Changes in rule evaluation in drools 10 in rules including non-dynamic objects

[daniel-clark-mint]

Hi all, I don't know if this is a defect or just a change in behavior we need to adapt for. This is the last (or second to last...) of the changes we have detected while preparing to upgrade from drools 7 to 10. We have approximately 3500 hand-written rules and 14000 unit test rules to test these in continuous integration, so I am confident we will turn up all "breaking changes" for our use cases.
When firing rules we fill the session with non-dynamic objects from our c++ application using the org.kie.api.runtime.rule.EntryPoint.insert(Object object) function before calling FireAllRules on the same stateful kiesession.
During the rule execution there are a lot of rules which insert objects via insertLogical. We have a lot of rules using the CountAccumulateFunction to inform logic based on the number of objects present, and based upon failing tests it seems some of these tests aren't re-evaluating as suspected (based on drools 7 behavior). I am including a reproducer below (I think it is too complicated but it took 3 days for me to get there from the original rules).

Most importantly for the reproducer:

• When the input objects are inserted by the rules themselves the rule evaluates twice, once before and once after the final object is inserted.
• When the objects are inserted from the java caller before the FireAllRules call, the rule only evaluates once, before insertion of the final object.

Additionally:

• If the "or not" node is removed the rule also fires twice...
• using exists instead of accumulate the rule doesn't seem to evaluate the second time even if the objects are inserted in the rule file.

I assume something has changed around the linking of nodes in the memory path when non-dynamic objects are present and they aren't re-added to the stack - but despite debugging I haven't been able to identify the change in the stack.
I would be happy for advice on if we need to change a runtime compiler setting or just use a different insert method up front (these objects do not change during runtime), or if this is a bug.

Thank you for your assistance!

Reproducer:

package com.example.reproducer;

declare ObjectReference
  id: Integer
  description: String
  type1: String
  type2: String
end

declare AlphaInputObject
  ref: ObjectReference
  value: Integer
end

declare BetaInputObject
  ref: ObjectReference
  type3: String
  answered: Boolean
end

declare BetaInferenceObject2
  ref: ObjectReference
end

// rule "test" evaluates twice with this reproducer, once before rule "3" ($count == 0) and once after ($count == 1).
// however when these facts are inserted from java via EntryPoint.Insert() only the first activation occurs, there is no update after the insert from rule "3".
rule "1" salience 100
  when
  then
    insert( new AlphaInputObject( new ObjectReference( 1, "Input Object 1", "", "" ), 2 ) );
    insert( new BetaInputObject( new ObjectReference( 1, "Input Object 1", "type1", "type2" ), "type3", true ) );
end

rule "test"
  when
    AlphaInputObject(
      $ref: ref,
      value in ( 1, 2 )
    )
    BetaInputObject(
      ref.id == $ref.id,
      ref.description == $ref.description,
      ref.type1 == "type1",
      ref.type2 == "type2",
      type3 == "type3",
      answered == true
    )
    or not BetaInputObject(
      ref.id == $ref.id,
      ref.description == $ref.description,
      ref.type1 == "type1",
      type3 == "type3",
      answered == true
    )
    accumulate(
      BetaInferenceObject2(
        ref.id == $ref.id,
        ref.description == $ref.description
      );
      $count: count()
    )
  then
    System.err.println( Long.toString( $count ) );
end

rule "3"
  when
    AlphaInputObject( $ref: ref, ref.id == 1, ref.description == "Input Object 1" );
  then
    insertLogical( new BetaInferenceObject2( $ref ) );
end

[daniel-clark-mint]

@tkobayas would you also be able to take a look at this for a potential issue, or pass it on to somebody else?
thank you very much in advance!

[tkobayas]

Hello @daniel-clark-mint ,

I reproduced rule "test" evaluates twice with the reproducer (Output is 0 and 1).

Then, I tried inserted from java with the below rules and java code.

https://github.com/tkobayas/kiegroup-examples/blob/master/Ex-evaluation-10.1/src/test/java/com/sample/DroolsTest.java
https://github.com/tkobayas/kiegroup-examples/blob/master/Ex-evaluation-10.1/src/main/resources/com/sample/Sample.drl

But it also evaluates twice (Output is 0 and 1).

Am I misunderstainding your description?

[daniel-clark-mint]

@tkobayas you understood correctly, maybe there is something different about how we our building our kiebase, I'll keep digging, thank you for looking into this!

[daniel-clark-mint]

FYI we also used your reproducer and it worked as expected, and the behavior is also as expected in our productive environment. So we have narrowed the issue down to our test framework, where each test is run on a different thread and enforces locks on various objects which are use to generate the input objects for the test executions.
We will keep looking into this to try and determine more exactly the source or the behavior.
Would you have any tips about changes in the thread-safe behavior of the rule execution between drools 10 and pre-10?

[tkobayas]

Would you have any tips about changes in the thread-safe behavior of the rule execution between drools 10 and pre-10?

Nothing specific comes to mind regarding changes in thread-safety between Drools 10 and pre-10.

Do you share one ksession across mutiple threads? Anyway, ksession is thread-safe unless you explicitly disable it by ThreadSafeOption.

[daniel-clark-mint]

Seems to be related to proprietary framework and not drools.