Extending JMeter

There are several ways to extend JMeter and add functionality. JMeter is designed

to make this task easier.

[A good overview of the process of extending JMeter](/JMeter Extension Scenario/)
Creating your own Timer
Creating your own SampleListener (such as a visualizer, or reporter)
Creating your own Config Element
Creating your own logic SamplerController
Creating your own test sample SamplerController
Creating your own Sampler
Making your custom elements play nice as a JMeter UI component
Making your custom elements saveable and loadable from within JMeter

Creating your own Timer

The timer interface:

public long delay();

Not too complicated. Your delay method must, each time it is called, return a

long representing the number of milliseconds to delay. The constant timer returns the

same number every time it’s called. A random timer returns a different number each time.

Creating your own SampleListener

The SampleListener interface:

public void sampleOccurred(SampleEvent e);

      public void sampleStarted(SampleEvent e);

      public void sampleStopped(SampleEvent e);

sampleOccurred is the method called when a sample is completed, and the data has been

collected. The SampleEvent object should contain all the information gathered

from the sample. If your sample listener is primarily concerned with collecting the

data from a test run, you can implement this method - the other two are for other purposes and

can be ignored (though the methods have to be there for your class to compile).

sampleStarted and sampleStopped are used to indicate the state of the sampling thread.

This is useful for visualizers that show the user the state of all running threads

(ie, they are running and waiting for response, or they’re stopped and waiting

to begin again).

Creating your own Config Element

The ConfigElement interface:

public void addConfigElement(ConfigElement config);

      public boolean expectsModification();

      public Object clone();

The ConfigElement interface is sparse. All ConfigElements are expected to implement

a public clone() method. The reason for this is that config elements will be cloned

for each different sampling thread, and most will be cloned for each sample.

If your config element expects to be modified in the process of a test run,

and you want those modifications to carry over from sample to sample (as in

a cookie manager - you want to save all cookies that gets set throughout

the test), then return true for the expectsModification() method. Your config element will not be

cloned for each sample. If your config elements are more static in nature,

return false. If in doubt, return false.

addConfigElement() is required so that config elements can be layered. For

instance, let’s say a user creates a URL entry that contains default values -

they might use this to specify a server. Then, all their test samples configure

individual test cases, but leave out the server field. This information is combined

via the addConfigElement() method. Your custom config elements should do the right

thing when this method is called. Normally, this involves ignoring such calls unless

the passed in ConfigElement is of the same type as yours, and then only merging in

values that are not already set in the object receiving the call (ie you probably

don’t want to overwrite any values).

You may have noticed there’s no specification on how to get the config information

out of a ConfigElement. This raises the question, who is going to use it?

At the end of the line, there will be a Sampler that will need the information held

in your config element. The sampler that uses your config element needs to know more

about the class than the rest of JMeter - that information is not part of this interface.

If at all possible, extend AbstractConfigElement when creating your own. By doing so,

and by following some simple rules, you will get cloning and saving to XML of your

config element for free (as in, you don’t have to do anything!). AbstractConfigElement

stores all its values in a Map, and provides getProperty and putProperty methods. Your

config element can provide getXXX() and setXXX() methods, but these should delegate

to getProperty() and setProperty(), probably using static Strings as keys in the Map.

You can store any type of object, provided the objects are cloneable and Saveable

(Strings, Integer, Long, Double, Float are all good in this regard).

One caveat - if your config element has been restored from file, all the values

held in the Map will be String objects (except for elements that implement Saveable

on their own), and you may have to do casting and parsing. Example: an Integer will

have to be converted from a String to an int, so your getXXX() method should check

for this possibility to avoid exceptions.

Creating your own logic SamplerController

The SamplerController interface looks as follows:

Entry nextEntry();

      Collection getListeners();

      void addSamplerController(SamplerController controller);

      void addConfigElement(ConfigElement config);

      Object clone();

Again, clone() is a method that must be implemented to all SamplerControllers to avoid

contamination between sampling threads.

The nextEntry() method is the essential job of a SamplerController - to deliver

Entry objects to be sampled. An Entry object encapsulates all the information needed

by a Sampler to do its job. The nextEntry() method should work like an iterator and

continuously return new Entry objects.

There are two boundary conditions that need to be handled. If the Controller has no

more Entries to give, for the rest of the test, it should return null. Therefore,

if your Controller has sub-controllers it is receiving Entries from, it should remove

them from its list of controllers to get Entries from. The other condition is when

your controller reaches the end of its list of Entries, and it needs to start over

from the beginning. The parent Controller needs to know this so that it can move

on to its next controller in its list. Therefore, at the end of each iteration,

your SamplerController needs to return a CycleEntry object instead of a normal Entry.

Conversely, this means that if your Controller receives a CycleEntry object, it should

move on to the next Controller in its list.

A logic controller does not generate Entries on its own, but simply regulates

the flow of Entries from its sub-controllers. A logic controller might provide

looping logic, or it might modify the Entries that pass through it, or whatever.

GenericController provides an implementation that does absolutely nothing but

pass Entries on from its sub-controllers. This class is useful both for reference

purposes and to extend, since it provides a lot of methods you’re likely to find

useful

getListeners() is an odd member of this Class. It’s there to serve those who

want their controller to receive sample data. This would be useful for a controller

that modified Entry objects based on previous sample results (like an HTML spider

that dynamically reacted to previously sampled webpages for links and forms). The

responsibility of the controller implementer is to collect all potential listeners

from the sub-controller list, and add themselves if desired. Most SamplerControllers

that extend GenericController don’t have to do anything.

addSamplerController(SamplerController controller) is the method used to

add sub controllers to your SamplerController.

addConfigElement(ConfigElement config) Your SamplerController should also

be capable of holding configuration elements and adding them to Entries as they

pass through your controller. Again, see GenericController for reference. Essentially,

all Entry objects that get returned by nextEntry() are handed all the ConfigElements

of the controller.

Creating your own test sample SamplerController

A SamplerController that generates Entry objects is just like a logic controller

except that it creates its own Entry objects instead of gathering them from

sub-controllers (although, to be fully correct, your test sample SamplerController

should handle both possibilities). Your test sample SamplerController can also

benefit from extending GenericController. By doing so, most of your cloning and

saving needs are handled (but probably not entirely). See HttpTestSample as

reference.

Creating your own Sampler

The Sampler interface:

public SampleResult sample(Entry e)

Your Sampler has two responsibilities. Of lesser importance, it should do whatever

it is you want to do, given an Entry object that hopefully contains information

about what is to be sampled. Of greater importance, your sampler should return

a SampleResult object that holds information about the sampling. Information such

as how long the sample took, the text response from the sample (if appropriate), and

a string that describes the location of what was sampled. The SampleResult interface

is essentially a Map with public static Strings as keys.

Making your custom elements play nice as a JMeter UI component

In order to take part in the JMeter UI, your component needs to implement the

JMeterComponentModel interface:

Class getGuiClass();

      public String getName();

      public void setName(String name);

      public Collection getAddList();

      public String getClassLabel();

      public void uncompile();

Most of this stuff is easy, boring, and tedious. getName(), setName() is a simple

String property that is the name of the object. getClassLabel() should return

a String that describes the class. This string will be displayed to the user and

so should be short but meaningful. getGuiClass() should return a Class object for

the class that will be used as a GUI component. This class should be a subclass

of java.awt.Container, and preferably a subclass of javax.swing.JComponent.

getAddList() should return a list of either Strings or JMenus. These Strings

represent the Classes that can be added to your SamplerController. Each String

should correspond to the target class’s getClassLabel() String. MenuFactory is

a class that will return some preset menu lists (such as all available SamplerControllers,

all available ConfigElements, etc).

uncompile() is a cleanup method used between sampling runs. When the user

hits “Start”, JMeter “compiles” the objects in the tree. Child nodes are added

to their parent objects recursively until there is one TestPlan object, which is

then submitted for testing. Afterward, these elements have to un-added from their

parent objects, or uncompiled. To uncompile your class, simply clear all your

data structures that are holding sub-elements. For your SamplerController, this

will be the list of sub-controllers and the list of ConfigElements.

That’s it, except for your GUI class. If your SamplerController has no

configuration needs, just return org.apache.jmeter.gui.NamePanel, and the user will

at least be able to change the name of your component. Otherwise, create a gui class

that implements the ModelSupported interface:

void setModel(Object model);

      public void updateGui();

setModel() is used to hand your JMeterModelComponent class to the GUI class when

it is instantiated. It is your responsibility for providing the means by which

the Gui class updates the values in the model class. For updating in the other

direction, there is updateGui(), which the model class can call if necessary.

Note, normally, this call is made for you automatically whenever the Gui is brought

to the screen. If you are creating a Visualizer, then you may need to use updateGui().

For reference, refer to UrlConfigGui (in org.apache.jmeter.protocol.http.config.gui).

If you have done all this correctly, there’s just one more step. If you compile

your classes into the ApacheJMeter.jar file, then you’re done. Your classes will

be automatically found and used. Otherwise, you will need to modify jmeter.properties.

The search_paths property should be modified to include the path where your

classes are. This does not obviate the need for your classes to be in the JVM’s

CLASSPATH - it is an additional requirement. Otherwise, your classes will not be

detected, and the Gui will not make them available to the user.

Making your custom elements saveable and loadable from within JMeter

The Saveable interface has just one method:

public Class getTagHandlerClass()

This method simply returns the Class object that represents the Class that handles

the saving and loading of your component.

To write this SaveHandler, make a class that extends TagHandler

(from org.apache.jmeter.save.xml). Note, if your component extends AbstractConfigElement,

it is already fully Saveable - provided you only have information stored in

the Map from AbstractConfigElement.

To write your own TagHandler, you will have to implement the following methods:

public abstract void setAtts(Attributes atts) throws Exception

      public String getPrimaryTagName()

      public void save(Saveable objectToSave,Writer out) throws IOException

getPrimaryTagName() should return the String that is the XML tagname that your

class handles. When you save your object, it should all be contained within an

XML tag of the same name. This will ensure that when JMeter’s parser hits that tag,

your class will be called upon to handle the data.

setAtts(Attributes atts) is called when the parser first hits your tag.

If this primary tag has any attributes, this method represents your chance to save

the information.

save(Saveable objectToSave,Writer out) - when the user selects “Save”,

JMeter will call this method and hand the Saveable object to be saved (it will be

the object that specified your TagHandler as the class responsible for saving it).

This method should use the given Writer object to print all the XML necessary to

save the current state of the objectToSave.

There’s more you have to do to handle creating a new Object when JMeter parses

an XML file. However, there’s no standard interface you need to implement, but rather,

JMeter uses reflection to generate method calls into your class. When JMeter hits

a tag that corresponds to your PrimaryTagName, an instance of your TagHandler will

be created, and its setAtts() method will get called. Thereafter, methods are called

depending on subsequent tags and character data. For every tag, JMeter calls

<tag-name>TagStart(Attributes atts), and for every end tag, JMeter calls

<tag-name>TagEnd().

Additionally, JMeter will call a method that corresponds to all tags that are

current. So, for instance, if JMeter runs into a tag name “foo”, then

foo(Attributes atts) will be called. If JMeter then parses character data,

then foo(String data) will be called. If JMeter parses a tag within foo, called

“nestedFoo”, then JMeter will call foo_nestedFoo(Attributes atts) and

foo_nestedFoo(String data). And so on.

An annotated example:

public class AbstractConfigElementHandler extends TagHandler

{

    private AbstractConfigElement config;

    private String currentProperty;

    public AbstractConfigElementHandler()

    {

    }

    /**

     * Returns the AbstractConfigElement object parsed from the XML.  This method

     * is required to fulfill the SaveHandler interface.  It is used by the XML

     * routines to gather all the saved objects.

     */

    public Object getModel()

    {

        return config;

    }

    /**

     * This is called when a tag is first encountered for this handler class to handle.

     * The attributes of the tag are passed, and the SaveHandler object is expected

     * to instantiate a new object.

     */

    public void setAtts(Attributes atts) throws Exception

    {

        String className = atts.getValue("type");

        config = (AbstractConfigElement)Class.forName(className).newInstance();

    }

    /**

     * Called by reflection when a &lt;property&gt; tag is encountered.  Again, the

     * attributes are passed.

     */

    public void property(Attributes atts)

    {

        currentProperty = atts.getValue("name");

    }

    /**

     * Called by reflection when text between the begin and end &lt;property&gt;

     * tag is encountered.

     */

    public void property(String data)

    {

        if(data != null && data.trim().length() &gt; 0)

        {

            config.putProperty(currentProperty,data);

            currentProperty = null;

        }

    }

    /**

     * Called by reflection when the &lt;property&gt; tag is ended.

     */

    public void propertyTagEnd()

    {

        // Here's a tricky bit.  See below for explanation.

        List children = xmlParent.takeChildObjects(this);

        if(children.size() == 1)

        {

            config.putProperty(currentProperty,((TagHandler)children.get(0)).getModel());

        }

    }

      /**

    * Gets the tag name that will trigger the use of this object's TagHandler.

    */

    public String getPrimaryTagName()

    {

        return "ConfigElement";

    }

  /**

    * Tells the object to save itself to the given output stream.

    */

    public void save(Saveable obj,Writer out) throws IOException

    {

        AbstractConfigElement saved = (AbstractConfigElement)obj;

        out.write("&lt;ConfigElement type=\"");

        out.write(saved.getClass().getName());

        out.write("\">\n");

        Iterator iter = saved.getPropertyNames().iterator();

        while (iter.hasNext())

        {

            String key = (String)iter.next();

            Object value = saved.getProperty(key);

            writeProperty(out,key,value);

        }

        out.write(&lt;/ConfigElement&gt;");

    }

    /**

     * Routine to write each property to xml.

     */

    private void writeProperty(Writer out,String key,Object value) throws IOException

    {

        out.write("&lt;property name=\"");

        out.write(key);

        out.write("\">\n");

        JMeterHandler.writeObject(value,out);

        out.write("\n&lt;/property&gt;\n");

    }

In the propertyTagEnd() method, takeChildObjects() is called on the xmlParent

instance variable. xmlParent is inherited from TagHandler - the DocumentHandler

object that is running the show. xmlParent takes an XML file that represents a portion of

the test configuration tree, and recreates a tree-like data structure. When it is

done, it will convert its tree-like data structure into the test configuration tree

structure.

However, sometimes, a tree element has sub objects that you do not want represented

in the tree - rather, they are part of your object. But, they may

be complicated enough to warrant their own SaveHandler class, and thus, the xmlParent

picks them up as part of its tree. When the tag is done, and you know that there are

child objects you want to grab, you can call the takeChildObjects() method and get a

List object containing them all. This will remove them from the tree, and you can add

them to your object that you’re creating.

UrlConfig is good example. It extends AbstractConfigElement, so it uses exactly the

code above to save and reload itself from XML. However, one of the pieces of data

that UrlConfig stores is an Arguments object. Arguments is too complicated to save

to file as a simple string, so it has its own Handler object (ArgumentsHandler). In

the above code, when the call to JMeterHandler.writeObject(value,out) is made, the

writeObject method detects whether the object implements Saveable, and if so, calls

the object’s SaveHandler class to deal with it. This means, however, that when

reading that XML file, the Argument object will show up as a separate entity in

the data tree, whereas it originally was just part of the data of the UrlConfig

object. In order to preserve that relationship, it’s necessary for the

AbstractConfigElementHandler to check after each property tag is done for child

objects in the tree, and take them for its own use.

Study the other SaveHandler objects and the TagHandler class to learn more

about how saving is accomplished. Once you understand the design, writing your

own SaveHandler is very easy.

Extending JMeter