run(KieRepositoryScannerImpl.java:198)
at java.util.TimerThread.mainLoop(Timer.java:555)
at java.util.TimerThread.run(Timer.java:505)

The following is the test rule I'm using to show the issues.

package org.drools.examples.shopping

import java.lang.Number;
import java.lang.String;

function boolean doSomething(String name) {
System.out.println("Function Called ZZZZZ");
return true ;
}

rule "SecondRule" no-loop
dialect "java"
when
$s : String( )
eval(doSomething($s))
then
System.out.println("Firing Second Rule XXX " + $s);
end

The first time after the rule gets compiled and kjar gets installed in local maven repo(or uploaded to business central) everything works fine.

When the BRMS client application keeps running and a change is made over the rule and a new kjar gets built, for example,
change
System.out.println("Firing Second Rule XXX " + $s);
to
System.out.println("Firing Second Rule XXXxxx " + $s);
after the kjar is installed in local maven repo(do "vn clean install" against the kjar project), this rule stop firing any more.

If a change is made to remove the function call from the rule, for example,
change
eval(doSomething($s))
to
//eval(doSomething($s))
after the kjar is installed in maven repo, the NPE error I mentioned above is thrown.

— Additional comment from Gary Hu on 2014-10-16 22:26:32 EDT —

— Additional comment from JBoss Product and Program Management on 2014-10-16 22:30:07 EDT —

Since this issue was entered in Red Hat Bugzilla, the release flag has been
set to ? to ensure that it is properly evaluated for this release.

— Additional comment from Gary Hu on 2014-10-16 22:40:38 EDT —

I've attached a reproducer that could demonstrate the issue.

The reproducer zip file contains two maven projects bpms.test.01256902.sample.kjar and bpms.test.01256902.

The first one is to build the kjar and second one is the test client which uses junit.

Steps to reproduce the issue:

1) unzip the reproduce and go to project bpms.test.01256902.sample.kjar, run command
mvn clean install
This should install the kjar in your local maven repo.

2) go to project bpms.test.01256902 and run command
mvn clean package
or
mvn clean test
This should build the test client and start the junit test.
From console you should see some printouts that shows something like:
Firing Second Rule XXX 0
Firing Second Rule XXX 1
Firing Second Rule XXX 2
Firing Second Rule XXX 3
Firing Second Rule XXX 4

This shows the rule "SecondRule" gets fired without issue.

3) keep the test client running, and make a change to the "SecondRule".
for example:
change
System.out.println("Firing Second Rule XXX " + $s);
to
System.out.println("Firing Second Rule XXXxxx " + $s);

and run command
mvn clean install

After the maven command finish successfully, you can see the rule stop firing. It doesn't show on the console any more.

4) make further change, say, removing the function call from the rule.
for example,
change
eval(doSomething($s))
to
//eval(doSomething($s))
and run command
mvn clean install
After it finishes successfully, you should see NPE error:

Exception in thread "Timer-0" java.lang.NullPointerException
at org.drools.core.reteoo.ReteooBuilder.removeRule(ReteooBuilder.java:156)
at org.drools.core.reteoo.ReteooRuleBase.removeRule(ReteooRuleBase.java:1415)
at org.drools.core.reteoo.ReteooRuleBase.removeRule(ReteooRuleBase.java:1406)
at org.drools.compiler.compiler.PackageBuilder.preProcessRules(PackageBuilder.java:1146)
at org.drools.compiler.compiler.PackageBuilder.compileRules(PackageBuilder.java:1089)
at org.drools.compiler.compiler.PackageBuilder.compileAllRules(PackageBuilder.java:973)
at org.drools.compiler.compiler.CompositeKnowledgeBuilderImpl.buildRules(CompositeKnowledgeBuilderImpl.java:266)
at org.drools.compiler.compiler.CompositeKnowledgeBuilderImpl.buildPackages(CompositeKnowledgeBuilderImpl.java:98)
at org.drools.compiler.compiler.CompositeKnowledgeBuilderImpl.build(CompositeKnowledgeBuilderImpl.java:87)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.rebuildAll(KieContainerImpl.java:302)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.update(KieContainerImpl.java:178)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.updateDependencyToVersion(KieContainerImpl.java:114)
at org.kie.scanner.KieRepositoryScannerImpl.updateKieModule(KieRepositoryScannerImpl.java:231)
at org.kie.scanner.KieRepositoryScannerImpl.scanNow(KieRepositoryScannerImpl.java:218)
at org.kie.scanner.KieRepositoryScannerImpl$ScanTask.run(KieRepositoryScannerImpl.java:198)
at java.util.TimerThread.mainLoop(Timer.java:555)
at java.util.TimerThread.run(Timer.java:505)

— Additional comment from Mario Fusco on 2014-10-17 13:17:10 EDT —

I couldn't reproduce any NPE on master. Nevertheless using the attached reproducer I found that incremental compilation doesn't work correctly when a rule uses a function.

I fixed this problem with this commit: https://github.com/droolsjbpm/drools/commit/d905672bf

I also checked 6.0.x branch and found that it suffers of exactly the same problem I found on master. Moreover the fix I pushed on master also solved this issue on 6.0.x

— Additional comment from Mario Fusco on 2014-10-17 13:41:18 EDT —

Backported to 6.0.x with https://github.com/droolsjbpm/drools/commit/f044ef89c

— Additional comment from Gary Hu on 2014-10-20 09:56:11 EDT —

Hi Mario,

I've created a "patch" for BPMS 6.0.3 based on your fix in https://github.com/droolsjbpm/drools/commit/f044ef89c.

After some more tests, we can verify that it fixes the following scenarios:

1) A new version of the kjar includes a modification to a rule that calls an existing function
2) A new version of the kjar includes the removal of a function entirely from the DRL and modifications to any rules that previously called removed function.

However, the patched jar does not fix the following kie module update scenario:

3) A new version of the kjar includes the addition of a new function to the DRL file as well as a modification to a rule in the same DRL to call the new function.

Are you able to look into the issue again to see if we can get the 3) issue fixed?

— Additional comment from Gary Hu on 2014-10-20 10:22:58 EDT —

I just tested the scenario 3) mentioned in your previous post:

3) A new version of the kjar includes the addition of a new function to the DRL file as well as a modification to a rule in the same DRL to call the new function.

I added one new function in the test drl rule.
for example:
function boolean doSomething2(String name)

And I changed the rule
from
System.out.println("Firing Second Rule XXX " + $s);
to
System.out.println("Firing Second Rule XXXxxx " + $s);

What happened in my test is that I received the following error:

Exception in thread "Timer-0" java.lang.NoSuchMethodError: org.drools.core.reteoo.ReteooRuleBase.removeObjectsGeneratedFromResource(Lorg/kie/api/io/Resource;)Z
at org.drools.compiler.compiler.PackageBuilder.removeObjectsGeneratedFromResource(PackageBuilder.java:4181)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.updateResourcesIncrementally(KieContainerImpl.java:260)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.update(KieContainerImpl.java:161)
at org.drools.compiler.kie.builder.impl.KieContainerImpl.updateDependencyToVersion(KieContainerImpl.java:114)
at org.kie.scanner.KieRepositoryScannerImpl.updateKieModule(KieRepositoryScannerImpl.java:231)
at org.kie.scanner.KieRepositoryScannerImpl.scanNow(KieRepositoryScannerImpl.java:218)
at org.kie.scanner.KieRepositoryScannerImpl$ScanTask.run(KieRepositoryScannerImpl.java:198)
at java.util.TimerThread.mainLoop(Timer.java:555)
at java.util.TimerThread.run(Timer.java:505)

In the meanwhile, the client application continue to fire the old rule. Any changes to the rule are not applied.

— Additional comment from Mario Fusco on 2014-10-20 12:10:11 EDT —

I also fixed scenario 3) with this commit on master https://github.com/droolsjbpm/drools/commit/e4b1bff10 and backported the fix on 6.0.x branch https://github.com/droolsjbpm/drools/commit/dd4901a29

— Additional comment from Marek Winkler on 2014-12-16 10:15:36 EST —

Verified on BRMS 6.1.0 ER3.

I have encountered unexpected behaviour with LATEST versions - it has been filed as BZ 1174836.