Sunday, February 23, 2014

JBoss monitoring with Twiddle


This short article introduces twiddle, a command line utility to administer a JBoss instance and shows how you can monitor a given instance of JBoss with it through a shell script.

JBoss provides a simple command line tool that allows for interaction with a remote JMX server instance. This tool is called twiddle (for twiddling bits via JMX) and is located in the bin directory of the distribution. Twiddle is a command execution tool, not a general command shell.

The 2 following links show basic usage of twiddle:


All available metrics can be found using the jmx-console deployed by the JBoss instance.

Next, here is a simple shell script that automate the collect of JBoss metrics (free & total memories, active threads count and active sessions count):

#!/bin/sh

# path to the log file (the directory /var/log/jboss must exist. Adapt this path to reflect your needs)
LOGFILE=/var/log/jboss/jbossmon-`hostname`-`date +"%Y%m%d"`.log
echo Logging JBoss metrics to $LOGFILE

# function that uses twiddle to get jboss metrics, given:
#  a first parameter that is the path to the MBean
#  a second parameter that is the name of the attribute of the MBean
# Notice that you may need to adapt the call to twiddle to use authentication or a dedicated port
# e.g: ./twiddle.sh -u user -p password -s jnp://localhost:9099
# Also update the path to the twiddle.sh, if this script is not saved in the same directory, i.e the bin directory of the JBoss distribution
function getJbossMetric() {
 ./twiddle.sh get $1 $2 | cut -d'=' -f2
}

# collects metrics every 5 mn forever (naive approach)
# Notice that a better way would be to replace the loop by a scheduler like cron
until false 
do
 # collects some metrics within JBoss using twiddle
 FREEMEM=$(getJbossMetric jboss.system:type=ServerInfo FreeMemory)
 TOTALMEM=$(getJbossMetric jboss.system:type=ServerInfo TotalMemory)
 THREADS=$(getJbossMetric jboss.system:type=ServerInfo ActiveThreadCount)
 # NB: you need to update the path property to reflect the context path of your web application
 SESSIONS=$(getJbossMetric jboss.web:host=localhost,path=/contextPathOfMyWebApp,type=Manager activeSessions)

 echo "-------------------------------------"
 echo "Free memory:    $FREEMEM"
 echo "Total memory:   $TOTALMEM"
 echo "Active Threads: $THREADS"
 echo "Sessions:       $SESSIONS"
 echo "-------------------------------------"
 
 # append metrics to the log file in a CSV manner
 TIMESTAMP=`date +"%d/%m/%Y %T"`
 echo "$TIMESTAMP;$TOTALMEM;$FREEMEM;$SESSIONS;$THREADS" >> $LOGFILE

 # waits for 300s (5mn)
 sleep 300
done
I hope this will be useful to those who want to monitor JBoss in a simple way.

Monday, July 29, 2013

Compute Java Object Memory Footprint at runtime with JAMM (Java Agent for Memory Measurements)

This short article shows how to measure java object memory size at runtime with JAMM, a java agent dedicated to measure actual object memory use including JVM overhead.

JAMM uses the Instrumentation implementation provided by the JVM to compute memory usage of a given object by calling the getObjectSize(..) method.

It is quite simple to use, as explained by the author:

MemoryMeter meter = new MemoryMeter();
meter.measure(object);
meter.measureDeep(object);
meter.countChildren(object);

The only constraint is to attach JAMM java agent to the JVM before using it to measure memory usage, by starting the JVM with the -javaagent option pointing to the JAMM jar.

So, we are going to write a JUnit that shows how to use JAMM, but before we need to setup maven to achieve this:
    
        
            junit
            junit
            4.11
            test
        
    
        
        
            com.github.stephenc
            jamm
            0.2.5
            test
        
            
    
    
    
        
            
            
            
                org.apache.maven.plugins
                maven-dependency-plugin
                2.8
                
                    
                        copy-dependencies
                        generate-test-resources
                        
                            copy
                        
                        
                            
                                
                                    com.github.stephenc
                                    jamm
                                    0.2.5
                                    jar
                                    ${project.build.directory}
                                    jamm.jar
                                
                            
                        
                    
                
            
            
            
            
                org.apache.maven.plugins
                maven-surefire-plugin
                2.14
                
                    -javaagent:${project.build.directory}/jamm.jar
                
            
            
        
    

Next, write the following JUnit that explores the JAMM features:
package org.javabenchmark.memory;

import java.util.ArrayList;
import java.util.List;
import org.github.jamm.MemoryMeter;
import org.junit.Test;

public class MemoryMeterTest {

    private MemoryMeter meter = new MemoryMeter();
    
    @Test
    public void shouldMeasureMemoryUsage() {

        String st1 = "This is the string #1";
        measure(st1);
        
        String st2 = "This is the string #2 and it has more chars.";
        measure(st2);
        
        List aList = new ArrayList(0);
        measure(aList);
        
        aList.add(st1);
        measure(aList);
        
        aList.add(st2);
        measure(aList);
        
    }

    private void measure(Object anObject) {
        
        System.out.println("-----------------------------------");
        System.out.printf("size: %d bytes\n", meter.measure(anObject));
        System.out.printf("retained size: %d bytes\n", meter.measureDeep(anObject));
        System.out.printf("inner object count: %d\n", meter.countChildren(anObject));
    }
}

Running the test produces the following output on my computer:
-----------------------------------
size: 24 bytes
retained size: 88 bytes
inner object count: 2
-----------------------------------
size: 24 bytes
retained size: 128 bytes
inner object count: 2
-----------------------------------
size: 24 bytes
retained size: 40 bytes
inner object count: 2
-----------------------------------
size: 24 bytes
retained size: 136 bytes
inner object count: 4
-----------------------------------
size: 24 bytes
retained size: 264 bytes
inner object count: 6
To conclude, you can see how it is easy to monitor the memory usage of your objects. It is very handy when dealing with huge collections, or when using caches such as the ones provided by Guava or EHCache. That way you can setup trigger that alert when memory consumption is excessive.


UPDATE: 2013-07-30

To answer to the rxin's comment, i did a quick (trivial) test with a list of 1,000,000 random strings, that represents 116 Mo in memory.

For the test, i sized the heap with the following JVM options: -Xms256m -Xmx256m.

Below is the memory usage during the test:


From my point of view, the memory overhead introduced by JAMM is negligible in that simple test case, but notice that the measureDeep() method takes time (reflection is slow).


 
  

Sunday, May 19, 2013

Monitoring Memory Usage inside a Java Web Application with JSF, PrimeFaces and MemoryMXBean

This article explains how to monitor memory usage in your web application by requesting the MemoryMXBean and exposing collected metrics within a Primefaces LineChart component in a JSF page.

Principle

The monitoring of the memory usage involves:

The overall design is illustrated below:



  1. The LineChart component calls every minute the MonitorController instance.
  2. The MonitorController instance requests the MemoryMXBean instance to get the current memory usage.
  3. The MonitorController instance updates a CartesianChartModel instance by adding a memory usage snapshot.
  4. The CartesianChartModel instance is returned to the LineChart component that renders it inside a JSF page.

PrimeFaces LineChart

To enable PrimeFaces components into your web application, update your maven pom.xml with the following repository and dependency:
    
        
            primefaces-maven-repo
            PrimeFaces Maven Repository
            http://repository.primefaces.org
            default
        
    
        
            org.primefaces
            primefaces
            3.4.2
        
Next, inside a JSF page, add the following code to display the LineChart and make it polling the MonitorController every minute:
<html xmlns="http://www.w3.org/1999/xhtml"
      xmlns:h="http://java.sun.com/jsf/html"
      xmlns:f="http://java.sun.com/jsf/core"
      xmlns:p="http://primefaces.org/ui"
      xmlns:c="http://java.sun.com/jsp/jstl/core"
      xmlns:ui="http://java.sun.com/jsf/facelets">

    <h:head>
       ...
    </h:head>

    <h:body>
       ...
                <h:form id="form">  

                    <!-- polls every minute -->
                    <p:poll interval="60" update="memoryChart" />  

                    <!-- memory line chart -->
                    <p:lineChart id="memoryChart" value="#{monitorController.memoryModel}"
                                 legendPosition="ne" title="Memory Usage" style="height:300px;margin-top:20px"
                                 xaxisLabel="Minutes" yaxisLabel="Bytes" zoom="true"/>  

                </h:form>  
       ...

    </h:body>
</html>


MonitorController

The code corresponding to the monitor controller class is:
package org.javabenchmark;

import java.io.Serializable;
import java.lang.management.ManagementFactory;
import java.lang.management.MemoryMXBean;
import java.lang.management.MemoryUsage;
import javax.faces.bean.ManagedBean;
import javax.faces.bean.SessionScoped;
import org.primefaces.model.chart.CartesianChartModel;
import org.primefaces.model.chart.ChartSeries;
import org.slf4j.LoggerFactory;

/**
 * JSF Managed Bean to get JVM memory usage.
 *
 */
@ManagedBean
@SessionScoped
public class MonitorController implements Serializable {

    /**
     * memory model.
     */
    private CartesianChartModel memoryModel;
    /**
     * memory usage series.
     */
    private ChartSeries memoryUsageSerie, maxMemorySerie;
    /**
     * elapsed time in minutes since the monitoring starts
     */
    private long elapsedTime = -1;

    /**
     * instantiates a new monitoring controller.
     */
    public MonitorController() {
        
        createMemoryModel();
    }
    
    /**
     * initializes memory usage model.
     */
    private void createMemoryModel() {  
        // model
        memoryModel = new CartesianChartModel(); 
        // heap serie
        memoryUsageSerie = new ChartSeries();  
        memoryUsageSerie.setLabel("Heap");
        memoryModel.addSeries(memoryUsageSerie);
        // max serie
        maxMemorySerie = new ChartSeries();  
        maxMemorySerie.setLabel("Max");
        memoryModel.addSeries(maxMemorySerie); 
    }
    
    /**
     * gets the memory model that will be rendered within a LineChart
     * component. The LineChart component should call this method every minute.
     * @return the updated memory model
     */
    public CartesianChartModel getMemoryModel() {  
        
        // gets data
        MemoryMXBean memoryMxBean = ManagementFactory.getMemoryMXBean();
        MemoryUsage memoryUsage = memoryMxBean.getHeapMemoryUsage();
        
        // one more minute
        elapsedTime++;
        
        // updates series
        memoryUsageSerie.set(elapsedTime, memoryUsage.getUsed());
        maxMemorySerie.set(elapsedTime, memoryUsage.getMax()); 
        
        return memoryModel;  
    }
}


Screenshot

Finally, a screenshot of the PrimeFaces line chart representing the memory usage :


Notice that it is possible to zoom with drag and drop, if you have a lot of points displayed.

Conclusion

As you can see, it is quite easy to monitor the JVM and to display collected metrics within a JSF page. You can go further by monitoring CPU usage for instance, or by adding some features like starting/stopping/resetting the monitoring.

Sunday, May 5, 2013

Java instrumentation tutorial

This article explains how Java instrumentation works, and how you can write your own agent to do some basic profiling/tracing.

Overview

JVM instrumentation feature was introduced in JDK 1.5 and is based on byte code instrumentation (BCI). Actually, when a class is loaded, you can alter the corresponding byte code to introduce features such as methods execution profiling or event tracing. Most of Java Application Performance Management (APM) solutions use this mechanism to monitor JVM.

Instrumentation Agent

To enable JVM instrumentation,  you have to provide an agent (or more) that is deployed as a JAR file. An attribute in the JAR file manifest specifies the agent class which will be loaded to start the agent.

There is 2 ways to load the agent:
  • with a command-line interfaceby adding this option to the command-line: -javaagent:jarpath[=options] where jarpath is the path to the agent JAR file. options is the agent options. This switch may be used multiple times on the same command-line, thus creating multiple agents. More than one agent may use the same jarpath.
  • by dynamic loading: the JVM must implement a mechanism to start agents sometime after the the VM has started. That way, a tool can "attach" an agent to a running JVM (for instance profilers or ByteMan)


After the JVM has initializedthe agent class will be loaded by the system class loader. If the class loader fails to load the agent, the JVM will abort.



Next, the JVM instantiates an Instrumentation interface implementation and given the context, tries to invoke one of the two methods that an agent must implement: premain or agentmain.

The premain method is invoked after JVM initialization and the agentmain method is invoked sometime after the JVM has started (if the JVM provides such a mechanism). When the agent is started using a command-line option (with -javaagent), the agentmain method is not invokedThe agent class may also have a premain method for use when the agent is started using a command-line option. When the agent is started after JVM startup the premain method is not invoked.




The signatures of the premain method are:
public static void premain(String agentArgs, Instrumentation inst);
public static void premain(String agentArgs);

And the signatures of the agentmain method are:
public static void agentmain(String agentArgs, Instrumentation inst);
public static void agentmain(String agentArgs);
The agent needs to implement only one signature per method.  The JVM first attempts to invoke the first signature, and if the agent class does not implement it then the JVM will attempt to invoke the second signature.

Byte Code Instrumentation

With the premain and agentmain methods, the agent can register a ClassFileTransformer instance by providing an implementation of this interface in order to transform class files. To register the transformer, the agent can use the addTransformer method of the given Instrumentation instance.


Now, all future class definitions will be seen by the transformer, except definitions of classes upon which any registered transformer is dependent. The transformer is called when classes are loaded, when they are redefined. and optionally, when they are retransformed (if the transformer was added to the instrumentation instance with the boolean canRetransform set to true).

The following method of the ClassFileTransformer interface is responsible of any class file transformation:
byte[] transform(ClassLoader loader, String className, Class<?> classBeingRedefined,
                 ProtectionDomain protectionDomain, byte[] classfileBuffer);

This is where things get complicated because you need to manipulate raw byte code. To achieve this, i strongly suggest to rely on dedicated tools such as ASM or Javassist.

Basic Profiling

To illustrate how the class file transformer can be used, we are going to set up some basic method profiling with Javassist:
  1. Write a trivial class that will be instrumented
  2. Write a ClassFileTransformer to inject some code to print method execution time
  3. Write an agent that registers the previous transformer
  4. Write the corresponding JUnit test
The class to be instrumented is:
package org.javabenchmark.instrumentation;

public class Sleeping {
    
    public void randomSleep() throws InterruptedException {
        
        // randomly sleeps between 500ms and 1200s
        long randomSleepDuration = (long) (500 + Math.random() * 700);
        System.out.printf("Sleeping for %d ms ..\n", randomSleepDuration);
        Thread.sleep(randomSleepDuration);
    }
}

The transformer class is:
package org.javabenchmark.instrumentation;

import java.lang.instrument.ClassFileTransformer;
import java.lang.instrument.IllegalClassFormatException;
import java.security.ProtectionDomain;
import javassist.ClassPool;
import javassist.CtClass;
import javassist.CtMethod;

public class SleepingClassFileTransformer implements ClassFileTransformer {

    public byte[] transform(ClassLoader loader, String className, Class classBeingRedefined,
        ProtectionDomain protectionDomain, byte[] classfileBuffer) throws IllegalClassFormatException {

        byte[] byteCode = classfileBuffer;

        if (className.equals("org/javabenchmark/instrumentation/Sleeping")) {

            try {
                ClassPool cp = ClassPool.getDefault();
                CtClass cc = cp.get("org.javabenchmark.instrumentation.Sleeping");
                CtMethod m = cc.getDeclaredMethod("randomSleep");
                m.addLocalVariable("elapsedTime", CtClass.longType);
                m.insertBefore("elapsedTime = System.currentTimeMillis();");
                m.insertAfter("{elapsedTime = System.currentTimeMillis() - elapsedTime;"
                        + "System.out.println(\"Method Executed in ms: \" + elapsedTime);}");
                byteCode = cc.toBytecode();
                cc.detach();
            } catch (Exception ex) {
                ex.printStackTrace();
            }
        }

        return byteCode;
    }
}

The transformer checks if the class to transform is the Sleeping one, then it injects the code that prints the time elapsed by the execution of the randomSleep() method.

The agent class is:
package org.javabenchmark.instrumentation;

import java.lang.instrument.Instrumentation;

public class Agent {

    public static void premain(String agentArgs, Instrumentation inst) {
        
        // registers the transformer
        inst.addTransformer(new SleepingClassFileTransformer());
    }
}
And finally, the corresponding JUnit code:
package org.javabenchmark.instrumentation;

import org.junit.Test;

public class AgentTest {

    @Test
    public void shouldInstantiateSleepingInstance() throws InterruptedException {

        Sleeping sleeping = new Sleeping();
        sleeping.randomSleep();
    }
}
But, before executing the test you need to setup Maven to enable the JVM agent:
  • build the JAR that contains the agent's code before running test
  • add the JVM option -javagent to the JVM that runs the test
  • add the Javassist dependency
To achieve this, add the following XML code to your pom.xml file, inside the <build><plugins> ... </plugins></build> section:
            
            
                org.apache.maven.plugins
                maven-jar-plugin
                2.4
                
                    
                        process-classes
                        
                            jar
                        
                        
                            
                                
                                    org.javabenchmark.instrumentation.Agent
                                
                            
                        
                    
                
            

            
            
                org.apache.maven.plugins
                maven-surefire-plugin
                2.14
                
                    -javaagent:target/${project.build.finalName}.jar
                
            
Next, add the Javassist dependency:
        
            org.javassist
            javassist
            3.14.0-GA
            jar
        

Then, running the test should produce something like this:
-------------------------------------------------------
 T E S T S
-------------------------------------------------------
Running org.javabenchmark.instrumentation.AgentTest
Sleeping for 818 ms ..
Method Executed in ms: 820
Tests run: 1, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 1.001 sec
You can see that there is a new trace: Method Executed in ms: 820 proving that instrumentation works. Without instrumentation, you would only have the Sleeping for 818 ms .. trace.

Conclusion

It is easy to profile your code with the instrumentation API and the Javassist API: write transformers with Javassist, write an agent to register them, then use the -javaagent option and you're done !

Sunday, April 21, 2013

Fault injection in your JUnit with ByteMan

This article shows how to use ByteMan from JBoss to inject fault when testing Java code.

As usual, let us start with some trivial code to test: a simple class that appends strings to a given file.
package org.javabenchmark.byteman;

import java.io.File;
import java.io.FileOutputStream;
import java.io.IOException;
import java.util.logging.Level;
import java.util.logging.Logger;

/**
 * Logging helper class.
 *
 */
public class LogHelper {

    public static final String LOG_FILE_PROPERTY = "logHelperFile";
    public static final String DEFAULT_LOG_FILE = "sample.log";
    private static final File LOG_FILE = new File(System.getProperty(LOG_FILE_PROPERTY, DEFAULT_LOG_FILE));

    /**
     * logs the given trace to a dedicated file.
     *
     * @param trace a string to be written into the log file.
     */
    public static void log(String trace) {

        try {
            FileOutputStream fos = new FileOutputStream(LOG_FILE, true);
            fos.write(trace.getBytes());
            fos.write('\n');
            fos.flush();
            fos.close();
        } catch (IOException ex) {
            final String msg = "Log Helper can't write trace to the log file: " + LOG_FILE.getAbsolutePath();
            System.err.println(msg);
        }
    }
}


The next step is to write the corresponding test:
package org.javabenchmark.byteman;

import java.io.File;
import static org.fest.assertions.api.Assertions.*;
import org.jboss.byteman.contrib.bmunit.BMRule;
import org.jboss.byteman.contrib.bmunit.BMUnitRunner;
import org.junit.Test;
import org.junit.runner.RunWith;

/**
 * Unit Test dedicated to LogHelper.
 */
public class LogHelperTest {

    private static final String SUCCESSFULL_TEST_LABEL = "Successfull Test";

    private void deleteFile(File file) {
        // clean
        file.delete();
        assertThat(file).doesNotExist();
    }
    
    @Test
    public void shouldLog() {

        File logFile = new File(LogHelper.DEFAULT_LOG_FILE);

        // clean
        deleteFile(logFile);

        // one line test
        LogHelper.log(SUCCESSFULL_TEST_LABEL);

        // control
        assertThat(logFile).exists();
        assertThat(logFile).hasContent(SUCCESSFULL_TEST_LABEL);

        // several lines test
        LogHelper.log(SUCCESSFULL_TEST_LABEL);
        LogHelper.log(SUCCESSFULL_TEST_LABEL);

        // control
        assertThat(logFile).hasContent(SUCCESSFULL_TEST_LABEL + '\n' + SUCCESSFULL_TEST_LABEL + '\n' + SUCCESSFULL_TEST_LABEL);
        deleteFile(logFile);
    }
}

Running the test produces the following output:
Testsuite: org.javabenchmark.byteman.LogHelperTest
Tests run: 1, Failures: 0, Errors: 0, Time elapsed: 0.136 sec

And if you enable code coverage in your favorite IDE, you should have something like this:

Red regions show us parts of the code that are not covered by the test. In our case, we can see that the code that handles IO exception is never invoked. Then, it is not possible to cover all the code of the method without raising an IOException. Hopefully, there is a powerful tool for this: ByteMan.

As explained by the authors:
Byteman is a tool which simplifies tracing and testing of Java programs. Byteman allows you to insert extra Java code into your application, either as it is loaded during JVM startup or even after it has already started running.

Actually, ByteMan is a java agent that does instrumentation and we are going to use it to inject IO exception during the execution of our test to check the code that deals with it. The idea is to throw an IOException when the method FileOutputStream.write(byte[]) is called.

Firstly, the dependency for ByteMan:

  org.jboss.byteman
  byteman-bmunit
  2.1.2


Secondly, modify the previous test to add ByteMan to our JUnit and write a new method to test the code that handles IOException:
package org.javabenchmark.byteman;

import java.io.File;
import static org.fest.assertions.api.Assertions.*;
import org.jboss.byteman.contrib.bmunit.BMRule;
import org.jboss.byteman.contrib.bmunit.BMUnitRunner;
import org.junit.Test;
import org.junit.runner.RunWith;

/**
 * Unit Test dedicated to LogHelper.
 */
@RunWith(BMUnitRunner.class)
public class LogHelperTest {

    private static final String SUCCESSFULL_TEST_LABEL = "Successfull Test";

    private void deleteFile(File file) {
        // ...
    }
    
    @Test
    public void shouldLog() {
        // ...
    }

    @Test
    @BMRule(name = "IOException Rule",
            targetClass = "FileOutputStream",
            targetMethod = "write(byte[])",
            targetLocation = "AT ENTRY",
            condition = "TRUE",
            action = "throw new java.io.IOException()")
    public void shouldHandleIOException() {
        
        File logFile = new File(LogHelper.DEFAULT_LOG_FILE);

        // clean
        deleteFile(logFile);

        // one line test that should raise an io exception
        LogHelper.log(SUCCESSFULL_TEST_LABEL);
        
        // control
        assertThat(logFile).exists();
        assertThat(logFile).hasContent("");
        
        // clean
        deleteFile(logFile);
    }
}


The BMRule annotation defines the execution context for ByteMan:
  • name: to set the name of the rule
  • targetClass and targetMethod: to set the method that must be instrumented
  • targetLocation: to set when to inject  the code (at the beginning, at the end or somewhere in the body of the method)
  • condition: to set some condition to be met
  • action: the code to be injected
So, we have a rule named "IOException Rule" that is activated when the method write(byte[]) of the FileOutputStream class begins. The rule then injects a given code that throws an IOException. When the shouldHandleIOException test runs, every time that write(byte[]) is called then a java.io.IOException is raised.

Now, running all tests produces:
Testsuite: org.javabenchmark.byteman.LogHelperTest
byteman jar is /home/javabenchmark/.m2/repository/org/jboss/byteman/byteman/2.1.2/byteman-2.1.2.jar
Log Helper can't write trace to the log file: /home/javabenchmark/Documents/projects/javabenchmark/sample.log
Tests run: 2, Failures: 0, Errors: 0, Time elapsed: 0.534 sec

And the code coverage looks like this:

The log(String) method is now fully covered (the implicit constructor of the LogHelper class is not tested, explaining the missing percents to reach 100%).

Conclusion

ByteMan is very useful to test code that deals with Exception, especially fault tolerant one. That way, you can check that your fault tolerance strategy is really resilient, or that your code gives you relevant traces when errors occur.

Thursday, April 11, 2013

Check that your code is thread-safe with JUnit and ContiPerf

This short article shows how to check that you write thread-safe code by transforming a JUnit test into a concurrent one with ContiPerf.

Suppose that you have to test a date formatter component that uses the famous SimpleDateFormat class, as illustrated below:
package org.javabenchmark;

import java.text.SimpleDateFormat;
import java.util.Date;

/**
 * Helper dedicated to format date in a standard way.
 */
public class NonThreadSafeDateFormatHelper {

    /**
     * the date format for standard representation.
     */
    private SimpleDateFormat standardDateFormat = new SimpleDateFormat("yyyy-MM-dd");

    /**
     * formats the given date using the standard date format: yyyy-MM-dd.
     *
     * @param date the date to format
     * @return a literal representation of the given date.
     */
    public String toStandardString(Date date) {
        return standardDateFormat.format(date);
    }
}

Then, you could write a junit test like this:
package org.javabenchmark;

import org.junit.Test;

import java.util.Calendar;

import static org.fest.assertions.api.Assertions.*;

public class NonThreadSafeDateFormatHelperTest {

    @Test
    public void shouldFormatRandomDate() {

        // random date
        Calendar calendar = Calendar.getInstance();
        calendar.set(Calendar.YEAR, (int) (1000 + Math.random() * 1000));
        calendar.set(Calendar.DAY_OF_YEAR, (int) (Math.random() * 365));

        // test
        NonThreadSafeDateFormatHelper dateFormatHelperToTest = new NonThreadSafeDateFormatHelper();
        String randomDateString = dateFormatHelperToTest.toStandardString(calendar.getTime());

        // general controls
        assertThat(randomDateString).isNotNull();
        assertThat(randomDateString).hasSize(10);
        // year control
        String literalYear = String.valueOf(calendar.get(Calendar.YEAR));
        assertThat(literalYear).isEqualTo(randomDateString.substring(0, 4));
        // month control
        String literalMonth = String.valueOf(calendar.get(Calendar.MONTH) + 1);
        if (literalMonth.length() == 1) {
            literalMonth = "0" + literalMonth;
        }
        assertThat(literalMonth).isEqualTo(randomDateString.substring(5, 7));
        // day control
        String literalDayh = String.valueOf(calendar.get(Calendar.DAY_OF_MONTH));
        if (literalDayh.length() == 1) {
            literalDayh = "0" + literalDayh;
        }
        assertThat(literalDayh).isEqualTo(randomDateString.substring(8));
    }
}

Next, running the test will produce the following output:
Testsuite: org.javabenchmark.NonThreadSafeDateFormatHelperTest
Tests run: 1, Failures: 0, Errors: 0, Time elapsed: 0.173 sec

The test passes, and you could think that everything is fine, but what will happen if the DateFormatHelper class is used in a concurrent way, for instance from a JSF page to display the current date ?

To check that your code can handle concurrency, you can modify the previous JUnit test like this:
package org.javabenchmark;

import org.junit.Rule;
import org.junit.Test;

import java.text.ParseException;
import java.util.Calendar;
import org.databene.contiperf.PerfTest;
import org.databene.contiperf.junit.ContiPerfRule;

import static org.fest.assertions.api.Assertions.assertThat;

public class NonThreadSafeDateFormatHelperPerfTest {

    @Rule
    public ContiPerfRule i = new ContiPerfRule();
    /**
     * the date format helper to test.
     */
    private NonThreadSafeDateFormatHelper dateFormatHelperToTest = new NonThreadSafeDateFormatHelper();

    @Test
    @PerfTest(invocations = 1000, threads = 2)
    public void shouldFormatRandomDatesConcurrently() throws ParseException {

        // random date
        Calendar calendar = Calendar.getInstance();
        calendar.set(Calendar.YEAR, (int) (1000 + Math.random() * 1013));
        calendar.set(Calendar.DAY_OF_YEAR, (int) (Math.random() * 365));
        
        // test
        String randomDateString = dateFormatHelperToTest.toStandardString(calendar.getTime());

        // general controls
        assertThat(randomDateString).isNotNull();
        assertThat(randomDateString).hasSize(10);
        // year control
        String literalYear = String.valueOf(calendar.get(Calendar.YEAR));
        assertThat(literalYear).isEqualTo(randomDateString.substring(0, 4));
        // month control
        String literalMonth = String.valueOf(calendar.get(Calendar.MONTH) + 1);
        if (literalMonth.length() == 1) {
            literalMonth = "0" + literalMonth;
        }
        assertThat(literalMonth).isEqualTo(randomDateString.substring(5, 7));
        // day control
        String literalDayh = String.valueOf(calendar.get(Calendar.DAY_OF_MONTH));
        if (literalDayh.length() == 1) {
            literalDayh = "0" + literalDayh;
        }
        assertThat(literalDayh).isEqualTo(randomDateString.substring(8));
    }
}

This is the same test, except that the test method is invoked 1000 times by 2 threads, producing this output:
Testsuite: org.javabenchmark.NonThreadSafeDateFormatHelperPerfTest
org.javabenchmark.NonThreadSafeDateFormatHelperPerfTest.shouldFormatRandomDatesConcurrently
samples: 999
max:     19
average: 0.04804804804804805
median:  0
Tests run: 1, Failures: 0, Errors: 1, Time elapsed: 0.377 sec
So, there is now an error, indicating that the DateFormatHelper component is not thread-safe: do not let it go into production :)

Summary

Good unit tests are not sufficient when you are writing components that will evolve in multi-threaded environment, like web applications. You can easily check if your code is vulnerable to race condition with contiperf.

Sunday, April 7, 2013

JVM Biased Locking and micro benchmark

This is a short article that focuses on HotSpot JVM Biased Locking feature. Before explaining what is it, let's take a simple example:
package org.javabenchmark;

public class SynchronizeCounter {
    
    private static long count = 0l;
    
    public synchronized void increment() {
        count = count + 1;
    }
    
    public long getCount() {
        return count;
    }
}
This class manages a thread safe counter. The code uses the synchronized keyword to ensure that only one thread at a time can invoke the increment() method.

Now, a trivial micro benchmark with only one thread incrementing the counter:
package org.javabenchmark;

import org.junit.Test;

public class OneThreadCounterTest {

    @Test
    public void incrementCounterLoop() {
        
        SynchronizeCounter counter = new SynchronizeCounter();
        long time = System.currentTimeMillis();
        for (int i = 0; i < 10000000; i++) {
            counter.increment();
        }
        time = System.currentTimeMillis() - time;
        System.out.printf("Counter: %d - Elapsed time: %d ms\n", counter.getCount(), time);
    }
}

The output is: Counter: 10000000 - Elapsed time: 67 ms

Next, the same test but with a 4.5s wait before:
package org.javabenchmark;

import org.junit.Test;

public class OneThreadCounterTest {

    @Test
    public void incrementCounterLoop() throws InterruptedException {

        Thread.sleep(4500);
        
        SynchronizeCounter counter = new SynchronizeCounter();
        long time = System.currentTimeMillis();
        for (int i = 0; i < 10000000; i++) {
            counter.increment();
        }
        time = System.currentTimeMillis() - time;
        System.out.printf("Counter: %d - Elapsed time: %d ms\n", counter.getCount(), time);
    }
}

The output is: Counter: 10000000 - Elapsed time: 12 ms
Wow, that is a big improvement considering that the only difference between the two tests is the line with Thread.sleep(..).
So, what happened ?

In the second case, the code benefits from the biased locking: we are in the context of un-contended lock, meaning that a lock mechanism is used but without any concurrency (single thread). In that case, the JVM uses biased locking and the running thread does not have to rely on expensive atomic instructions to acquire the lock.

Biased locking is enabled by default but only after 4 seconds, that is why the second test benefits from it. You can suppress this 4 seconds wait with the following JVM option:
-XX:BiasedLockingStartupDelay=0

Conclusion

When dealing with un-contended micro-benchmark be aware of the biased locking startup delay and use the previous JVM option to disabled it.