We are just getting our heads around using Optaplanner for a project. We have a very simple solution setup as per following:
Job -> PlanningEntity, PlanningVariable=Resource from resourcesList
Resource -> POJO
Solution
- List<Job> PlanningEntityCollectionProperty
- List<Resource> ProblemFactCollectionProperty, resourcesList
We have setup some rules for testing. The first rule is simply to say, don't assign more than three Jobs to a Resource:
rule "noMoreThan3JobsPerResource"
when
$resource : Resource()
$totalJobsOnResource : Number(intValue > 3) from accumulate (
Job(
resource == $resource,
$count : 1),
sum($count)
)
then
scoreHolder.addHardConstraintMatch(kcontext, 3 - $totalJobsOnResource.intValue());
end
What we want to understand is HOW and WHEN the drools rules are evaluated. For example, if we add these two rules:
rule "logWhenResource"
when
$resource: Resource()
then
System.out.println("RESOURCE encountered");
end
rule "logWhenJob"
when
$job : Job()
then
System.out.println("JOB encountered");
end
We get "JOB encountered" in the log, but never "RESOURCE encountered". And yet, our first rule has $resource : Resource() in the when? Does optaplanner fire a rule when a job is placed (in our example)? We are just a bit unclear on why logWhenResource doesn't fire, but noMoreThan3JobsPerResource does (when they both try and 'match' a Resource object? Is Resource the resource that a job has been moved to?
Thanks in advance
After some discussions on IRC, (and a lot of patient help from Geoffrey!), hopefully the following will serve as a helper for other people.
1. Turn Logging on
First off, make sure you turn on trace logging for the Optaplanner package (and maybe turn it off for drools). This really helps as it shows exactly when optaplanner is triggering score calculations. It also shows the candidate score calculation:
Move index (0), score (-3init/-2hard/0medium/0soft), move (Job 7 {null -> Resource 1}).
in addition to the final step selection:
CH step (6), time spent (110), score (-3init/-2hard/0medium/0soft), selected move count (2), picked move (Job 7 {null -> Resource 1}).
You can also log in your "then" part of Rules, by doing something like:
LoggerFactory.getLogger("org.optaplanner").debug("...);
This makes sure it gets logged in the right order as Logging vs println can be asynchronous and things may not be in time ascending order.
2. Understand when Optaplanner calculates scores, and when it doesn't
This is a pretty useful summary of the 'event loop' of optaplanner:
doMove()
fireAllRules()
undoMove()
doMove()
fireAllRules()
undoMove()
doStep()
doMove()
fireAllRules()
undoMove() ...
etc. One thing that is interesting, as per our chat on IRC is the following:
"Notice that it doesn't do fireAllRules() after an undoMove or after doStep() because it can predict the score". Neat.
3. FULL_ASSERT
To check whether you are corrupting the score, turn on FULL_ASSERT.
<environmentMode>FULL_ASSERT</environmentMode>
This is useful to determine if your score calculation isn't right (ours wasn't).
Turn on TRACE logging. It fires all rules (that have changed since last time because it's incremental calculation) every time there's a move line in that log.
Related
We have rest resource
/tasks/{task-type}
and only GET methods available.
GET /tasks/{task-type}
GET /tasks/{task-type}/{id}
Task entity contains meta info like created, finished, status, ref key and try counts for scheduled tasks.
Now we faced with problem, when task may contains incorrect data and its execution always failed.
Due to scheduler invoked tasks every 5 min there are a lot of errors in logs and largest try counts around 500k. The solution i found is to limit try_count to five (for example). And now we need way to manual discard try-count to zero. So i found two solutions:
1.
PATCH /tasks/{task-type}/{id}/discard-try-count - no response body
This solution look pretty simple, but violates the REST convention, because we use action(verb) in naming. But if we need to change other fields, then we will make a lot of endpoints in this style.
2a.
PATCH /tasks/{task-type}/{id}
body:
{
"tryCounts": int
}
This looks like REST want to see it and we can easy add new fields to modify, but now client can set any value for tryCount.
2b
PATCH /tasks/{task-type}/{id}
body:
{
"tryCounts": int // validate that try count can be only zero
}
Differs from the previous one by the presence of validation.
This looks like the most reliable solution. Is it really the best fit?
The non-verb convention is not a standard, you can violate it if you want to, though it can be worked around with very simple stuff, just convert the verb into a noun and you will be ok, something like:
POST /tasks/{task-type}/{id}/try-count-discarding
Another way is setting the try count to zero:
PUT /tasks/{task-type}/{id}/try-count 0
Yet another solution is combining the two, which I like the most:
PATCH /tasks/{task-type}/{id}/try-count {"op": "reset"}
Or another variant:
PATCH /tasks/{task-type}/{id} {"op": "discard-try-count"}
rule "attaching AV and impact rating"
agenda-group "evaluate likelihood"
dialect "java"
when
Application($threatList:getThreatList())
$av:AttackVector()
exists $threat:Application.Threats(impact == "Disclose Information")from $threatList
exists AttackVector($av == AttackVector.REQUEST_MANIPULATION)
then
RiskRating riskRating=new RiskRating($threat.getImpactRating(),$av.getLikelihood(),$av.getName());
insertLogical(riskRating);
end
I am working on getting the object $threat in THEN part of the above-mentioned rule. If I run the above rule, it says:
Rule Compilation error : [Rule name='attaching AV and impact rating']
referee/security/attack/Rule_attaching_AV_and_impact_rating1426933818.java (7:1053) : $threat cannot be resolved
If I loop through it and get the value in the THEN part, it causes a CARTESIAN product and inserts the values a number of times in the session. My rule looks like this when I get the cartesian product.
rule "attaching AV and impact rating"
agenda-group "evaluate likelihood"
dialect "java"
when
Application($threatList:getThreatList())
$av:AttackVector()
exists $threat:Application.Threats(impact == "Disclose Information")from $threatList
$threat:Application.Threats(impact == "Disclose Information")from $threatList
exists AttackVector($av == AttackVector.REQUEST_MANIPULATION)
then
RiskRating riskRating=new RiskRating($threat.getImpactRating(),$av.getLikelihood(),$av.getName());
insertLogical(riskRating);
end
How do I get the value of $threat in THEN part without having the cartesian product?
Remove the exists operation entirely.
rule "attaching AV and impact rating"
agenda-group "evaluate likelihood"
dialect "java"
when
Application($threatList:getThreatList())
$av: AttackVector()
$threat: Application.Threats(impact == "Disclose Information")from $threatList
exists(AttackVector($av == AttackVector.REQUEST_MANIPULATION))
then
RiskRating riskRating=new RiskRating($threat.getImpactRating(),$av.getLikelihood(),$av.getName());
insertLogical(riskRating);
end
exists means "there is a thing in working memory that matches these conditions/looks like this". It's not used to actually extract or provide a reference to that matching instance. Simply remove the operator and it works as you need -- if there is an Application.Threats that matches your conditions, the rule triggers and the matching value is assigned to the $threat variable.
What you're running into is the fact that you have multiple threats that mean your condition, which is why you're having multiple triggers of the rule -- it will trigger once per matching Application.Threats. The exists keyword mitigates this because it only cares that at least one match exists, but you don't get a reference (because if there's four matches which one will be assigned to the variable? it doesn't make sense, logically.)
So you need to change your rule so that it won't fire multiple times and will instead only fire once when it finds a match. Usually you'd do this by making the consequences do something to working memory that makes the rule no longer eligible to be fired. In your example, you insert a RiskRating object; you could, then, check that no risk rating exists in your conditions:
not( RiskRating( /* insert criteria here or leave empty */ ) )
Alternatively you could retract something from working memory that your rule relies on to be present or a match. For example, if you don't need it for anything later on, you could retract the attack vector:
retract( $av )
Yet another option might be to try and update your getThreatList() implementation to return a Set instead so you don't have duplicates (assuming threats are considered duplicates based on the 'impact' field.) Or you could try to remove all Application.Threats instances that match the criteria from the threatlist being returned.
We simply don't know enough about your use case or rule set to know what data you need or what it looks like, but at the end of the day you simply need the rule to fire once and only once, so to do this you need to somehow update the rule to know that it's no longer valid.
I would like the "then" clause to only execute one time, but it is executing for each child object in the list that matches.
If any item in the list meets the condition I want to break out and have only one execution of the then clause.
rule "Profile - Delinquent"
when
$c : CreditReportAll( $creditLiability : creditLiability )
$cs: CreditLiability( paymentPatternData.contains("X")) from $creditLiability
then
CreditUserSegment $cu = new CreditUserSegment();
$cu.setSegmentCode("delinquent");
$c.addUserSegmentToList($cu);
end
In order to only fire the rule once when any item in the $creditLiability list meets the condition, you either need to write your rule such that it doesn't iterate across the list, or you need to update your rule so that once it does fire, it changes the facts in working memory to not allow it to fire again.
No iteration
The easiest way to do this is to change your rule to not iterate across the list. To do this, we use the exists keyword like this:
rule "Profile - Delinquent"
when
$c : CreditReportAll( $creditLiability : creditLiability )
exists( CreditLiability( paymentPatternData.contains("X")) from $creditLiability )
then
CreditUserSegment $cu = new CreditUserSegment();
$cu.setSegmentCode("delinquent");
$c.addUserSegmentToList($cu);
end
The exists keyword will match when the there is at least one element present that matches the required condition. Note that we don't assign a variable to the result anymore, because it doesn't make any sense (eg. there's no assignment of $cs here; it would be ambiguous as to what it would even refer to.)
The downside to this approach is that if you update working memory in any other rule (eg by calling insert, modify, update, and so on), you may end up triggering this rule again because the conditions on the left hand side will still remain valid and matching. To alleviate this you may be able to leverage the no-loop rule attribute (depending on your setup). Otherwise you'll want to update your rule (or data in working memory) so that your rule is no longer valid.
Invalidate the rule
The other way to only trigger the rule once is to make the rule no longer valid to fire after it fires once. A trivial way to do this (likely not best practice in this case) would be to insert a flag into working memory and check on its presence. In this case I will use a simple string "DELINQUENT" as the flag.
rule "Profile - Delinquent"
when
not(String(this == "DELINQUENT"))
$c : CreditReportAll( $creditLiability : creditLiability )
$cs: CreditLiability( paymentPatternData.contains("X")) from $creditLiability
then
CreditUserSegment $cu = new CreditUserSegment();
$cu.setSegmentCode("delinquent");
$c.addUserSegmentToList($cu);
insert("DELINQUENT");
end
When the rule fires, it inserts a string that says "DELINQUENT" into working memory. The rule conditions are such that the rule only fires if this string doesn't exist in working memory. Thus after the first execution, the rule will not execute again unless a rule retracts that string.
This solution increases the memory footprint of the rule execution because there is more information in working memory. However unlike the other solution (which is more elegant), this version will not re-fire if another rule retriggers execution (eg. via update.)
I am newbie to drools. Need understanding how to cascade filtered data from one condition to next one as shown below.
rule "rulename"
when
$var1 : MainObj(filter1)
$var2 : SubObj() from $var1.getSubObj // need all SubObj without any filter here for then clause.
$var3 : SubObj(conditions) from $var2 // Need a subset of SubObj for another then clause
then
$var2.doSomething()
$var3.doVar3Stuff()
end
How can I achieve that?
Continuation of my original post..
i have below rule computing statistics of a rule execution as below.
rule "myrule"
when
$a:wrapper(isActive("ruleName") && dataTypeCode.equals("typeCodeinDB"))
$total : Number() from accumulate($entity: MyObj(ObjAttribute=='testValue') from $a.getMyObj() , count( $entity ))
$filter: MyObj(ObjAttribute=="testValue" && ObjAttribute2.equals("ValidValue")) from $a.getMyObj()
then
$filter.addToResults($filter.getRuleConstituents(),1,drools.getRule().getName(),null); // Add data for insertion in DB at later stage
$filter.addPayloadExecution($a.dataTypeCode(),$a.getMyObj().toString(),$a.getMyObj().size()); //Total Count of MyObj
$filter.addRuleExecutions(drools.getRule().getName(), $total); // anotehr slice of data
end
Here.. i am computing two stats.. Total count of MYObj(), apply some filter for data validation and compute count of valid list ($filter is valid data)
When I run that code I am getting error as below
Exception executing consequence for rule "MyRule" in moneta.rules: [Error: $filter.addRuleExecutions(drools.getRule().getName(), $total): null]
[Near : {... $filter.addToResults($filter.get ....}]
^
[Line: 1, Column: 1]
Need advice on resolving this.
I can't tell if you're trying to change the data in one rule and make it visible in another, or if you're trying to do a single complex condition in a single rule. So I'll outline how both work.
Complex conditions
The pseudo-code in your rule indicates that you have a main object with a filter and a collection of sub-objects. You need to get a list of all sub-objects without a filter, and also to get a list of sub-objects that meet some sort of condition.
To collect these sublists, we have two available operations: accumulate and collect. Since you didn't give a real example, just some psuedo-code, I can't say for sure which is the correct one for you to use. The example I present below uses collect because it is appropriate for this use case. For more information about these operations, refer to the Drools documentation. I've linked to the section on operators; scroll down to see the details for collect and accumulate.
So imagine I have an app that is modeling a school, and my KindergartenClass object has a List of Student objects. In this case, KindergartenClass is the main object, and Student is the sub-object. My Students have names, ages, and an Allergies object that indicates food allergies. Students without Allergies have no allergy object. This is going to be the filter for the example rule -- finding the students without allergies.
The rule would then look like:
rule "EXAMPLE"
when
KindergartenClass( $students: students ) // $var1
// Find all students with no allergies, $var2
$var2: List() from collect( Student( allergies == null ) from $students )
// Find all students who are older than 5, $var3
$var3: List() from collect( Student( age > 5 ) from $students )
then
...
end
Obviously your right-hand-side would have to be adjusted because 'List' has no doSomething() and doVar3Stuff() methods. But without an actual example rule for what you're attempting I can't be more specific here.
You could get the individual students from the lists like this:
$studentWithoutAllergies: Student() from $var2
$studentOlderThan5: Student() from $var3
... but then your then-clause would trigger once for every single $studentWithoutAllergies and $studentOlderThan5. Most people don't want a cartesian product all elements in two lists and then having their consequences firing for each of those products.
Of course, you also talk about "cascading", though it's rather unclear what you mean. If you mean that you want $var3 to be a subset of what you found in $var2, then it's as simple as changing the from clause. So in the example I've been working with, if you actually want all students older than 5 who have no allergies, you can simply change the when clause to be:
// Find all students with no allergies, $var2
$var2: List() from collect( Student( allergies == null ) from $students )
// Find the subset of the non-allergy students who are older than 5, $var3
$var3: List() from collect( Student( age > 5 ) from $var2 )
Changing the data that triggers the rule
There are two keywords available for changing the data available to the left hand side (when clause): update and insert. Which one you use depends on what you're trying to do, and come with very different considerations in terms of performance.
When you ask Drools to fire a set of rules, it first goes through all of the rules and determines if the rule fires, given the data you've inputted. Basically it just goes through, in order, all of the when clauses and decides if the rule is activated or not. Then, once it has the sub-set of rules, in order, it goes through, one by one, executing the right hand side.
If you call update in your right hand side (then clause), then this process repeats: all rules are evaluated again, from the top, including the rule you just fired. If you call insert in your then clause, then the new data is put into working memory, and Drools re-evaluates all of the remaining rules in its execution list. (So, for example, if Drools had decided that rules A, B, C, and D met their when criteria, and B inserts a new fact into working memory, then C and D would be re-evaluated to make sure that they were still valid.)
This is how you'd use them:
rule "Example rule with Update"
when
$input: MyAwesomeInputObject(someValue == 0)
then
$input.setSomeValue(99);
$input.doSomeStuff();
update($input);
end
rule "Example rule with Insert"
when
Person(name == "Bob")
not( Age() )
then
Age age = new Age(18);
insert(age); // there is now an Age object in working memory
end
I'm not 100% familiar with your use case, but I'm going to assume you want to fire all of the filtering rules a second time after your initial rule fires, with the updated data. So I'd do something like this:
rule "rulename"
when
$var1 : MainObj(filter1)
$var2 : SubObj() from $var1.getSubObj // need all SubObj without any filter here for then clause.
$var3 : SubObj(conditions) from $var2 // Need a subset of SubObj for another then clause
then
$var2.doSomething();
$var3.doVar3Stuff();
update($var1); // update the var1 in working memory
end
I have no idea what your when clause is trying to do, since it doesn't appear to match the comments and the syntax is completely wonky. So only the right hand side (then clause) has been adjusted. I have added update($var1) so it will refire the rules with the new instance of $var1 that should have your changes in it.
Of course, once you start re-firing all rules, you run the risk of having rules loop. If this does happen, you will see spikes of CPU and apparent thread deadlocks (this is useful to alert on.) Note that there is a no-loop property, but it won't help when you're calling update.
relative new comer to drools and rules engines.
I have three rules that essentially check if different instances of the same object exist, if they don't, they create one and insert it into the working memory. After performing a series of other related rules, theres a rule that checks for validity of the models.
The goal here is to have three different instances of the same object that represent three different entities and proceed as long as 2 of them are in a valid state.
They're defined as follows:
rule "SetupA"
salience 80
lock-on-active
when
not InvalidModel(name == "A-Name")
then
InvalidModel invalidA = new InvalidModel();
invalidA.setName("A-Name");
insert(invalidA);
end
rule "SetupB"
salience 80
lock-on-active
when
not InvalidModel(name == "B-Name")
then
InvalidModel invalidB = new InvalidModel();
invalidB.setName("B-Name");
insert(invalidB);
end
rule "SetupC"
salience 80
lock-on-active
when
not InvalidModel(name == "C-Name")
then
InvalidModel invalidC = new InvalidModel();
invalidC.setName("C-Name");
insert(invalidC);
end
rule "VALIDATE_MODEL_FOR_ANY_INVALID_FLAGS"
salience -10
when
$invalidList: List(size > 1) from collect(InvalidModel(isNotFound || isFlagged || isNotActive))
then
setDecision(result, "Denied");
end
However, what's happening is that it seems like these rules are matched on every request to this ruleset which I interpret as the facts not existing in the Working Memory. I'd expect the rules to be fired once, when each of the facts for the respective names dont exist, and then to not fire again as there is nothing that would remove them from the working memory. However, my audit log of the rules that fire show that for every request those rules get fired.
Furthermore, the fourth rule's condition seems to not be performing as expected as it returns a response like such
=============== DEBUG MESSAGE: illegal bytecode sequence - method not verified ================
My ask is for some guidance on where my logic/drools language structure may have gone astray. Thanks for your time.