Most web and enterprise Java applications can be split into three parts: a front end to talk to the user, a service layer to talk to back-end systems such as databases, and business logic in between. While it is now common practice to use frameworks for both front- and back-end requirements (e.g., Struts, Cocoon, Spring, Hibernate, JDO, and Entity Beans), there is no standard way of structuring business logic. Frameworks like EJB and Spring do this at a high level, but don't help us in organizing our code. Wouldn't it would be great if we could replace messy, tangled if...then statements with a framework that gave us the same benefits of configurability, readability, and reuse that we already enjoy in other areas? This article suggests using the Drools rules engine as a framework to solve the problem.
The sample code below gives a sample of the problem we're trying to avoid. It shows some business logic in a typical Java application.
if ((user.isMemberOf(AdministratorGroup)
&& user.isMemberOf(teleworkerGroup))
|| user.isSuperUser(){
// more checks for specific cases
if((expenseRequest.code().equals("B203")
||(expenseRequest.code().equals("A903")
&&(totalExpenses<200)
&&(bossSignOff> totalExpenses))
&&(deptBudget.notExceeded)) {
//issue payments
} else if {
//check lots of other conditions
}
} else {
// even more business logic
}
We've all come across similar (or even more complex) business logic. While this has been the standard way of implementing business logic in Java, there are many problems with it.
- What if the business users come up with another form ("C987") that needs to be added to the already hard-to-understand code? Would you want to be the person to maintain it, once all of the original programmers had moved on?
- How do we check that these rules are correct? It's hard enough for technical people--never mind commercial folks--to review. Do we have any methodical way of testing this business logic?
- Many applications have similar business rules--if one of the rules change, can we be sure that it is changed consistently across all systems? If a new application uses some of these rules, but also adds some new ones, do we need to rewrite all of the logic from scratch?
- Is the business logic easily configurable, not so firmly tied to Java code that we need to recompile/redeploy every time that a small change is made?
- What if other (scripting) languages want to leverage the existing investment in business rule logic?
J2EE/EJB and "inversion of control" frameworks (such as Spring, Pico, and Avalon) give us the ability to organize our code at a high level. While they are very good at providing reusability, configuration, and security, none of them would replace the "spaghetti code" in the above example. Ideally, whatever framework we choose will be compatible with not only J2EE applications, but also "normal" Java (J2SE--Standard Edition) programs, and most of the widely used presentation and persistence frameworks. Such a framework should allow us to do the following:
- Business users should be able to easily read and verify the business logic.
- Business rules should be reusable and configurable across applications.
- The framework should be scalable and performant under heavy load.
- It should be as easy to use for Java programmers as existing front-end (Struts, Spring) and back-end (object-relational mapping) frameworks.
An additional problem is that while there are only so many ways to organize web pages and database access, business logic tends to differ widely between applications. Our framework should be able to cope with this and still promote code reuse. Ideally, our application would be "frameworks all the way down." By using frameworks in this way, we can a large amount of our application "out of the box," allowing us to write only the parts that add value for the customer.
Rule Engines to the Rescue
How are we going to solve this problem? One solution that is gaining traction is to use a rule engine. Rule engines are frameworks for organizing business logic that allow the developer to concentrate on things that are known to be true, rather than the low-level mechanics of making decisions.
Often, business users are more comfortable with expressing things that they know to be true, than to express things in an if...then format. Examples of things that you might hear from a business expert are:
- "FORM 10A is used for expense claims over 200 Euro."
- "We only trade shares in quantities of 10,000 or more."
- "Purchases over €10m need the approval of a company director."
By focusing on what we know to be true, rather than the mechanics of how to express it in Java code, the above statements are clearer than our previous code sample. Still, clear as they may be, we still need a mechanism to apply these rules to the facts that we know and get a decision. Such a mechanism is a rule engine.
Rule Engines in Java
JSR 94, the javax.rules API, sets a common standard for interacting with rule engines, much as JDBC allows us to interact with varying databases. What JSR-94 does not specify is how the actual rules are written, leaving plenty of choice among the most widely used Java rule engines:
- Jess is perhaps the most mature Java rule engine, with good tool support (including Eclipse plugins) and documentation. However it is a commercial product, and it writes its rules in a Prolog-style notation, which can be intimidating for many Java programmers.
- Jena is an open source framework, originally from HP. While it has a rules engine, and is especially strong for those interested in the Semantic Web, it is not fully JSR-94-compliant.
- Drools is a JSR-94-complaint rules engine, and is fully open source under an "Apache-style" license. Not only does it express rules in familiar Java and XML syntax, it has a strong user and developer community. For the examples in this article, we'll be using Drools, as it has the easiest to use Java-like syntax and it has the most open license.
Starting a Java Application using Drools
Imagine this scenario: minutes after reading this article, your boss asks your to prototype a stock trading application. As the business users still haven't fully defined the business logic, you think it a good idea to implement it using a rules engine. The final system will be accessible over an intranet and will need to communicate with back-end database and messaging systems. To get started, download the Drools framework (with dependencies). Create a new project in your favorite IDE and make sure all of the .jars are referenced in it, as per Figure 1. This screenshot is Eclipse-based, but the setup will be similar for other IDEs.
Figure 1. Libraries needed to run Drools
Due to the huge potential losses if our stock trading system went amok, it's vital that we have some sort of simulator to put our system through its paces. Such a simulator also gives you confidence that the decisions made by the system are those that are intended, even after rule changes are made. We'll borrow some tools from the Agile toolbox and use JUnit as a framework for our simulations.
The first code we write is the JUnit Test/simulator, as per the following listing. Even if we can't test every combination of values likely to be input into our application, some tests are better than none at all. In this example, all of our files and classes (including unit tests) are in one folder/package, but in reality, you would implement a proper package and folder structure. We'd also use Log4j instead of the System.out calls in the sample code.
import junit.framework.TestCase;
/*
* JUnit test for the business rules in the
* application.
*
* This also acts a 'simulator' for the business
* rules - allowing us to specify the inputs,
* examine the outputs and see if they match our
* expectations before letting the code loose in
* the real world.
*/
public class BusinessRuleTest extends TestCase {
/**
* Tests the purchase of a stock
*/
public void testStockBuy() throws Exception{
//Create a Stock with simulated values
StockOffer testOffer = new StockOffer();
testOffer.setStockName("MEGACORP");
testOffer.setStockPrice(22);
testOffer.setStockQuantity(1000);
//Run the rules on it
BusinessLayer.evaluateStockPurchase(testOffer);
//Is it what we expected?
assertTrue(
testOffer.getRecommendPurchase()!=null);
assertTrue("YES".equals(
testOffer.getRecommendPurchase()));
}
}
This is a basic JUnit test, as we know that our (very simple!) system should buy all stocks with a price of less than 100 Euro. Obviously, this won't compile without our data holding class (StockOffer.java) and our business layer class (BusinessLayer.java). These are provided in the following listings.
/**
* Facade for the Business Logic in our example.
*
* In this simple example, all our business logic
* is contained in this class but in reality it
* would delegate to other classes as required.
*/
public class BusinessLayer {
/**
* Evaluate whether or not it is a good idea
* to purchase this stock.
* @param stockToBuy
* @return true if the recommendation is to buy
* the stock, false if otherwise
*/
public static void evaluateStockPurchase
(StockOffer stockToBuy){
return false;
}
}
The StockOffer class looks like this:
/**
* Simple JavaBean to hold StockOffer values.
* A 'Stock offer' is an offer (from somebody else)
* to sell us a Stock (or Company share).
*/
public class StockOffer {
//constants
public final static String YES="YES";
public final static String NO="NO";
//Internal Variables
private String stockName =null;
private int stockPrice=0;
private int stockQuantity=0;
private String recommendPurchase = null;
/**
* @return Returns the stockName.
*/
public String getStockName() {
return stockName;
}
/**
* @param stockName The stockName to set.
*/
public void setStockName(String stockName) {
this.stockName = stockName;
}
/**
* @return Returns the stockPrice.
*/
public int getStockPrice() {
return stockPrice;
}
/**
* @param stockPrice The stockPrice to set.
*/
public void setStockPrice(int stockPrice) {
this.stockPrice = stockPrice;
}
/**
* @return Returns the stockQuantity.
*/
public int getStockQuantity() {
return stockQuantity;
}
/**
* @param stockQuantity to set.
*/
public void setStockQuantity(int stockQuantity){
this.stockQuantity = stockQuantity;
}
/**
* @return Returns the recommendPurchase.
*/
public String getRecommendPurchase() {
return recommendPurchase;
}
}
We run BusinessRuleTest through the JUnit extension of our favorite IDE. If you're not familiar with JUnit, more information can be found at JUnit's website. Not surprisingly, our test fails at the second assertion, shown in Figure 2, as we don't (yet) have the appropriate business logic in place. This is reassuring, as it shows that our simulator/unit tests are highlighting the problems that they should.
Figure 2. JUnit test results
Writing the Business Logic using Rules
At this point, we need to write some business logic that says, "If the stock price is less than 100 Euro, then we should buy it." To do this, we will modify BusinessLayer.java to read:
import java.io.IOException;
import org.drools.DroolsException;
import org.drools.RuleBase;
import org.drools.WorkingMemory;
import org.drools.event.DebugWorkingMemoryEventListener;
import org.drools.io.RuleBaseLoader;
import org.xml.sax.SAXException;
/**
* Facade for the Business Logic in our example.
*
* In this simple example, all our business logic
* is contained in this class but in reality it
* would delegate to other classes as required.
* @author default
*/
public class BusinessLayer {
//Name of the file containing the rules
private static final String BUSINESS_RULE_FILE=
"BusinessRules.drl";
//Internal handle to rule base
private static RuleBase businessRules = null;
/**
* Load the business rules if we have not
* already done so.
* @throws Exception - normally we try to
* recover from these
*/
private static void loadRules()
throws Exception{
if (businessRules==null){
businessRules = RuleBaseLoader.loadFromUrl(
BusinessLayer.class.getResource(
BUSINESS_RULE_FILE ) );
}
}
/**
* Evaluate whether or not to purchase stock.
* @param stockToBuy
* @return true if the recommendation is to buy
* @throws Exception
*/
public static void evaluateStockPurchase
(StockOffer stockToBuy) throws Exception{
//Ensure that the business rules are loaded
loadRules();
//Some logging of what is going on
System.out.println( "FIRE RULES" );
System.out.println( "----------" );
//Clear any state from previous runs
WorkingMemory workingMemory
= businessRules.newWorkingMemory();
//Small ruleset, OK to add a debug listener
workingMemory.addEventListener(
new DebugWorkingMemoryEventListener());
//Let the rule engine know about the facts
workingMemory.assertObject(stockToBuy);
//Let the rule engine do its stuff!!
workingMemory.fireAllRules();
}
}
This class now has some important methods:
loadRules(), which loads the rules from the BusinessRules.drl file.
- An updated
evaluateStockPurchase(), which evaluates these business rules. Some points to note about this method are:
- We can reuse the same
RuleSet over and over (as business rules in memory are stateless).
- We use a new
WorkingMemory for every evaluation, as this is our knowledge of what we know to be true at this time. We use assertObject() to place known facts (as Java Objects) into this memory.
- Drools has an event listener model, to allow us to "see" what is going on within the event model. Here we use it to print debug information.
- The
fireAllRules() method on the working memory class causes the rules to be evaluated and updated (in this case, stock offer).
Before we can run the example again, we need to create our BusinessRules.drl file, as follows:
<?xml version="1.0"?>
<rule-set name="BusinessRulesSample"
xmlns="http://drools.org/rules"
xmlns:java="http://drools.org/semantics/java"
xmlns:xs
="http://www.w3.org/2001/XMLSchema-instance"
xs:schemaLocation
="http://drools.org/rules rules.xsd
http://drools.org/semantics/java java.xsd">
<!-- Import the Java Objects that we refer
to in our rules -->
<java:import>
java.lang.Object
</java:import>
<java:import>
java.lang.String
</java:import>
<java:import>
net.firstpartners.rp.StockOffer
</java:import>
<!-- A Java (Utility) function we reference
in our rules-->
<java:functions>
public void printStock(
net.firstpartners.rp.StockOffer stock)
{
System.out.println("Name:"
+stock.getStockName()
+" Price: "+stock.getStockPrice()
+" BUY:"
+stock.getRecommendPurchase());
}
</java:functions>
<rule-set>
<!-- Ensure stock price is not too high-->
<rule name="Stock Price Low Enough">
<!-- Params to pass to business rule -->
<parameter identifier="stockOffer">
<class>StockOffer</class>
</parameter>
<!-- Conditions or 'Left Hand Side'
(LHS) that must be met for
business rule to fire -->
<!-- note markup -->
<java:condition>
stockOffer.getRecommendPurchase() == null
</java:condition>
<java:condition>
stockOffer.getStockPrice() < 100
</java:condition>
<!-- What happens when the business
rule is activated -->
<java:consequence>
stockOffer.setRecommendPurchase(
StockOffer.YES);
printStock(stockOffer);
</java:consequence>
</rule>
</rule-set>
This rules file has several interesting parts:
- Just after the XML-Schema definitions come the Java objects we reference in our rules. These objects can come from any Java library as required.
- Next comes our functions, which can incorporate standard Java code. In this case, we incorporate a logging function to help us see what is going on.
- After that comes our rule set, consisting of one or more rules.
- Each rule can take parameters (the
StockOffer class), one or more conditions that need to be fulfilled, and a consequence that is carried out if and when the conditions are met.
Having modified and compiled our code, we run the JUnit test simulations again. This time, the business rules are called, our logic evaluates correctly, and our tests pass, as seen in Figure 3. Congratulations--you've just built your first rule-based application!
Figure 3. Successful JUnit test
Smarter Rules
Fresh from building the application, you demonstrate the prototype above to the business users, and they remember a few more rules that they forgot to mention earlier. One of the new rules is that we shouldn't trade stocks where the quantity is a negative number (<0). "No problem," you say, and return to your desk, secure in the knowledge that you can quickly evolve your system.
The first thing you do is to update your simulator, and add the following code to BusinessRuleTest.java:
/**
* Tests the purchase of a stock
* makes sure the system will not accept
* negative numbers.
*/
public void testNegativeStockBuy()
throws Exception{
//Create a Stock with our simulated values
StockOffer testOffer = new StockOffer();
testOffer.setStockName("MEGACORP");
testOffer.setStockPrice(-22);
testOffer.setStockQuantity(1000);
//Run the rules on it
BusinessLayer
.evaluateStockPurchase(testOffer);
//Is it what we expected?
assertTrue("NO".equals(
testOffer.getRecommendPurchase()));
}
This tests for the new rule described by the business users. If we run this JUnit test, our new test fails, as expected. We need to add a new rule to our .drl file, as follows.
<!-- Ensure that negative prices
are not accepted-->
<rule name="Stock Price Not Negative">
<!-- Parameters we can pass into
the business rule -->
<parameter identifier="stockOffer">
<class>StockOffer</class>
</parameter>
<!-- Conditions or 'Left Hand Side' (LHS)
that must be met for rule to fire -->
<java:condition>
stockOffer.getStockPrice() < 0
</java:condition>
<!-- What happens when the business rule
is activated -->
<java:consequence>
stockOffer.setRecommendPurchase(
StockOffer.NO);
printStock(stockOffer);
</java:consequence>
</rule>
This rule is similar in format to the previous one, expect that our <java:condition> is different (testing for negative numbers) and the <java:consequence> sets the recommend purchase to No. We run our unit tests/simulator again, and this time the test passes.
At this point, if you're used to procedural programming (like most Java programmers), you may be scratching your head: here we have a file containing two separate business rules, yet we haven't told the rule engine which is more important. However, our stock price (of -22) satisfies both rules (i.e., it is less than 0 and it is less than 100). Despite this, we get the correct result, even if we swap the order of the rules around. How does this work?
The extract of the console output below helps us to see what is going on. We see that both rules are firing (the [activationfired] line), and that the Recommend Buy is first set to Yes and then to No. How does Drools know to fire these rules in the correct order? If you look at the Stock Price Low Enough rule, you will see that one of the conditions is that recommendPurchase() is null. This is enough for the Drools rule engine to decide that the Stock Price Low Enough rule should be fired before the Stock Price Not Negative rule. This process is called conflict resolution.
FIRE RULES
----------
[ConditionTested: rule=Stock Price Not Negative;
condition=[Condition: stockOffer.getStockPrice()
< 0]; passed=true; tuple={[]}]
[ActivationCreated: rule=Stock Price Not Negative;
tuple={[]}]
[ObjectAsserted: handle=[fid:2];
object=net.firstpartners.rp.StockOffer@16546ef]
[ActivationFired: rule=Stock Price Low Enough;
tuple={[]}]
[ActivationFired: rule=Stock Price Not Negative;
tuple={[]}]
Name:MEGACORP Price: -22 BUY:YES
Name:MEGACORP Price: -22 BUY:NO
If you're a procedural programmer, no matter how clever you think this is, you still may not trust it completely. That is why we have our unit tests/simulator: "hard" JUnit tests (using normal Java code) ensure that the rule engine makes its decisions along the lines we want it to. (And doesn't spend billions on worthless stock!) At the same time, the power and the flexibility of our rule engine allows us to quickly develop the business logic.
Later on, we will see more sophisticated forms of conflict resolution.
Conflict Resolution
Now the folks on the business side are really impressed and are starting to think through the possible options. They've come across a problem with stocks of XYZ Corp and have decided to implement a new rule: Only buy stocks of XYZ Corp if they are less than 10 Euro.
As before, you add the test to our simulator and include the new business rule in our rules file, as per the following listings. First, we add a new method to BusinessRuleTest.java:
/**
* Makes sure the system will buy stocks
* of XYZ corp only if it really cheap
*/
public void testXYZStockBuy() throws Exception{
//Create a Stock with our simulated values
StockOffer testOfferLow = new StockOffer();
StockOffer testOfferHigh = new StockOffer();
testOfferLow.setStockName("XYZ");
testOfferLow.setStockPrice(9);
testOfferLow.setStockQuantity(1000);
testOfferHigh.setStockName("XYZ");
testOfferHigh.setStockPrice(11);
testOfferHigh.setStockQuantity(1000);
//Run the rules on it and test
BusinessLayer.evaluateStockPurchase(
testOfferLow);
assertTrue("YES".equals(
testOfferLow.getRecommendPurchase()));
BusinessLayer.evaluateStockPurchase(
testOfferHigh);
assertTrue("NO".equals(
testOfferHigh.getRecommendPurchase()));
}
Next, we need a new <rule> in BusinessRules.drl:
<rule name="XYZCorp" salience="-1">
<!-- Parameters we pass to rule -->
<parameter identifier="stockOffer">
<class>StockOffer</class>
</parameter>
<java:condition>
stockOffer.getStockName().equals("XYZ")
</java:condition>
<java:condition>
stockOffer.getRecommendPurchase() == null
</java:condition>
<java:condition>
stockOffer.getStockPrice() > 10
</java:condition>
<!-- What happens when the business
rule is activated -->
<java:consequence>
stockOffer.setRecommendPurchase(
StockOffer.NO);
printStock(stockOffer);
</java:consequence>
</rule>
Note that in the business rules file, after the rule name, we set our salience to -1 (i.e., the lowest priority of all of the rules we have specified so far). Most of the rules in our system conflict, meaning Drools must make some decision on the order in which to fire rules, given that the conditions for all of the rules will be met. The default way of deciding is:
Salience: A value we assign, as per the above listing.
Recency: How many times we have used a rule.
Complexity: Specific rules with more complicated values fire first.
LoadOrder: The order in which rules are loaded.
If we did not specify the saliency of our rule in this example, what would happen is:
- The XYZ Corp rule ("Don't buy XYZ if the price is more than 10 Euro") would fire first (the status of the
Recommend Buy flag would be set to No).
- Then the more general rule ("Buy all stock under 100") fires, setting the
Recommended Buy flag to yes.
This would give a result that we don't want. However, since our example does set the saliency factor, the test and our business rules work as expected.
While most of the time, writing clear rules and setting the saliency will give enough information to Drools for it to choose the proper order in which to fire rules, sometimes we want to change the entire manner in which rule conflicts are resolved. An example of how to change this is given below, where we tell the rule engine to fire the simplest rules first. A word of warning: be careful when changing conflict resolution, as it can fundamentally change the behavior of the rule engine--a lot of problems can be solved first with clear and well-written rules.
//Generate our list of conflict resolvers
ConflictResolver[] conflictResolvers =
new ConflictResolver[] {
SalienceConflictResolver.getInstance(),
RecencyConflictResolver.getInstance(),
SimplicityConflictResolver.getInstance(),
LoadOrderConflictResolver.getInstance()
};
//Wrap this up into one composite resolver
CompositeConflictResolver resolver =
new CompositeConflictResolver(
conflictResolvers);
//Specify this resolver when we load the rules
businessRules = RuleBaseLoader.loadFromUrl(
BusinessLayer.class.getResource(
BUSINESS_RULE_FILE),resolver);
For our simple application, driven by JUnit tests, we don't need to alter the way the Drools resolves rule conflicts. It is useful to know how conflict resolution works, especially when your application grows to meet more complex and demanding requirements.
Conclusion
This article demonstrated a problem that most programmers have had to face: how to put some order on the complexity of business logic. We demonstrated a simple application using Drools as a solution and introduced the notion of rule-based programming, including how these rules are resolved at runtime. Later on, a follow-up article will take these foundations and show how to use them in an enterprise Java application.
Resources
Paul Browne , based in Dublin, Ireland, has been consulting in enterprise Java with FirstPartners.net for almost seven years.