1 Rules Programming Concepts
This chapter introduces Oracle Business Rules RL Language (RL Language) concepts.
This chapter includes the following sections:
-
Section 1.1, "Starting the Oracle Business Rules RL Language Command-Line"
-
Section 1.6, "Integrating RL Language Programs with Java Programs"
1.1 Starting the Oracle Business Rules RL Language Command-Line
The Oracle Business Rules environment is implemented in a JVM or in a J2EE container by the classes supplied with rl.jar. Start the RL Language command-line interface using the following command:
java -jar $ORACLE_HOME/soa/modules/oracle.rules_11.1.1/rl.jar -p "RL> "
Where ORACLE_HOME is where SOA modules are installed (for example, c:/Oracle/Middleware). The –p option specifies the prompt.
The RL Language command-line interface provides access to an Oracle Business Rules RuleSession. The RuleSession is the API that allows Java programmers to access the RL Language in a Java application (the command-line interface uses a RuleSession internally).
You can run the program in Example 1-1 using the command-line interface by entering the text shown at the RL> prompt.
Example 1-1 Using the Command-Line Interface
RL> println(1 + 2); 3 RL> final int low = -10; RL> final int high = 10; RL> println(low + high * high); 90 RL> exit;
See Also:
-
Chapter 3, "Using the Command-line Interface" for more details and for a list of command-line options
-
Chapter 4, "Using a RuleSession" for details on Oracle Business Rules RuleSession API
1.2 Introducing Rules and Rulesets
An RL Language ruleset provides a namespace, similar to a Java package, for RL classes, functions, and rules. In addition, you can use rulesets to partially order rule firing. A ruleset may contain executable actions, may include or contain other rulesets, and may import Java classes and packages.
An RL Language rule consists of rule conditions, also called fact-set-conditions, and an action-block or list of actions. Rules follow an if-then structure with rule conditions followed by rule actions.
Example 1-2 shows a program that prints, "Hello World." This example demonstrates a program that contains a single top-level action in the default ruleset (named main). Example 1-2 contains only an action, and does not define a rule, so the action executes immediately at the command-line.
See Also:
Understanding and Controlling Rule Firing for details on rule firing
1.2.1 Rule Conditions
A rule condition is a component of a rule that is composed of conditional expressions that refer to facts.
In the following example the conditional expression refers to a fact (Driver instance d1), followed by a test that the fact's data member, age, is less than 16.
if (fact Driver d1 && d1.age < 16)
Example 1-3 shows the complete rule, written in RL Language (the rule includes a rule condition and a rule action).
The Oracle Rules Engine activates a rule whenever there is a combination of facts that makes the rule's conditional expression true. In some respects, a rule condition is like a query over the available facts in the Oracle Rules Engine, and for every row that returns from the query, the rule activates.
Note:
Rule activation is different from rule firing. For more information, see Section 1.4, "Understanding and Controlling Rule Firing".
1.2.2 Rule Actions
A rule action is activated if all of the rule conditions are satisfied. There are several kinds of actions that a rule's action-block might perform. For example, an action in the rule's action-block can add new facts by calling the assert function or remove facts by calling the retract function. An action can also execute a Java method or perform an RL Language function (Example 1-3 uses the println function). Using actions, you can call functions that perform a desired task associated with a pattern match.
1.3 Introducing Facts and RL Language Classes
This section describes Oracle Business Rules facts and includes the following sections:
1.3.1 What Are Facts?
Oracle Business Rules facts are asserted objects. For Java objects, a fact is a shallow copy of the object, meaning that each property is cloned, if possible, and if not, then the fact is a copy of the Java object reference.
In RL Language, a Java object is an instance of a Java class and an RL Object is an instance of an RL Language class. You can use Java classes in the classpath or you can define and use RL Language classes in a ruleset. You can also declare additional properties that are associated with the existing properties or methods of a Java class using a fact class declaration. You can hide properties of a Java class that are not needed in facts using a fact class declaration.
An RL Language class is similar to a Java Bean without methods. An RL class contains set of named properties. Each property has a type that is either an RL class, a Java object, or a primitive type.
Using Oracle Business Rules, you typically use Java classes, including JAXB generated classes that support the use of XML, to create rules that examine the business objects in a rule enabled application, or to return results to the application. You typically use RL classes to create intermediate facts that can trigger other rules in the Oracle Rules Engine.
1.3.2 Adding Facts to Working Memory with Assert
Oracle Business Rules uses working memory to contain facts (facts do not exist outside of working memory). A RuleSession contains the working memory.
A fact in RL Language is an asserted instance of a class. Example 1-4 shows the assert function that adds an instance of the RL class enterRoom as a fact to working memory. A class that is the basis for asserted facts may be defined in Java or in RL Language.
In Example 1-4 the sayHello rule matches facts of type enterRoom, and for each such fact, prints a message. The action new, shown in the assert function, creates an instance of the enterRoom class.
In Example 1-4 the run function fires the sayHello rule.
Note:
The RL Language new keyword extends the Java new functionality with the capability to specify initial values for properties.
1.3.3 Using RL Language Classes as Facts
You can use RL Language classes in a rules program to supplement a Java application's object model, without having to change the application code for the Java application that supplies Java Objects.
Example 1-5 shows the goldCust rule uses a Java class containing customer data, cust; the rule's action asserts an instance of the GoldCustomer RL class, representing a customer that spends more than 500 dollars in a three month period. The Java Customer class includes a method SpentInLastMonths that is supplied an integer representing a number of months of customer data to add.
rule goldCust {
if (fact Customer cust && cust.SpentInLastMonths(3) > 500 ){
assert (new GoldCustomer(cust: cust));
}
}
Example 1-6 shows the goldDiscount rule uses the RL fact GoldCustomer to infer that if a customer spent $500 within the past 3 months, then the customer is eligible for a 10% discount.
rule goldDiscount {
if (fact Order ord & fact GoldCustomer(cust: ord.customer) )
{
ord.discount = 0.1;
assert(ord);
}
}
Example 1-7 shows the declaration for the GoldCustomer RL class (this assumes that you also have the Customer class available in the classpath).
1.3.4 Using Java Classes as Facts
You can use asserted Java objects as facts in an RL Language program. You are not required to explicitly define or declare the Java classes. However, you must include the Java classes in the classpath when you run the program. This lets you use the Java classes in rules, and allows a rules program to access and use the public attributes, public methods, and bean properties defined in the Java class (bean properties are preferable for some applications because the Oracle Rules Engine can detect that a Java object supports PropertyChangeListener; in this case it uses that mechanism to be notified when the object changes).
In addition, Fact class declarations can fine tune the properties available to use in an RL program, and may be required for certain multiple inheritance situations.
When you work with Java classes, using the import statement lets you omit the package name (see Example 1-8).
Example 1-8 Sample Java Fact with Import
ruleset main
{
import example.Person;
import java.util.*;
rule hasNickNames
{
if (fact Person p && ! p.nicknames.isEmpty() )
{
// accessing properties as fields:
println(p.firstName + " " + p.lastName + " has nicknames:");
Iterator i = p.nicknames.iterator();
while (i.hasNext())
{
println(i.next());
}
}
}
See Also:
1.4 Understanding and Controlling Rule Firing
This section covers the following topics:
1.4.1 Rule Activation and the Agenda
The Oracle Rules Engine matches facts against the rule conditions (fact-set-conditions) of all rules as the state of working memory changes. The Oracle Rules Engine only checks for matches when the state of working memory changes, typically when a fact is asserted or retracted. A group of facts that makes a given rule condition true is called a fact set row. A fact set is a collection of all the fact set rows for a given rule. Thus a fact set consists of the facts that match the rule conditions for a rule. For each fact set row in a fact set, an activation, consisting of a fact set row and a reference to the rule is added to the agenda (the agenda contains the complete list of activations).
Figure 1-1 shows a RuleSession with an agenda containing activations in working memory.
Figure 1-1 RuleSession with Working Memory and the Agenda Containing Activations
Description of "Figure 1-1 RuleSession with Working Memory and the Agenda Containing Activations"
The run, runUntilHalt, and step functions execute the activations on the agenda, that is, these commands fire the rules (use the step command to fire a specified number of activations).
Rules fire when the Oracle Rules Engine removes activations, by popping the activations off the agenda and performing the rule's actions.
The Oracle Rules Engine may remove activations without firing a rule if the rule conditions are no longer satisfied. For example, if the facts change or the rule is cleared then activations may be removed without firing. Further, the Oracle Rules Engine removes activations from the agenda when the facts referenced in a fact set row are modified or the facts are retracted, such that they no longer match a rule condition (and this can also happen in cases where new facts are asserted, when the ! operator applies).
Note the following concerning rule activations:
-
Activations are created, and thus rules fire only when facts are asserted, modified, or retracted (otherwise, the rules would fire continuously).
-
If a rule asserts a fact that is mentioned in the rule condition, and the rule condition is still true, then a new activation is added back to the agenda and the rule fires again (in this case the rule would fire continuously). This behavior is often a bug.
-
The actions associated with a rule firing can change the set of activations on the agenda, by modifying facts, asserting facts, or retracting facts, and this can change the next rule to fire.
-
Rules fire sequentially, not in parallel.
See Also:
1.4.2 Watching Facts, Rules, and Rule Activations
You can use the functions watchActivations, watchFacts, watchRules, and showFacts to help write and debug RL Language programs.
This section covers the following topics:
1.4.2.1 Watching and Showing Facts in Working Memory
Example 1-9 shows the watchFacts function that prints information about facts entering and leaving working memory.
As shown in Example 1-9, the watchFacts function prints ==> when a fact is asserted. Each fact is assigned a short identifier, beginning with f-, so that the fact may be referenced. For example, activations include a reference to the facts that are passed to the rule actions.
In Example 1-9, notice that the program uses the default ruleset main. This ruleset contains the enterRoom class.
Example 1-9 Using watchFacts with enterRoom Facts
RL> watchFacts();
RL> class enterRoom {String who;}
RL> assert(new enterRoom(who: "Rahul"));
==> f-1 main.enterRoom(who : "Rahul")
RL> assert(new enterRoom(who: "Kathy"));
==> f-2 main.enterRoom(who : "Kathy")
RL> assert(new enterRoom(who: "Tom"));
==> f-3 main.enterRoom(who : "Tom")
RL>
You can use showFacts to show the current facts in working memory. Example 1-10 shows that the Oracle Rules Engine asserts the initial-fact, f-0 (the Oracle Rules Engine uses this fact internally).
Example 1-10 Show Facts in Working Memory
RL> showFacts(); f-0 initial-fact() f-1 main.enterRoom(who : "Rahul") f-2 main.enterRoom(who : "Kathy") f-3 main.enterRoom(who : "Tom") For a total of 4 facts.
Use retract to remove facts from working memory, as shown in Example 1-11. When watchFacts is enabled, the Oracle Rules Engine prints <== when a fact is retracted.
1.4.2.2 Watching Activations and Rule Firing
The watchActivations function monitors the Oracle Rules Engine and prints information about rule activations entering and leaving the agenda. The watchRules function prints information about rules firing.
Example 1-12 shows how run causes the activations to fire. Notice that Rahul is greeted last even though he entered the room first (this is due to the firing order).
Note:
Activations may be removed from the agenda before they are fired if their associated facts no longer make the condition true.
Example 1-12 Using WatchActivations and WatchRules
RL> clear;
RL> class enterRoom {String who;}
RL> assert(new enterRoom(who: "Rahul"));
RL> assert(new enterRoom(who: "Kathy"));
RL> assert(new enterRoom(who: "Tom"));
RL> watchActivations();
RL> rule sayHello {
if (fact enterRoom) {
println("Hello " + enterRoom.who);
}
}
==> Activation: main.sayHello : f-1
==> Activation: main.sayHello : f-2
==> Activation: main.sayHello : f-3
RL> watchRules();
RL> run();
Fire 1 main.sayHello f-3
Hello Tom
Fire 2 main.sayHello f-2
Hello Kathy
Fire 3 main.sayHello f-1
Hello Rahul
RL>
1.4.3 Ordering Rule Firing
To understand the ordering algorithm for firing rule activations on the agenda, we introduce the ruleset stack. Each RuleSession includes one ruleset stack. The RuleSession's ruleset stack contains the top of the stack, called the focus ruleset, and any non focus rulesets that are also on the ruleset stack. You place additional rulesets on the ruleset stack using either the pushRuleset or setRulesetStack built-in functions. You can manage the rulesets on the ruleset stack with the clearRulesetStack, popRuleset, and setRulesetStack functions. In this case, the focus of the ruleset stack is the current top ruleset in the ruleset stack (see Example 1-13).
Example 1-13 Ruleset Stack - Picture
RuleSet Stack
Focus Ruleset --> Top_Ruleset
Next_down_Ruleset
Lower_Ruleset
Bottom_Ruleset
When activations are on the agenda, the Oracle Rules Engine fires rules when run, runUntilHalt, or step executes. The Oracle Rules Engine sequentially selects a rule activation from all of the activations on the agenda, using the following ordering algorithm:
-
The Oracle Rules Engine selects all the rule activations for the focus ruleset, that is the ruleset at the top of the ruleset stack (see the pushRuleset and setRulesetStack built-in functions).
-
Within the set of activations associated with the focus ruleset, rule priority specifies the firing order, with the higher priority rule activations selected to be fired ahead of lower priority rule activations (the default priority level is 0).
-
Within the set of rule activations of the same priority, within the focus ruleset, the most recently added rule activation is the next rule to fire. However, note that in some cases multiple activations may be added to the agenda at the same time, the ordering for such activations is not defined.
-
When all of the rule activations in the current focus fire, the Oracle Rules Engine pops the ruleset stack, and the process returns to Step 1, with the current focus.
If a set of rules named R1 must all fire before any rule in a second set of rules named R2, then you have two choices:
-
Use a single ruleset and set the priority of the rules in R1 higher than the priority of rules in R2.
-
Use two rulesets R1 and R2, and push R2 and then R1 on the ruleset stack.
Generally, using two rulesets with the ruleset stack is more flexible than using a single ruleset and setting the priority to control when rules fire. For example if some rule R in R1 must trigger a rule in R2 before all rules in R1 fire, a return in R pops the ruleset stack and allows rules in R2 to fire.
If execution must alternate between two sets of rules, for example, rules to produce facts and rules to consume facts, it is easier to alternate flow with different rulesets than by using different priorities.
Example 1-14 shows that the priority of the keepGaryOut rule is set to high, this is higher than the priority of the sayHello rule (the default priority is 0). If the activations of both rules are on the agenda, the higher priority rule fires first. Notice that just before calling run, sayHello has two activations on the agenda. Because keepGaryOut fires first, it retracts the enterRoom(who: "Gary") fact, which removes the corresponding sayHello activation, resulting in only one sayHello firing.
The rule shown in Example 1-14 illustrates two additional RL Language features.
-
The
factoperator, also known as a fact set pattern, uses the optionalvarkeyword to define a variable, in this case the variableg, that is bound to the matching facts. -
You can remove facts in working memory using the
retractfunction.
Example 1-14 Using Rule Priority with keepGaryOut Rule
RL> final int low = -10;
RL> final int high = 10;
RL> rule keepGaryOut {
priority = high;
if (fact enterRoom(who: "Gary") var g) {
retract(g);
}
}
RL> assert(new enterRoom(who: "Gary"));
==> f-4 main.enterRoom(who: "Gary")
==> Activation: main.sayHello : f-4
==> Activation: main.keepGaryOut : f-4
RL> assert(new enterRoom(who: "Mary"));
==> f-5 main.enterRoom(who: "Mary")
==> Activation: main.sayHello : f-5
RL> run();
Fire 1 main.keepGaryOut f-4
<== f-4 main.enterRoom(who: "Gary")
<== Activation: main.sayHello : f-4
Fire 2 main.sayHello f-5
Hello Mary
RL>
Example 1-15 shows the sayHello rule that includes a condition that matches the asserted enterRoom fact; this match adds an activation to the agenda. Example 1-15 demonstrates the following RL Language programming features.
-
The Oracle Rules Engine matches facts against the rule conditions (fact-set-conditions) of all rules as the state of working memory changes. Thus, it does not matter whether facts are asserted before the rule is defined, or after.
-
The run function processes any activations on the agenda. No activations on the agenda are processed before calling run.
Example 1-15 enterRoom Class with sayHello Rule
RL> class enterRoom { String who; }
RL> rule sayHello {
if ( fact enterRoom ) {
println("Hello " + enterRoom.who);
}
}
RL> assert(new enterRoom(who: "Bob"));
RL> run();
Hello Bob
RL>
Notes for ordering rule firing.
-
When you use the
returnaction, this changes the behavior for firing rules. Areturnaction in a rule pops the ruleset stack, so that execution continues with the activations on the agenda that are from the ruleset that is currently at the top of the ruleset stack.If rule execution was initiated with either the run or step functions, and a return action pops the last ruleset from the ruleset stack, then control returns to the caller of the run or step function.
If rule execution was initiated with the runUntilHalt function, then a return action does not pop the last ruleset from the ruleset stack. The last ruleset is popped with runUntilHalt when there are not any activations left. The Oracle Rules Engine then waits for more activations to appear. When they do, it places the last ruleset on the ruleset stack before resuming ruleset firing.
-
Rule priority is only applicable within rules in a given ruleset. Thus, the priority of rules in different rulesets are not comparable.
1.5 Using Effective Dates
By default, the value of the effective date is managed implicitly by the rules engine. In this case, when a run family of built-in functions is invoked, the effective date is updated to the current system date. This is done before any rules fire so that the new effective date is applied before rules begin to fire. In the case of runUntilHalt, this update occurs each time there is a transition from 0 rules on the agenda to > 0 rules on the agenda.
In Oracle Business Rules RL Language, the effective start and end dates and the active property are only applied to rules (and do not apply for rulesets). The effective start and end date properties of a rule can be specified in the rule.
For example,
rule myrule2 {
active = true;
effectiveDateForm = Rule.EDFORM_DATETIME:
effectiveStartDate = JavaDate.fromDateTimeString("2008-11-01");
effectiveEndDate = JavaDate.fromDateTimeString("2008-11-16");
if (fact Foo)
{
.
.
}
}
If you use the RuleSession Java API, you can access the effective start and end date.
Setting a property from RL Language requires a long expression or several statements.
For example, given a ruleset:
ruleset MyRules {
rule myRule { if fact foo { }}
}
To set the active property, use the following:
Rule r = getRuleSession().getRuleset("MyRules").getRule("myRule");
r.setActive(false);
1.6 Integrating RL Language Programs with Java Programs
This section describes integrating RL Language programs with Java programs. This section covers the following topics:
See Also:
"Working with Rules SDK Decision Point API" in the Oracle Business Rules User's Guide
1.6.1 Using Java Beans Asserted as Facts
Example 1-16 shows the Java source for a simple bean. Use the javac command to compile the bean, example.Person shown in Example 1-16 into a directory tree.
The following shows how an RL Language command-line can be started that can access this Java bean:
java -classpath $ORACLE_HOME/soa/modules/oracle.rules_11.1.1/rl.jar;BeanPath oracle.rules.rl.session.CommandLine -p "RL> "
Where BeanPath is the classpath component to any supplied Java Bean classes.
Example 1-16 Java Source for Person Bean Class
package example;
import java.util.*;
public class Person
{
private String firstName;
private String lastName;
private Set nicknames = new HashSet();
public Person(String first, String last, String[] nick) {
firstName = first; lastName = last;
for (int i = 0; i < nick.length; ++i)
nicknames.add(nick[i]);
}
public Person() {}
public String getFirstName() {return firstName;}
public void setFirstName(String first) {firstName = first;}
public String getLastName() {return lastName;}
public void setLastName(String last) {lastName = last;}
public Set getNicknames() {return nicknames;}
}
Example 1-17 shows how the RL Language command-line can execute an RL Language program that uses example.Person. The import statement, as in Java, allows a reference to the Person class using "Person" instead of "example.Person". Rules reference the Person bean class and its properties and methods. In order to create a Person fact you must assert a Java Person bean.
Example 1-17 uses the new operator to create an array of Person objects, named people. The people array is declared final so that reset does not create more people. The numPeople variable is not declared final so that reset re-invokes the assertPeople function and re-asserts the Person facts using the existing Person objects.
Example 1-17 Ruleset Using Person Bean Class
ruleset main
{
import example.Person;
import java.util.*;
rule hasNickNames
{
if (fact Person(nicknames: var nns) p && !nns.isEmpty())
{
// accessing properties as fields:
println(p.firstName + " " + p.lastName + " has nicknames:");
Iterator i = nns.iterator();
while (i.hasNext())
{
println(i.next());
}
}
}
rule noNickNames
{
if fact Person(nicknames: var nns) p && nns.isEmpty()
{
// accessing properties with getters:
println(p.getFirstName() + " " + p.getLastName() + " does not have nicknames");
}
}
final Person[] people = new Person[] {
new Person("Robert", "Smith", new String[] { "Bob", "Rob" }), // using constructor
new Person(firstName: "Joe", lastName: "Schmoe") // using attribute value pairs
};
function assertPeople(Person[] people) returns int
{
for (int i = 0; i < people.length; ++i) {
assert(people[i]);
}
return people.length;
}
int numPeople = assertPeople(people);
run();
}
Note the following when working with Java beans as facts:
-
The
factoperator can include a pattern that matches or retrieves the bean properties. The properties are defined by getter and setter methods in the bean class. -
The
newoperator can include a pattern that sets property values after invoking the default no-argument constructor, or can pass arguments to a user-defined constructor. -
Outside of the
factandnewoperators, the bean properties may be referenced or updated using getter and setter methods, or using the property name as if it were a field. -
If a bean has both a property and a field with the same name, then the field cannot be referenced in RL Language.
If Example 1-18 executes using the same RuleSession following the execution of Example 1-17, the output is identical to the Example 1-17 results (both person facts are reasserted).
Note:
The RL Language command-line interpreter internally creates a RuleSession when it starts (and when you use the clear command).
1.6.2 Using RuleSession Objects in Java Applications
Java programs can use the RuleSession interface to execute rulesets, invoke RL Language functions passing Java objects as arguments, and redirect RL Language watch and println output. Example 1-19 and Example 1-20 each contain a Java program fragment that uses a RuleSession that prints "hello world". Like many Java program fragments, these examples are also legal RL Language programs.
The RL Language environment provides multiple rule sessions. Each rule session can be used by multiple threads, but rules are fired by a single thread at a time.
Each rule RuleSession has its own copy of facts and rules. To create a fact from a Java Object, use a call such as:
rs.callFunctionWithArgument("assert", Object;);
To create a rule, a function, or an RL Language class, define a string containing a ruleset, and use the executeRuleset method.
Example 1-19 Using a RuleSession Object with callFunctionWithArgument
import oracle.rules.rl.*;
try {
RuleSession rs = new RuleSession();
rs.callFunctionWithArgument("println", "hello world");
} catch (RLException rle) {
System.out.println(rle);
}
Example 1-20 Using a RuleSession with ExecuteRuleset
import oracle.rules.rl.*;
try {
RuleSession rs = new RuleSession();
String rset =
"ruleset main {" +
" function myPrintln(String s) {" +
" println(s);" +
" }" +
"}";
rs.executeRuleset(rset);
rs.callFunctionWithArgument("myPrintln", "hello world");
} catch (RLException rle) {
System.out.println(rle);
}
1.7 Using Decision Tracing
Using Oracle Business Rules, a decision trace is a trace of rule engine execution that includes information on the state of the rule engine, including the state of facts when rule fire. The Oracle Business Rules rule engine constructs and returns a decision trace using JAXB generated Java classes generated from the decision trace XML schema.
1.7.1 Introduction to Rule Engine Level Decision Tracing
To provide a business analyst friendly presentation of a decision trace requires that the associated rule dictionary is available. Using the rule dictionary associated with a trace allows for a more flexible and efficient approach, as the trace output does not need to include all of the related dictionary content information in the trace.
The XML schema is in the file decisiontrace.xsd and it is part of the Jar file rl.jar as: oracle/rules/rl/trace/decisiontrace.xsd. The packages of interest are oracle.rules.rl.trace, oracle.rules.rl.extensions.trace, and oracle.rules.sdk2.decisiontrace. The Java classes packages generated from the decisiontrace XML schema are in the package oracle.rules.rl.trace and are included in the Javadoc. For more information, see Oracle Business Rules Java API Reference.
1.7.2 Using Rule Engine Level Decision Tracing
A decision trace is a set of XML elements showing rule engine events that occur during rule evaluation. The types of events that are added to a decision trace depend on the trace level selected, and can include:
-
Fact operations (assert, retract, modify)
-
Rules fired
-
Rule activations added or removed from the agenda
-
Ruleset stack changes
-
Rule compilation
-
Reset (which is needed for maintaining state for decision trace analysis)
Each trace contains information about a particular event. For example, a fact operation event entry consists of:
-
The operation type (assert, modify, retract)
-
The ID of the fact in working memory
-
Fact type name (fact classed in RL)
-
Runtime object type name
-
The fact object data, including the property name and value for zero or more fact properties
-
Name of rule, RL name, if the operation was the result of a rule action
-
Millisecond timestamp
In a fact operation event trace, the fact object content reflects the structure of the object as a Java Bean. If the bean properties are references to other beans the related bean content is included in the trace. The value of a bean property can be one of the following alternatives.
-
A string representation of the property. This is the case for primitive types and classes in the java.* and javax.* packages.
-
A nested bean object with its property values.
-
A fact ID. This occurs when the property value is an object which has itself been asserted as a fact. The data for the fact at the time of the trace can be retrieved from the RuleEngineState using the fact ID when analyzing the trace.
-
A collection of values accessed as a
Listin the trace. -
An array of values accessed as a
Listin the trace.
At runtime, to determine which alternative is included in the trace you can test for null; only the correct alternative has a non-null value.
Table 1-1 shows the RL functions that provide control over decision tracing in the rule engine and provide access to a decision trace.
Table 1-1 RL Decision Trace Functions
| Function | Description |
|---|---|
|
|
Returns the current trace and starts a new trace. |
|
|
Gets the current decision trace level. |
|
|
Returns the current limit on the number of events in a trace. |
|
|
Sets the decision trace level to the specified level. |
|
|
Sets the limit on the number of events in a trace. Default limit value is 10000. |
The decision trace level may be set by invoking the setDecisionTraceLevel function. You can also configure the initial trace level in a RuleSession or in a RuleSessionPool by including the RuleSession.CFG_DECISION_TRACE_LEVEL initialization parameter and specifying a level in the configuration Map passed to the RuleSession or RuleSessionPool constructor. This sets the decision trace level at the time a RuleSession is created.
You can invoke the setDecisionTraceLevel function on a RuleSession or a RuleSessionPool object after initialization. When you invoke reset(), this function returns the decision trace level to the configured value (if the level was changed during rule execution). Thus, the reset() function resets the decision trace limit to the value set during initialization of a RuleSession or a RuleSessionPool. In these cases, reset() restores the values established using the initialization parameters.
Note:
These reset() semantics for a RuleSession are only valid for a RuleSession initialized with either or both of the CFG_DECISION_TRACE_LIMIT and the CFG_DECISION_TRACE_LEVEL initialization parameters (or that is obtained from a RuleSessionPool when the pool is created with either or both of the CFG_DECISION_TRACE_LIMIT and the CFG_DECISION_TRACE_LEVEL initialization parameters.
The size of a trace is limited by limiting the number of entries in a decision trace. This necessary to avoid infinite rule fire loops, due to a possible bug in the rules, from creating a trace that consumes all available heap in the JVM. Set the trace limit with the setDecisionTraceLimit function. The limit may also be configured in a RuleSession (or RuleSessionPool) by including the RuleSession.CFG_DECISION_TRACE_LIMIT initialization parameter with the desired limit in the configuration Map passed to the RuleSession or RuleSessionPool constructor.
For rules applications that use runUntilHalt, it is the responsibility of the application to invoke getDecisionTrace before the trace limit is hit.
The decision trace provides structure to the trace data so that it can be manipulated programmatically. However, the trace by itself can be cumbersome to analyze. A trace analysis class (oracle.rules.rl.extensions.trace.TraceAnalysis) analyzes a decision trace and facilitates exploration of the trace. Use this class to construct the state of working memory, the agenda, and the ruleset stack from the trace.
The TraceAnalysis API supports the following:
-
Obtain a list of fact types that appear in the trace.
-
Obtain a list of names of the rules that fired in the trace.
-
Obtain a list of the last fact traces for each instance of a specific fact type.
-
Obtain the last fact trace for a specific fact identified by its fact ID.
-
Obtain all of the fact traces for a fact identified by its fact ID.
-
For a fact trace, if the fact trace was created by a rule action, get the rule trace that rule firing in which the action executed.
-
For a rule trace, get the list of fact traces for each fact that matched and resulted in the rule firing.
-
Get the next or previous trace. Exploration of the trace is typically not an iteration over the trace. For example, obtaining a rule trace from a fact trace is essentially jumping to that rule trace. The traces near the rule trace can be explored directly.
-
Obtain a list of rule traces for a rule identified by its name.
-
Obtain the rule engine state for a trace entry. The rule engine state reflects the state of the rule engine after the activity that generated the trace. This API enables inspecting the rule engine state at the time of each trace. This API is most useful with development level tracing. With production level tracing, only the facts in working memory can be tracked and they will not include any fact contents.
Example 1-21 shows a code sample that uses the decision trace analysis API.
Example 1-21 Decision Trace Analysis API Usage
DecisionTrace trace;
...
TraceAnalysis ta = new TraceAnalysis(trace);
// Get all of the last fact traces for a fact type.
List<FactTrace> factTraces = ta.getLastFactTraces("com.example.MyFactType");
// From a particular fact trace, how it was arrived at may be explored, first by
// obtaining the rule that asserted or modified the fact.
// From the FactRecord, the rule that resulted in the record can be obtained.
FactTrace factTrace = factTraces.get(0); // assumes there is one
RuleTrace ruleTrace = ta.whichRule(factTrace);
// The ruleTrace will be null if the fact operation was not part of a rule action.
System.out.print("Fact " + factTrace.getFactId() + ", a " + factTrace.getFactType() + " " + factRecord.getFactOp());
if (ruleTrace != null)
System.out.println(" by rule " + ruleTrace.getRuleName());
else
System.out.println("");
// The analysis can continue by obtaining the list of FactRecords that matched the rule and
// proceeding to analyze the trace back in time.
List<FactTrace> matchingFacts = ta.getRuleMatchedFacts(ruleTrace);
1.7.3 Decision Trace Samples for Production and Development Level Tracing
Example 1-22 shows a sample production level trace document.
Example 1-22 Sample Production Level Decision Trace
<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
<decision-trace xmlns="http://xmlns.oracle.com/rules/decisiontrace">
<trace-entries xsi:type="rule-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975549890</timestamp>
<rule-name>OrderDiscount.goldCustomer</rule-name>
<token-time>0</token-time>
<sequence-number>1</sequence-number>
</trace-entries>
<trace-entries xsi:type="rule-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975549893</timestamp>
<rule-name>OrderDiscount.goldCustomerDiscount</rule-name>
<token-time>0</token-time>
<sequence-number>2</sequence-number>
</trace-entries>
<trace-entries xsi:type="rule-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975549894</timestamp>
<rule-name>OrderDiscount.applyDiscount</rule-name>
<token-time>0</token-time>
<sequence-number>3</sequence-number>
</trace-entries>
</decision-trace>
Example 1-23 Sample Development Level DecisionTrace
<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
<decision-trace xmlns="http://xmlns.oracle.com/rules/decisiontrace">
<trace-entries xsi:type="fact-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975491008</timestamp>
<fact-id>1</fact-id>
<operation>assert</operation>
<fact-type>com.example.Customer</fact-type>
<object-type>com.example.Customer</object-type>
<fact-object>
<properties>
<name>YTDOrderAmount</name>
<value>
<string>2000.0</string>
</value>
</properties>
<properties>
<name>level</name>
<value>
<string>null</string>
</value>
</properties>
<properties>
<name>name</name>
<value>
<string>OneLtd</string>
</value>
</properties>
<properties>
<name>pastDue</name>
<value>
<string>false</string>
</value>
</properties>
</fact-object>
</trace-entries>
<trace-entries xsi:type="activation-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975491024</timestamp>
<rule-name>OrderDiscount.goldCustomer</rule-name>
<token-time>2</token-time>
<fact-ids>1</fact-ids>
<operation>add</operation>
</trace-entries>
<trace-entries xsi:type="fact-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975491025</timestamp>
<fact-id>2</fact-id>
<operation>assert</operation>
<fact-type>com.example.Order</fact-type>
<object-type>com.example.Order</object-type>
<fact-object>
<properties>
<name>customerName</name>
<value>
<string>OneLtd</string>
</value>
</properties>
<properties>
<name>discount</name>
<value>
<string>0.0</string>
</value>
</properties>
<properties>
<name>grossAmount</name>
<value>
<string>400.0</string>
</value>
</properties>
<properties>
<name>netAmount</name>
<value>
<string>0.0</string>
</value>
</properties>
<properties>
<name>number</name>
<value>
<string>1001</string>
</value>
</properties>
</fact-object>
</trace-entries>
<trace-entries xsi:type="activation-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975491035</timestamp>
<rule-name>OrderDiscount.goldCustomerDiscount</rule-name>
<token-time>5</token-time>
<fact-ids>2</fact-ids>
<fact-ids>1</fact-ids>
<operation>add</operation>
</trace-entries>
<trace-entries xsi:type="rule-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975491036</timestamp>
<rule-name>OrderDiscount.goldCustomerDiscount</rule-name>
<token-time>5</token-time>
<fact-ids>2</fact-ids>
<fact-ids>1</fact-ids>
<sequence-number>2</sequence-number>
</trace-entries>
...
<trace-entries xsi:type="rule-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975491036</timestamp>
<rule-name>OrderDiscount.applyDiscount</rule-name>
<token-time>7</token-time>
<fact-ids>2</fact-ids>
<sequence-number>3</sequence-number>
</trace-entries>
...
<trace-entries xsi:type="ruleset-stack-trace" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<timestamp>1248975491037</timestamp>
<operation>pop</operation>
<ruleset-name>OrderDiscount</ruleset-name>
</trace-entries>
</decision-trace>
1.8 Building a Coin Counter Rules Program
This section shows a sample that uses RL Language to solve a puzzle:
How many ways can 50 coins add up to $1.50?
The rules program that solves this puzzle illustrates an important point for rule-based programming; knowledge representation, that is, the fact classes that you select, can be the key design issue. It is often worthwhile to write procedural code to shape your data into a convenient format for the rules to match and process.
To use this example, first copy the RL Language program shown in Example 1-25 to a file named coins.rl. You can include this from the RL Language command-line using the include command. Before you include the coins program, use the clear; command to erase everything in the current rule session, as follows:
RL> clear; RL> include file:coins.rl; RL>
Example 1-24 shows the debugging functions that show the count coins sample facts, activations, and rules for the coin counter. All facts are asserted, and activations for all solutions are placed on the agenda. Notice that the facts are matched to the rule condition as they are generated by populate_facts, and that find_solution prints the matches.
Example 1-24 Using Debugging Functions with Coins Example
RL> watchFacts(); RL> watchActivations(); RL> watchRules(); RL> reset(); RL> showActivations(); RL> run(); The rule is fired for each activation, printing out the solutions RL>
In Example 1-25, the keyword final in front of a global variable definition such as coinCount and totalAmount marks that variable as a constant, as in Java. You can reference constants in rule conditions, but you cannot reference variables in rule conditions.
In RL Language, you must initialize all variables. The initialization expression for a final variable is evaluated once when the variable is defined. The initialization expression for a non-final variable is evaluated when the variable is defined, and again each time the reset function is called. Because the reset function retracts all facts from working memory, it is good practice to assert initial facts in a global variable initialization expression, so that the facts are re-asserted when reset is called.
Example 1-25 illustrates how to use global variable initialization expressions. The initialized global variable is initialized with the populate_facts function. This function is re-executed whenever reset is called. The populate_facts function has a while loop nested within a for loop. The for loop iterates over an array of coin denomination Strings. For each denomination, the while loop asserts a fact that expresses a count and a total that does not exceed the total amount of $1.50. For example, for half dollars:
coin(denomination "half-dollar", count:0, amount:0) coin(denomination "half-dollar", count:1, amount:50) coin(denomination "half-dollar", count:2, amount:100) coin(denomination "half-dollar", count:3, amount:150)
With such facts in working memory, the rule find_solution matches against each denomination with a condition that requires that the counts sum to coinCount and the amounts sum to totalAmt. The run function fires the find_solutions activations.
Example 1-25 Count Coins Program Source
final int coinCount = 50;
final int totalAmt = 150;
final String[] denominations = new String[]
{"half-dollar" , "quarter", "dime", "nickel", "penny" };
class coin {
String denomination;
int count;
int amount;
}
function populate_facts() returns boolean
{
for (int i = 0; i < denominations.length; ++i) {
String denom = denominations[i];
int count = 0;
int total = 0;
int amount = 0;
if (denom == "half-dollar" ) { amount = 50; }
else if (denom == "quarter" ) { amount = 25; }
else if (denom == "dime" ) { amount = 10; }
else if (denom == "nickel" ) { amount = 5; }
else { amount = 1; }
while (total <= totalAmt && count <= coinCount)
{
assert(new coin(denomination: denom,
count : count,
amount : total));
total += amount;
count ++;
}
}
return true;
}
boolean initialized = populate_facts();
rule find_solution
{
if(fact coin(denomination: "penny") p
&& fact coin(denomination: "nickel") n
&& fact coin(denomination: "dime") d
&& fact coin(denomination: "quarter") q
&& fact coin(denomination: "half-dollar") h
&& p.count + n.count + d.count + q.count + h.count == coinCount
&& p.amount + n.amount + d.amount + q.amount + h.amount == totalAmt)
{
println("Solution:"
+ " pennies=" + p.count
+ " nickels=" + n.count
+ " dimes=" + d.count
+ " quarters=" + q.count
+ " half-dollars=" + h.count
);
}
}
run();