Java Operator Development Guide

Edit me

Java is a deeply integrated part of Streams. Making Java easy in Streams has been a priority for years because of its extensive libraries and ubiquitous developer base. The goal of this guide is to take a two-tiered approach. The videos and instructions are going to help you get your Java operators up and running as quickly and easily as possible (Tier 1). However, Streams Studio abstracts many details away from the developer to make their life easy, so we also go more into depth on how Java and Streams work together under the cover (Tier 2). Following the CLI versions of instructions will give you a deeper understanding of Java operators in Streams, while following the videos and Streams Studio instructions will get your code running as fast as possible.

Operator Life Cycle

Unless you’re writing an extremely simple Java operator, it’s important to understand how the operator life cycle works. All operators written in Java must implement the following interface and implement a no-argument constructor:

com.ibm.streams.operator.Operator   

public interface Operator {
	public void initialize(OperatorContext context) throws Exception;
	public void allPortsReady() throws Exception;
	public void process(StreamingInput<Tuple> port, Tuple tuple) throws Exception;
	public void processPunctuation(StreamingInput<Tuple> port, Punctuation mark) throws Exception;
	public void shutdown() throws Exception;
}      

In general, you will extend AbstractOperator, which implements the Operator interface and takes default actions. You will then override the methods that need to be customized. Here is a brief explanation of the required methods:

  • void initialize(OperatorContext context) - This is called once before any tuples are processed and before any of the other required methods. Here is where you should put code to establish connections to external systems and data. You will have access to parameters set in the SPL operator invocation.

  • void allPortsReady() - This is called once after the initialize() method has returned and all input and output ports are connected and ready to receive/submit tuples. Operators that process incoming tuples generally won’t use this, since the process method is used to handle tuples. In the case of a source operator, there are no incoming tuples so this is where the tuple producing threads are started. We will cover this more in the source operator section.

  • void process(StreamingInput<Tuple> port, Tuple tuple) - This is where the manipulation of incoming tuples will take place, followed by submission to an output port or an external connection (in the case of a sink operator). The performance of your operator is almost completely dependent on how efficient your process method is. See the Improving Performance section for some tips.

  • void processPunctuation(StreamingInput<Tuple> port, Punctuation mark) - This is where incoming punctuation markers that arrived on the specified port are processed. The two types of punctuation to be handled are window and final.

  • void shutdown() - This is where connections are closed and resources related to any external system or data store are released. This method is invoked when a job is cancelled or an operator is stopped.

Creating Your First Java Operator

Types of Operators

  • Process Operator - Data comes in as a tuple on an input data stream. That data is processed and the result is submitted as a tuple to an output stream. Process operators make up most of a typical Streams operator graph.
  • Source Operator - Data is read from an external system (or generated) and submitted to an output port as a tuple stream.
  • Sink Operator - Data comes in as a tuple on an input data stream, and is then written to an external system.

In our first Java primitive operator, we will create an operator that converts incoming strings to upper case. This example is really simple, but it demonstrates some basic concepts about Java primitive operators. This guide will build on this example to demonstrate the concepts in each section.

StringToCaps.java

@PrimitiveOperator()
@InputPorts(@InputPortSet(cardinality=1))
@OutputPorts(@OutputPortSet(cardinality=1))
public class StringToCaps extends AbstractOperator {
    @Override
    public final void process(StreamingInput<Tuple> inputStream, Tuple tuple)
            throws Exception {
        // Create a new tuple for output port 0
        StreamingOutput<OutputTuple> outStream = getOutput(0);
        OutputTuple outTuple = outStream.newTuple();
        
		// Copy incoming attributes to output tuple if they're in the output schema
        outTuple.assign(tuple);
        
        // Get attribute from input tuple and manipulate
        String myString = tuple.getString("myString");
        myString = myString.toUpperCase();
        outTuple.setString("myString", myString);

        // Submit new tuple to output port 0
        outStream.submit(outTuple);
    }
}
package stringToCaps;

import com.ibm.streams.operator.AbstractOperator;
import com.ibm.streams.operator.OutputTuple;
import com.ibm.streams.operator.StreamingInput;
import com.ibm.streams.operator.StreamingOutput;
import com.ibm.streams.operator.Tuple;
import com.ibm.streams.operator.model.InputPortSet;
import com.ibm.streams.operator.model.InputPorts;
import com.ibm.streams.operator.model.OutputPortSet;
import com.ibm.streams.operator.model.OutputPorts;
import com.ibm.streams.operator.model.PrimitiveOperator;


@PrimitiveOperator()
@InputPorts(@InputPortSet(cardinality=1))
@OutputPorts(@OutputPortSet(cardinality=1))
public class StringToCaps extends AbstractOperator {
    @Override
    public final void process(StreamingInput<Tuple> inputStream, Tuple tuple)
            throws Exception {
        // Create a new tuple for output port 0
        StreamingOutput<OutputTuple> outStream = getOutput(0);
        OutputTuple outTuple = outStream.newTuple();
        
		// Copy incoming attributes to output tuple if they're in the output schema
        outTuple.assign(tuple);
        
        // Get attribute from input tuple and manipulate
        String myString = tuple.getString("myString");
        myString = myString.toUpperCase();
        outTuple.setString("myString", myString);

        // Submit new tuple to output port 0
        outStream.submit(outTuple);
    }
}

Key points to note from this example:

  • Extend AbstractOperator - The Java primitive operator needs to implement the Operator interface. The simplest way to do that is to extend from the AbstractOperator and override the methods that you want to customize. In this cases, the process() method is the only method that needs to be overridden.

  • @PrimitiveOperator - The @PrimitiveOperator annotation marks the StringToCaps class to be the implementation class of a Java primitive operator. The toolkit indexer looks for any classes that has this annotation and will include it as part of the containing Streams toolkit. You may configure other high-level operator properties such as operator name, namespace and description using this annotation.

  • @InputPorts - Defines one or more @InputPortSets that describe the ports for incoming tuples.
    • @InputPortSet - Contains the property definitions for one or more ports. The available properties are description, cardinality, optional, windowingMode, and windowPunctuationInputMode. Only the last @InputPortSet within an @InputPorts definition can have a cardinality of -1 (define multiple input ports).
  • @OutputPorts - Defines one or more @OutputPortSets that describe the ports for outgoing tuples.
    • @OutputPortSet - Contains the property definitions for one or more ports. The available properties are description, cardinality, optional, windowingMode, and windowPunctuationOutputMode. Only the last @OutputPortSet within an @OutputPorts definition can have a cardinality of -1 (define multiple input ports).
  • process(StreamingInput<Tuple> inputStream, Tuple tuple) - This is the main method that gets called when a tuple is received by the operator. In our example, the operator first copies any incoming attributes to the outgoing tuple (in cases where attribute names match). Next, the operator gets the myString attribute from the incoming stream. It converts the string to upper case, and assigns the resulting string to the output tuple. The operator then submits the output tuple to the output port. This example demonstrates a common pattern in the process method for a Java primitive operator:
    1. Forward all attributes that aren’t getting manipulated by this operator.
    2. Get attributes that you want to manipulate from the incoming tuple using getter methods.
    3. Manipulate the attributes in the desired way.
    4. Write attributes to the output tuple using setter methods.
    5. Submit output tuple.

Building Java Primitive Operator


The steps for building the simple Java primitive operator from above on the command line are as follows:

  1. Create a directory for your SPL toolkit (MyJavaOp in this case). Create these directories as well:
  2. MyJavaOp/impl/java/src/
    MyJavaOp/impl/java/src/stringToCaps
    MyJavaOp/impl/java/bin/
    
  3. Place your StringToCaps.java operator class in
    MyJavaOp/impl/java/src/stringToCaps
  4. Compile the Java operator class from the SPL toolkit directory (MyJavaOp) using:
  5. javac -cp $STREAMS_INSTALL/lib/com.ibm.streams.operator.jar impl/java/src/stringToCaps/StringToCaps.java -d impl/java/bin/
  6. Index the toolkit from the SPL toolkit directory. This will generate the operator model and build the toolkit directory structure.
  7. spl-make-toolkit -i ./

The steps for building the simple Java primitive operator using Streams Studio are as follows:

  1. File -> New -> Project
  2. Expand InfoSphere Streams Studio
  3. Select SPL Project
  4. Name your project (MyJavaOp in this case) and then click Finish
  5. In the generated project, right-click and select New -> Java Primitive Operator
  6. Set the namespace to stringToCaps and the Name to be StringToCaps
  7. Click Finish
  8. In the generated template, StringToCaps.java, modify the process operator to look like the example above.
  9. Save your changes and Studio will automatically build your operator model and create the toolkit directory structure.


When the Java compiler is run, by having the com.ibm.streams.operator.jar file in the class path, the Java compiler invokes the Streams annotation processor to process the annotations on the Java primitive operator class. The Streams annotation processor generates an operatorName$StreamsModel.class which contains all the information required to run the Java operator from SPL. Using this information, the annotation processor then generates the operator model for the Java primitive operator.

When spl-make-toolkit is run, the toolkit indexer scans the toolkit for operator models. It includes operators defined in the operator model into the toolkit index. This step is required to make the Java primitive operator accessible from SPL.

In both the CLI and the Streams Studio cases, the standard toolkit structure will look approximately like this:

/+ <toolkit>
   /* toolkit.xml
   /+ <name-space>
      /+ <operator>
      	/* <operator model>
   /+ impl
      /+ bin
      /+ include
      /+ java
          /+ src
          /+ bin
      /+ lib
      /+ src             

If you followed the CLI instructions, this should be your directory structure:

/+ MyJavaOp
   /* toolkit.xml
   /+ stringToCaps
      /+ StringToCaps
      	/* StringToCaps.xml
   /+ impl
      /+ java
          /+ src
          	/+ stringToCaps
          		/+ StringToCaps.java
          /+ bin
          	/+ stringToCaps
          		/* StringToCaps.class   
          		/* StringToCaps$StreamsModel.java 
          		/* StringToCaps$StreamsModel.class          
  • [toolkit] - root directory of the toolkit. Typically the name of the root directory matches the name of the toolkit in info.xml.
  • toolkit.xml - a generated resource by the SPL toolkit indexing processing. This file should not be modified and is used by the compiler and various components in Streams.
  • [name-space] - similar to packages in Java, allow you to group related SPL toolkit artifacts together.
    • [operator] - a directory containing all artifacts relative to a Java / C++ primitive operator. Each operator has its own separate directory.
  • impl - the impl directory contains implementation code of primitive operators or native functions
    • impl/java/src - contains Java primitive operator or native function code
    • impl/java/bin - contains Java class files for the Java implementation code
  • lib - contains any external libraries that this toolkit may require

Note: If you use Streams Studio to create a new SPL project, Streams Studio will automatically create the correct directory structure for you.

Running and Testing Your Operator

SPL Toolkit Dependency

The best way to start testing your Java operator is to create a simple SPL application. Typically, you want this test application in a separate SPL toolkit, such that the test code is independent from the Java primitive operator. We will create a new toolkit called TestJavaOp.

For your test application to be able to access the Java primitive operator, the TestJavaOp toolkit must add the MyJavaOp toolkit to the info.xml file as a toolkit dependency.


Create `TestJavaOp` directory for your toolkit. In that directory, create a file named `info.xml`. In the info.xml, specify toolkit dependency as shown in the file below.
  1. We specify `MyJavaOp` toolkit as a dependency for `TestJavaOp`
  2. We specify that `TestJavaOp` depends on version `1.0.0` to `2.0.0` (exclusive) of the `TestJavaOp` toolkit.

There are two ways to do this in Studio. You can specify the dependency on MyJavaOp during the creation of a new SPL Application Project (as shown in the video above). If the application is already created, you can edit the dependencies.

  1. Add dependency during project creation:
    1. File -> New -> Project. Select SPL Application Project and click Next.
    2. Type in your Project, Namespace, and Main composite names. Click Next.
    3. In the Dependencies section, expand Workspace Projects and select MyJavaOp. In cases where you are using an external toolkit, they will show up in this box for selection as well. Click Finish.
  2. Add dependency to an existing project by editing dependencies.
    1. Expand the SPL Application Project that you want to add a dependency to.
    2. Right-click on Dependencies and select Edit Dependencies...
    3. Click Add... and select the toolkit dependency you want to add. Note: If you don't see the toolkit that you want to add a dependency on, then you probably haven't added the toolkit location to your workspace. Read how to do that here.


The resulting info.xml file should look like this:

<?xml version="1.0" encoding="UTF-8"?>
<info:toolkitInfoModel xmlns:common="http://www.ibm.com/xmlns/prod/streams/spl/common" xmlns:info="http://www.ibm.com/xmlns/prod/streams/spl/toolkitInfo">
  <info:identity>
    <info:name>TestJavaOp</info:name>
    <info:description></info:description>
    <info:version>1.0.0</info:version>
    <info:requiredProductVersion>4.1.0.0</info:requiredProductVersion>
  </info:identity>
  <info:dependencies>
    <info:toolkit>
      <common:name>MyJavaOp</common:name>
      <common:version>[1.0.0,2.0.0)</common:version>
    </info:toolkit>
  </info:dependencies>
</info:toolkitInfoModel>

Test Application

In the TestJavaOp toolkit, we create a test application that calls the Java primitive operator, StringsToCap. A common pattern for testing a primitive operator is Beacon -> Java Operator -> FileSink as in the following example.

Streams Studio

SPL Code for TestJavaOp (TestJavaOp.spl, located in /TestJavaOp/application/):

namespace application ;

use stringToCaps::StringToCaps ;

composite TestJavaOp
{
  graph
    (stream<rstring myString> Beacon_1_out0) as Beacon_1 = Beacon()
    {
      param
        period : 1.0 ;
      output
        Beacon_1_out0 : myString = "lowercase" +(rstring) IterationCount() ;
    }
    (stream<rstring myString> StringToCaps_2_out0) as StringToCaps_2 =
    StringToCaps(Beacon_1_out0){}

    () as FileSink_3 = FileSink(StringToCaps_2_out0 as inputStream)
    {
      param
        file : "results.txt" ;
        flush : 1u;
    }
}

Key points to note from this example:

  • Beacon - The Beacon operator is provided as part of the SPL Standard Toolkit. It is a great tool for testing primitive operators as it can generate a continuous data stream of any type. In the example, the Beacon operator generates an rstring of lowercase + InterationCount(). InterationCount() is an output function provided by the Beacon operator to query how many tuples have been generated by the operator.
  • StringToCaps - The output stream from the Beacon operator is fed into the StringToCaps operator. For each tuple received by this Java primitive operator, its process method is called. In our case, the Java primitive operator converts each of the incoming strings to uppercase and submits it in an output tuple.
  • FileSink - The FileSink operator receives the incoming tuples from StringToCaps and prints the result to a file named results.txt. The flush parameter tells the operator to flush output to file for each tuple it has received.

Running a Test Application

Compile and launch the TestJavaOp main composite. The quickest way to test an application is to run it in stand-alone mode.

cd [root directory  of TestJavaOp toolkit]

sc -M application::TestJavaOp --output-directory=[build-output-dir] --data-directory=data -T -a -t ../MyJavaOp

Key points to note from the command above:

  • -T - compiles the application in stand-alone mode. To compile in distributed mode, remove this option.
  • -t ../MyJavaOp - specifies the toolkit path for the compiler to find the MyJavaOp toolkit. Alternatively, set this location with using the STREAMS_SPL_PATH environment variable.
  • –output-directory - specifies where the output of the compilation should go.
  • –data-directory - specifies the default location for files to be read from and written to (make sure you create a data directory at your application root)

To run this program:

cd [build-output-dir]
standalone

You should see the following output in <workspace>/TestJavaOp/data/results.txt:

"LOWERCASE0"
"LOWERCASE1"
"LOWERCASE2"
"LOWERCASE3"

You may also compile and run the application in distributed mode as shown in the video.

Referencing External Libraries

Your Java operators can take advantage of all JARs and existing Java code you already have. This makes connecting to any servers, databases, etc that have Java clients easy, and it makes converting your Java code to Streams simple.

There are two simple steps that are required to start using external libraries in your Java operator:

  1. Make JARs of interest available to the Java operator at Streams runtime.
  2. Use the @Libraries annotation to add the JARs to the operator class path.

Accessing JARS at Runtime

When the Java primitive operator is executed, the operator code must be able to access the dependent JAR files at runtime. In designing your operator, you need to determine how the Java primitive operator can access the external JARS. There are two general approaches, each with its own pros and cons.

  1. Package the JAR as part of the toolkit

    In this approach, the JAR file is packaged as part of the toolkit, and eventually as part the Streams application bundle when the application is built. This approach simplifies the set up of the distributed environment, as you do not to need to manually set up these JARs on the resources in the domain. When the application is run and deployed, the Streams runtime automatically copies these dependencies to the remote systems. The disadvantage of this approach is that it makes the size of the application bundle larger. This affects the time required to submit the application to the Streams instance.

    For more information about application bundles, refer to this documentation: Application bundle files.

  2. Using Environment Variables

    In this approach, the JAR files are stored at the same file location on all resources that can run the operator. The operator then defines an environment variable for the end user to indicate where to find the JAR files. When running the application, the environment variable must be set on the instance. The advantage of this approach is that it makes the application bundle smaller, thereby, reducing the time required to submit a job to an instance. The disadvantage is that the end user must now manually set up the environment before the operator can be used.

Setting up the Classpath

To set up the classpath for the Java primitive operator, use the @Libraries annotation. This annotation sets up the classpath for the Java primitive operator at runtime.

Options for specifying classpaths using the @Libraries annotation:

  • Specific JARs: @Libraries("opt/Reverse.jar")
  • Entire directories using wildcard (this adds all files with extension .jar or .JAR): @Libraries("opt/*")
  • Environment variables: @Libraries("@REVERSE_HOME@")where the environment variable would be REVERSE_HOME.
    WARNING: The environment variable is resolved at compile-time, resulting in an application bundle that is not relocatable.

To specify multiple locations for JARs, simply comma separate your locations:
@Libraries({"opt/Reverse.jar" , "opt/downloaded/*", "@REVERSE_HOME@"})

Added Library Example

The example in the video above uses a JAR with a simple function that reverses a String (download here).


The steps for using an external JAR with the command-line are as follows:

  1. In your Java primitive operator toolkit directory, create an opt/ directory. Place the Reverse.jar file in the MyJavaOp/opt/ directory.
  2. In your operator Java code, import the package:
  3. import reverse.Reverse;
  4. Use the @Libraries annotation to add the JAR to the operator's class path. The @Libraries parameter can be placed above the operator class definition with the other annotations. You will also need to import the @Libraries library (Yes, you're importing a library in order to import a library :-) ).
  5. import com.ibm.streams.operator.model.Libraries;
    ...
    @Libraries("opt/Reverse.jar")
  6. Go down to the process method, and before setting the "myString" output attribute, add this line to reverse the String:
  7.     @Override
        public final void process(StreamingInput<Tuple> inputStream, Tuple tuple)
                throws Exception {
            // Create a new tuple for output port 0
            StreamingOutput<OutputTuple> outStream = getOutput(0);
            OutputTuple outTuple = outStream.newTuple();
            
    		// Copy incoming attributes to output tuple if they're in the output schema
            outTuple.assign(tuple);
            
            // Get attribute from input tuple and manipulate
            String myString = tuple.getString("myString");
            myString = myString.toUpperCase();
            myString = Reverse.reverse(myString);
            outTuple.setString("myString", myString);
    
            // Submit new tuple to output port 0
            outStream.submit(outTuple);
        }
    
  8. Compile the Java operator class as you did before, but add the JAR to the class path:
  9. javac -cp $STREAMS_INSTALL/lib/com.ibm.streams.operator.jar:opt/Reverse.jar impl/java/src/stringToCaps/StringToCaps.java -d impl/java/bin/
  10. Index the toolkit again.
  11. spl-make-toolkit -i ./
  12. Rebuild your test application and submit. The output in results.txt will look like this:
  13. "0ESACREWOL"
    "1ESACREWOL"
    "2ESACREWOL"
    "3ESACREWOL"
    

The steps for using an external JAR with Streams Studio are as follows:

  1. In your Java operator project folder, create an opt directory. Add the Reverse.jar file to it.
  2. In the Project Explorer, refresh your Resources folder and expand opt. Right-click on Reverse.jar and select Build Path -> Add to Build Path.
  3. Import the package:
  4. import reverse.Reverse;
  5. Use the @Libraries annotation to add the JAR to the operator's class path. (If you ever get "class not found" exceptions once you submit your job, check your @Libraries annotation first).
  6. @Libraries("opt/Reverse.jar")
  7. Go down to the process method, and before setting the "myString" attribute, add this line to reverse the String:
  8.     @Override
        public final void process(StreamingInput<Tuple> inputStream, Tuple tuple)
                throws Exception {
            // Create a new tuple for output port 0
            StreamingOutput<OutputTuple> outStream = getOutput(0);
            OutputTuple outTuple = outStream.newTuple();
            
    		// Copy incoming attributes to output tuple if they're in the output schema
            outTuple.assign(tuple);
            
            // Get attribute from input tuple and manipulate
            String myString = tuple.getString("myString");
            myString = myString.toUpperCase();
            myString = Reverse.reverse(myString);
            outTuple.setString("myString", myString);
    
            // Submit new tuple to output port 0
            outStream.submit(outTuple);
        }
    
  9. Save and let Studio automatically build your toolkit.
  10. Rebuild your test application and submit. The output in results.txt will look like this:
  11. "0ESACREWOL"
    	"1ESACREWOL"
    	"2ESACREWOL"
    	"3ESACREWOL"

Making Your Operator Generic with Parameters

Parameters allow your Java primitive operator to be more generic and enables the end-user to configure and control the behavior of the operator.

The example in the video above shows how to generalize the StringToCaps operator so that it reverses (or doesn’t reverse) the incoming string based on a reverseString boolean parameter in the SPL operator code.

Below are the general steps to create a parameter for a Java primitive operator:

  1. Create a private member variable for the parameter in your operator class.
  2. Create a setter method for setting the newly define private field.
  3. Annotate the setter method with the @Parameter annotation. You may configure the parameter name, description, and cardinality, etc using this annotation. The annotated setter method will be recognized by the Streams runtime as the method for initializing the parameter value. The setter method will be called before the initialize method is called. Note: The default name for a parameter is the portion of the setter method name after “set”. For example, setReverseString defaults to an SPL parameter name of reverseString if the name property is not specified.
  4. Modify your operator logic/code to honor the parameter value.

To learn more about the different kinds of parameters and the supported SPL types, refer to this documentation: Parameter Annotation Javadoc

Example Parameter

In our example, we will introduce a boolean parameter named reverseString to control if the incoming String from a tuple should be reversed or not.

  1. In the operator class, add a Boolean member field that will be used to hold the parameter value. Default the parameter value to be false if it is not specified.

      
     private Boolean reverseString = false;
     
  2. Add a setter method for reverseString.
  3. Annotate the setter method with the @Parameter annotation. Since the parameter is of type Boolean, the cardinality is automatically set to 1, meaning that only a single value can be provided. (You will need to import: import com.ibm.streams.operator.model.Parameter;) l

    
       @Parameter(name = "reverseString", optional = true,
       	description = "Boolean value. Reverse strings if true. Default: false.")
       public void setReverseString(Boolean value){
       	reverseString = value;
       }
     
  4. In the process() method, modify the operator logic to only reverse the string if the reverseString field is true.

      public final void process(StreamingInput<Tuple> inputStream, Tuple tuple)
              throws Exception {
          // Create a new tuple for output port 0
          StreamingOutput<OutputTuple> outStream = getOutput(0);
          OutputTuple outTuple = outStream.newTuple();
    	        
          // Copy incoming attributes to output tuple if they're in the output schema
          outTuple.assign(tuple);
             
          String myString = tuple.getString("myString");
          myString = myString.toUpperCase();
          if (reverseString){
          	myString = Reverse.reverse(myString);
          }
          outTuple.setString("myString", myString);
          // Submit new tuple to output port 0
          outStream.submit(outTuple);
      }
    
  5. Save and build the operator.

  6. To test the parameter, add a param clause at the invocation of the StringToCaps operator:

    
     (stream<rstring myString> StringToCaps_2_out0) as StringToCaps_2 =
             StringToCaps(Beacon_1_out0)
         {
             param
                 reverseString : true;
         }
     
  7. Run the application. You should see output similar to the following:

    "0ESACREWOL"
    "1ESACREWOL"
    "2ESACREWOL"
    "3ESACREWOL"
    

More Parameter Examples

String parameter - SPL parameter queueName of type ustring (rstring also works).

    private String queueName = "";
    ...
    ...
    ...
    @Parameter(optional = true, description = "Name of the queue. If not specified, a name will be randomly generated name.")
    public void setQueueName(String value) {
        queueName = value;
    }

int parameter - SPL parameter messageSendRetryDelay of type int32.

    int messageSendRetryDelay = 10000;
    ...
    ...
    ...
    @Parameter(optional = true, description = "This optional parameter specifies the time in milliseconds before the retrying delivery.")
    public void setMessageSendRetryDelay(int value) {
        messageSendRetryDelay = value; 
    }

List<String> parameter - SPL parameter routingKey of type list<ustring>.

    private List<String> routingKeys = new ArrayList<String>();
    ...
    ...
    ...
    @Parameter(optional = true, description = "Routing key/keys to bind the queue to.")
    public void setRoutingKey(List<String> values) {
        if(values!=null)
            routingKeys.addAll(values);
    }

Creating a Source Operator

At this point, you have learned the most important Java operator basics. Before getting into more advanced topics, we will introduce Source and Sink operators.

Source operators are unique because they don’t rely on an incoming tuple to trigger the sending of an output tuple. Source operators are typically used to bring data into your Streams application from an external source such as a database or messaging server.

Below are the basic steps for creating a source operator:

  1. Override the initialize() method to:
    1. Initialize the AbstractOperator super class: super.initialize(context)
    2. Set up connections with external data sources.
    3. Initialize base state for the operator.
    4. Create a thread that will produce tuples using the produceTuples() method.
  2. Override the allPortsReady() method to start your production thread.
  3. Use a produceTuples() to generate tuples. Typically this is done within a while loop that terminates on operator shutdown.
  4. Override the shutdown() method to:
    1. Interrupt your production thread.
    2. Close and cleanup connections to external data sources.
    3. Shutdown the AbstractOperator super class: super.shutdown()

Example Source Operator

In this example, we will implement a source operator that gets data from an external server system. This is a simulated server. The server provides the following interface to get data from the server. While this example is simple, and the interface does not reflect a real external system, this example demonstrates the key concepts required for implementing a source operator. You can download Server.java or try with your own external system.


/*
 * Simulated Server interface
 */
public interface Server {
	/*
	 * Connect to the server to receive data
	 */
	public void initialize(String userId, String password) throws Exception;

	/*
	 * Read the next record from the server
	 */
	String getData() throws Exception;
	
	/*
	 * Send a record to the server
	 */
	void sendData(String message) throws Exception ;

	/*
	 * Disconnect from the server
	 */
	public void disconnect() throws Exception;
}

Below is the ServerSource operator that uses this interface. Place this code in its own “serverConnections” namespace directory (<workspace>/MyJavaOp/impl/java/src/serverConnections). Refer to the Creating Your first Java Operator section for help on how to do this.

@PrimitiveOperator()
@OutputPorts(@OutputPortSet(cardinality=1))
public class ServerSource extends AbstractOperator {
    private Thread processThread;
    private Server server;
    private boolean shutdown = false;

    @Override
    public synchronized void initialize(OperatorContext context)
            throws Exception {
        // Must call super.initialize(context) to correctly setup an operator.
        super.initialize(context);

        server = new Server();
        server.initialize("myUser", "myPassw0rd");

        processThread = getOperatorContext().getThreadFactory().newThread(
                new Runnable() {

                    @Override
                    public void run() {
                        try {
                            produceTuples();
                        } catch (Exception e) {
                            Logger.getLogger(this.getClass()).error("Operator error", e);
                        }                    
                    }

                });

        processThread.setDaemon(false);
    }

    @Override
    public synchronized void allPortsReady() throws Exception {
        processThread.start();
    }

    private void produceTuples() throws Exception  {
        final StreamingOutput<OutputTuple> out = getOutput(0);

        OutputTuple tuple = out.newTuple();
        while(!shutdown ){
            String myString = server.getData();
            tuple.setString("myString", myString);
            out.submit(tuple);
            Thread.sleep(1000);
        }        
    }
	
	@Override
    public synchronized void shutdown() throws Exception {
        shutdown = true;
        if (processThread != null) {
            processThread.interrupt();
            processThread = null;
        }
        server.disconnect();
        super.shutdown();
    }

}
package serverConnections;

import org.apache.log4j.Logger;

import com.ibm.streams.operator.AbstractOperator;
import com.ibm.streams.operator.OperatorContext;
import com.ibm.streams.operator.OutputTuple;
import com.ibm.streams.operator.StreamingOutput;
import com.ibm.streams.operator.model.OutputPortSet;
import com.ibm.streams.operator.model.OutputPorts;
import com.ibm.streams.operator.model.PrimitiveOperator;

@PrimitiveOperator()
@OutputPorts(@OutputPortSet(cardinality=1))
public class ServerSource extends AbstractOperator {
    private Thread processThread;
    private Server server;
    private boolean shutdown = false;

     /*
     * Initialize this operator. Called once before any tuples are processed.
     * @param context OperatorContext for this operator.
     * @throws Exception Operator failure, will cause the enclosing PE to terminate.
     */
    @Override
    public synchronized void initialize(OperatorContext context)
            throws Exception {
        // Must call super.initialize(context) to correctly setup an operator.
        super.initialize(context);

        server = new Server();
        server.initialize("myUser", "myPassw0rd");

        /*
         * Create the thread for producing tuples.
         * The thread is created at initialize time but started.
         * The thread will be started by allPortsReady().
         */
        processThread = getOperatorContext().getThreadFactory().newThread(
                new Runnable() {

                    @Override
                    public void run() {
                        try {
                            produceTuples();
                        } catch (Exception e) {
                            Logger.getLogger(this.getClass()).error("Operator error", e);
                        }                    
                    }

                });

        /*
         * Set the thread not to be a daemon to ensure that the SPL runtime
         * will wait for the thread to complete before determining the
         * operator is complete.
         */
        processThread.setDaemon(false);
    }

    /**
     * Notification that initialization is complete and all input and output ports
     * are connected and ready to receive and submit tuples.
     * @throws Exception Operator failure, will cause the enclosing PE to terminate.
     */
    @Override
    public synchronized void allPortsReady() throws Exception {
        processThread.start();
    }

    /**
     * Submit new tuples to the output stream
     * @throws Exception if an error occurs while submitting a tuple
     */
    private void produceTuples() throws Exception  {
        final StreamingOutput<OutputTuple> out = getOutput(0);

        while(!shutdown ){
					  OutputTuple tuple = out.newTuple();
            String myString = server.getData();
            tuple.setString("myString", myString);
            out.submit(tuple);
            Thread.sleep(1000);
        }        
    }

    /**
     * Shutdown this operator, which will interrupt the thread
     * executing the <code>produceTuples()</code> method.
     * @throws Exception Operator failure, will cause the enclosing PE to terminate.
     */
    @Override
    public synchronized void shutdown() throws Exception {
        shutdown = true;
        if (processThread != null) {
            processThread.interrupt();
            processThread = null;
        }
        server.disconnect();
        super.shutdown();
    }

}

Key things to note from this example:

  1. In the initialize() method:

    • Call super class’s initialize(context) method to properly initialize the operator.
    • Create a Server object and make a connection via its Server.initalize(...) method.
    • Use the operator context thread factory to create a thread. This thread will be used to read data from the server. The thread is set up in the initialize(...) method, but it is not started. When the operator is initializing, connections to downstream operators may not have been made yet. Therefore, it is not recommended to start reading from the external system at this point.
  2. In the allPortsReady() method:

    • At this point, connections to downstream operators have been established. Therefore, we can start the tuple processing thread and start reading from external system.
  3. In the produceTuples() method:

    This method is called by the processing thread that we have started from allPortsReady. In this method:

    • Get the output port from the operator: out
    • Create a while loop to start reading from server until the operator is shut down.
    • From the output port, create a new tuple.
    • Read data from the server.
    • Assign data to the output tuple.
    • Submit the tuple to the output port
  4. In the shutdown() method:

    This method is called when the operator is shutdown by the runtime. This is where the operator should clean up: In this method:

    • Stop the processing thread
    • Disconnect from the server
    • Always call super.shutdown() in the shutdown method

This covers the basics of creating a source operator. Streams Java Operator APIs come with many useful patterns and samples. You may extend from these patterns when implementing the source operator. To have access to these patterns, include the com.ibm.streams.operator.sample.jar in your classpath. For more information about these samples and patterns, refer to this documentation: Java Operator Samples

Creating a Sink Operator

Sink operators are unique because they don’t send output tuples (unless there is an error port). Sink operators are typically used to send data to an external system such as a database or messaging server.

Below are the basic steps for creating a sink operator:

  1. Override the initialize() method to:
    1. Initialize the AbstractOperator super class: super.initialize(context)
    2. Setup connections with external data sources.
    3. Initialize base state for the operator.
  2. Override the allPortsReady() method to start your production thread.
  3. Use a produceTuples() to generate tuples. Typically this is done within a while loop that terminates on operator shutdown.
  4. Override the shutdown() method to:
    1. Interrupt your production thread.
    2. Close and cleanup connections to external data sources.
    3. Shutdown the AbstractOperator super class: super.shutdown()

Example Sink Operator

In this example, we will implement a sink operator that sends data to the simulated server from the Source Operator section. While this example is simple, and the interface does not reflect a real external system, this example demonstrates the key concepts required for implementing a sink operator.

Below is the ServerSink operator:

@PrimitiveOperator()
@InputPorts({@InputPortSet(cardinality=1)})
public class ServerSink extends AbstractOperator {
    private Server server;
    private Logger trace = Logger.getLogger(this.getClass());
    
    @Override
    public synchronized void initialize(OperatorContext context)
            throws Exception {
        super.initialize(context);
  
        server = new Server();
        server.initialize("myUser", "myPassw0rd");
    }
    
    @Override
    public final void process(StreamingInput<Tuple> inputStream, Tuple tuple)
            throws Exception {      
        // Get message from tuple and submit to server
        String myString = tuple.getString("myString");
        
        try {
            server.sendData(myString);
        } catch (Exception e){
            e.printStackTrace();
            trace.log(TraceLevel.ERROR, e.getMessage());
        }
    }
    
    @Override
    public synchronized void shutdown() throws Exception {
        server.disconnect();
        super.shutdown();
    }
    
}
package serverConnections;


import org.apache.log4j.Logger;

import com.ibm.streams.operator.AbstractOperator;
import com.ibm.streams.operator.OperatorContext;
import com.ibm.streams.operator.StreamingInput;
import com.ibm.streams.operator.Tuple;
import com.ibm.streams.operator.log4j.TraceLevel;
import com.ibm.streams.operator.model.InputPortSet;
import com.ibm.streams.operator.model.InputPortSet.WindowMode;
import com.ibm.streams.operator.model.InputPortSet.WindowPunctuationInputMode;
import com.ibm.streams.operator.model.InputPorts;
import com.ibm.streams.operator.model.PrimitiveOperator;

/**
 * Class for an operator that consumes tuples and does not produce an output stream. 
 * This pattern supports a number of input streams and no output streams. 
 * <P>
 * The following event methods from the Operator interface can be called:
 * </p>
 * <ul>
 * <li><code>initialize()</code> to perform operator initialization</li>
 * <li>allPortsReady() notification indicates the operator's ports are ready to process and submit tuples</li> 
 * <li>process() handles a tuple arriving on an input port 
 * <li>processPuncuation() handles a punctuation mark arriving on an input port 
 * <li>shutdown() to shutdown the operator. A shutdown request may occur at any time, 
 * such as a request to stop a PE or cancel a job. 
 * Thus the shutdown() may occur while the operator is processing tuples, punctuation marks, 
 * or even during port ready notification.</li>
 * </ul>
 * <p>With the exception of operator initialization, all the other events may occur concurrently with each other, 
 * which lead to these methods being called concurrently by different threads.</p> 
 */
@PrimitiveOperator(name="ServerSink", namespace="serverConnections",
description="Java Operator ServerSink")
@InputPorts({@InputPortSet(description="Port that ingests tuples", cardinality=1, optional=false, windowingMode=WindowMode.NonWindowed, windowPunctuationInputMode=WindowPunctuationInputMode.Oblivious), @InputPortSet(description="Optional input ports", optional=true, windowingMode=WindowMode.NonWindowed, windowPunctuationInputMode=WindowPunctuationInputMode.Oblivious)})
public class ServerSink extends AbstractOperator {
    private Server server;
    private Logger trace = Logger.getLogger(this.getClass());
    
    /**
     * Initialize this operator. Called once before any tuples are processed.
     * @param context OperatorContext for this operator.
     * @throws Exception Operator failure, will cause the enclosing PE to terminate.
     */
    @Override
    public synchronized void initialize(OperatorContext context)
            throws Exception {
        // Must call super.initialize(context) to correctly setup an operator.
        super.initialize(context);
        server = new Server();
        server.initialize("myUser", "myPassw0rd");
    }

    /**
     * Process an incoming tuple that arrived on the specified port.
     * @param stream Port the tuple is arriving on.
     * @param tuple Object representing the incoming tuple.
     * @throws Exception Operator failure, will cause the enclosing PE to terminate.
     */
    @Override
    public void process(StreamingInput<Tuple> stream, Tuple tuple)
            throws Exception {
        // Get message from tuple and submit to server
        String myString = tuple.getString("myString");
        
        try {
            server.sendData(myString);
        } catch (Exception e){
            e.printStackTrace();
            trace.log(TraceLevel.ERROR, e.getMessage());
        }
    }


    /**
     * Shutdown this operator.
     * @throws Exception Operator failure, will cause the enclosing PE to terminate.
     */
    @Override
    public synchronized void shutdown() throws Exception {
        server.disconnect();
        super.shutdown();
    }
    
}

Key things to note from this example:

  1. In the initialize() method:

    • Call super class’s initialize(context) method to properly initialize the operator.
    • Create a Server object and make a connection via its Server.initalize(...) method.
  2. In the process() method:

    This method is called every time a tuple is received (as in the standard operator case). In this method:

    • Read the message we want to send from the incoming tuple.
    • Send the message to the server.
  3. In the shutdown() method:

    This method is called when the operator is shutdown by the runtime. This is where the operator should clean up. In this method:

    • Disconnect from the server
    • Always call super.shutdown() in the shutdown method

This covers the basics of creating a sink operator. Streams Java Operator APIs come with patterns and samples that you can extend when implementing a sink operator. For more information about these samples and patterns, refer to this documentation: Java Operator Samples

Handling Errors

Your Streams operator should never crash unless you want it to. There are cases where an exception will be thrown because of a loss of connection or other normal error causes, but it is important to handle those errors so that your operator can continue to process data.

When an exception is thrown, the best-practice response is to:

  1. Catch the exception.
  2. Log the exception as an error in the operator log.
  3. Write the error out to an error port.

There are some cases where you will actually want your operator to fail. These cases are typically only during the initial startup of your operator and are usually related to bad configuration, connections, or parameters.

In the code below, we enhanced our ServerSource operator to include error handling in the produceTuples() method.

@PrimitiveOperator()
@OutputPorts({@OutputPortSet(cardinality=1),@OutputPortSet(description="Error Port", cardinality=1)})
public class ServerSource extends AbstractOperator {
    private Thread processThread;
    private Server server;
    private boolean shutdown = false;
    private Logger trace = Logger.getLogger(this.getClass());
	.
	.
	.
    private void produceTuples() throws Exception  {
        final StreamingOutput<OutputTuple> out = getOutput(0);
        final StreamingOutput<OutputTuple> error = getOutput(1);

        OutputTuple tuple = out.newTuple();
        while(!shutdown ){
            try{
                String myString = server.getData();
                tuple.setString("myString", myString);
                out.submit(tuple);
                Thread.sleep(1000);
            } catch (Exception e){
                trace.log(TraceLevel.ERROR, "Error submitting tuple. Message: " + e.toString());
                OutputTuple errorTuple = error.newTuple();
                errorTuple.setString("error_message", "Error submitting tuple. Message: " + e.toString());
                error.submit(errorTuple);
            }
        }      
    }  
}
package serverSource;

import org.apache.log4j.Logger;
import com.ibm.streams.operator.AbstractOperator;
import com.ibm.streams.operator.OperatorContext;
import com.ibm.streams.operator.OutputTuple;
import com.ibm.streams.operator.StreamingOutput;
import com.ibm.streams.operator.log4j.TraceLevel;
import com.ibm.streams.operator.model.OutputPortSet;
import com.ibm.streams.operator.model.OutputPorts;
import com.ibm.streams.operator.model.PrimitiveOperator;

@PrimitiveOperator()
@OutputPorts({@OutputPortSet(cardinality=1),@OutputPortSet(description="Error Port", cardinality=1)})
public class ServerSource extends AbstractOperator {
    private Thread processThread;
    private Server server;
    private boolean shutdown = false;
    private Logger trace = Logger.getLogger(this.getClass());

    @Override
    public synchronized void initialize(OperatorContext context)
            throws Exception {
        super.initialize(context);

        server = new Server();
        server.initialize("myUser", "myPassw0rd");

        processThread = getOperatorContext().getThreadFactory().newThread(
                new Runnable() {

                    @Override
                    public void run() {
                        try {
                            produceTuples();
                        } catch (Exception e) {
                            Logger.getLogger(this.getClass()).error("Operator error", e);
                        }                    
                    }

                });

        processThread.setDaemon(false);
    }

    @Override
    public synchronized void allPortsReady() throws Exception {
        processThread.start();
    }

    private void produceTuples() throws Exception  {
        final StreamingOutput<OutputTuple> out = getOutput(0);
        final StreamingOutput<OutputTuple> error = getOutput(1);

        OutputTuple tuple = out.newTuple();
        while(!shutdown ){
            try{
                String myString = server.getData();
                tuple.setString("myString", myString);
                out.submit(tuple);
                Thread.sleep(1000);
            } catch (Exception e){
                trace.log(TraceLevel.ERROR, "Error submitting tuple. Message: " + e.toString());
                OutputTuple errorTuple = error.newTuple();
                errorTuple.setString("error_message", "Error submitting tuple. Message: " + e.toString());
                error.submit(errorTuple);
            }
        }      
    }

    public synchronized void shutdown() throws Exception {
        shutdown = true;
        if (processThread != null) {
            processThread.interrupt();
            processThread = null;
        }
        server.disconnect();
        super.shutdown();
    }

}

Adding Compile-Time and Runtime Checks

Adding compile-time and runtime checks to your Java operator is done using the @ContextCheck.

Common compile-time checks:

  • Check to see if an operator that does not support consistent region is within a consistent region.
  • Check incoming and outgoing port schemas. Make sure that the attributes you require are available at an SPL level.

Common runtime checks:

  • Check that a parameter has a valid value.

@ContextCheck - This operator annotation allows you to run checks at operator compile-time or before the initialize() method is called at runtime. The check is done by creating a public static method that takes an argument of type OperatorContextChecker. Runtime @ContextCheck methods are invoked before invocation of @Parameter annotated methods. The compile-time and runtime checks use Boolean values (default is compile = true):

  • @ContextCheck(compile = true)
  • @ContextCheck(runtime = true)

The following example is part of the StringToCaps operator. We have added compile-time checks to make sure that the incoming and outgoing SPL streams have an attribute named “myString” that is of type rstring. This code is placed inside the StringToCaps operator definition. If any of the checks run by checker fail, the compile is interrupted with an error message. Be sure to import the two libraries that are needed.

	import com.ibm.streams.operator.OperatorContext.ContextCheck;
	import com.ibm.streams.operator.compile.OperatorContextChecker;
	.
	.
	.
    @ContextCheck
    public static void checkAttributes(OperatorContextChecker checker){
        OperatorContext context = checker.getOperatorContext();

        //Check that myString attributes exist
        checker.checkRequiredAttributes(context.getStreamingInputs().get(0), "myString");
        checker.checkRequiredAttributes(context.getStreamingOutputs().get(0), "myString");

        //Check the myString attributes are of correct type
        Attribute incoming = context.getStreamingInputs().get(0).getStreamSchema().getAttribute("myString");
        checker.checkAttributeType(incoming, Type.MetaType.RSTRING);
        Attribute outgoing = context.getStreamingOutputs().get(0).getStreamSchema().getAttribute("myString");
        checker.checkAttributeType(outgoing, Type.MetaType.RSTRING);
    }

Using Windows

Windows are an important part of most Streams applications. Intelligent use of windows can allow you to use the same Streams application for real-time processing, as well as batch processing. Batch processing such as map reduce can be done in Streams by using large window sizes.

Types of Windows:

  • Tumbling - Tumbling windows operate in batches. When a tumbling window fills up, all of the tuples in the window are evicted.
  • Sliding - Sliding windows operate in an incremental fashion. When a sliding window fills up, the future tuple insertions result in evicting the oldest tuples in the window.

For more details, read Window Handling.

General strategy for implementing a windowed operator:

  1. Have your operator class extend AbstractWindowOperator (instead of AbstractOperator).
  2. Create a window handler class that implements StreamsWindowListener. This window handler class will be in place of a process method in most cases. Your window handler class should:
    • Have a constructor that takes a StreamingOutput<OutputTuple> argument.
    • Override the void handleEvent(StreamWindowEvent<Tuple> event) and develop a switch case to handle tumbling and/or sliding events. Tumbling:
        @Override
         public synchronized void handleEvent(StreamWindowEvent<Tuple> event) throws Exception {
      
           switch (event.getType()) {
             case INSERTION:
                  // handle insertion of tuples into the window
                  break;
             case EVICTION:
                  // handle the tumble
                  break;
             case FINAL:
                  // handle final mark
                  break;
             }
         }
       

      Sliding:

        @Override
         public synchronized void handleEvent(StreamWindowEvent<Tuple> event) throws Exception {
      
           switch (event.getType()) {
               case INSERTION:
                   // handle insertion of tuples into the window
                   break;
               case INITIAL_FULL:
                   // handle first time window fills
                   break;
               case TRIGGER:
                   // handle trigger events
                   break;
               case PARTITION_EVICTION:
                   // handle partial evictions
                   break;
               case EVICTION:
                   // handle the tumble
                   break;
               case FINAL:
                   // handle final mark
                   break;
           }
         }
       
  3. In the initialize(…) method of the operator code, register a StreamWindowListener with your window handler class:

     getInput(0).getStreamWindow().registerListener(new WindowHandler(getOutput(0)), false);
    

Window Example

In the example below, we implement a tumbling window operator that submits the alphabetic minimium String for a given window. The operator maintains a List<String> that is cleared every time the window tumbles. Read more details about tumbling and sliding windows by clicking on the links.

@PrimitiveOperator()
@InputPorts(@InputPortSet(cardinality=1, windowingMode=WindowMode.Windowed))
@OutputPorts(@OutputPortSet(cardinality=1))
public class MinWindowString extends AbstractWindowOperator {
    @Override
    public synchronized void initialize(OperatorContext context)
            throws Exception {
        super.initialize(context);
        getInput(0).getStreamWindow().registerListener(
                new WindowHandler(getOutput(0)), false);

    }
}
package minWindow;


import org.apache.log4j.Logger;

import com.ibm.streams.operator.OperatorContext;
import com.ibm.streams.operator.StreamingData.Punctuation;
import com.ibm.streams.operator.StreamingInput;
import com.ibm.streams.operator.Tuple;
import com.ibm.streams.operator.model.InputPortSet;
import com.ibm.streams.operator.model.InputPortSet.WindowMode;
import com.ibm.streams.operator.model.InputPortSet.WindowPunctuationInputMode;
import com.ibm.streams.operator.model.InputPorts;
import com.ibm.streams.operator.model.OutputPortSet;
import com.ibm.streams.operator.model.OutputPortSet.WindowPunctuationOutputMode;
import com.ibm.streams.operator.model.OutputPorts;
import com.ibm.streams.operator.model.PrimitiveOperator;
import com.ibm.streams.operator.window.AbstractWindowOperator;


@PrimitiveOperator()
@InputPorts(@InputPortSet(cardinality=1, windowingMode=WindowMode.Windowed))
@OutputPorts(@OutputPortSet(cardinality=1))
public class MinWindowString extends AbstractWindowOperator {
    @Override
    public synchronized void initialize(OperatorContext context)
            throws Exception {
        super.initialize(context);
        getInput(0).getStreamWindow().registerListener(
                new WindowHandler(getOutput(0)), false);

    }
}

</code></pre>
public class WindowHandler implements StreamWindowListener<Tuple> {
    private int tupleCount;
    private List<String> stringList = new ArrayList<String>();
    private final StreamingOutput<OutputTuple> output;
    
    public WindowHandler(StreamingOutput<OutputTuple> output) {
          this.output = output;
        }
    
    @Override
    public synchronized void handleEvent(StreamWindowEvent<Tuple> event) throws Exception {

        switch (event.getType()) {
        case INSERTION:
            for (Tuple tuple : event.getTuples()) {
                String myString = tuple.getString("myString");
                stringList.add(myString);
                tupleCount++;
            }
            break;
        case EVICTION:
            if (tupleCount != 0) {
                OutputTuple tuple = output.newTuple();
                java.util.Collections.sort(stringList);
                
                //get first alphabetical String
                String firstString = stringList.get(0);
                tuple.setString("myString", firstString);
                
                output.submit(tuple);
                output.punctuate(Punctuation.WINDOW_MARKER);
                stringList.clear();
                tupleCount = 0;
            }
            break;
        case FINAL:
            // handle final mark
            break;
        default:
            break;
        }
    }
}

package minWindow;

import java.util.ArrayList;
import java.util.List;
import com.ibm.streams.operator.OutputTuple;
import com.ibm.streams.operator.StreamingData.Punctuation;
import com.ibm.streams.operator.StreamingOutput;
import com.ibm.streams.operator.Tuple;
import com.ibm.streams.operator.window.StreamWindowEvent;
import com.ibm.streams.operator.window.StreamWindowListener;

public class WindowHandler implements StreamWindowListener<Tuple> {
    private int tupleCount;
    private List<String> stringList = new ArrayList<String>();
    private final StreamingOutput<OutputTuple> output;
    
    public WindowHandler(StreamingOutput<OutputTuple> output) {
          this.output = output;
        }
    
    @Override
    public synchronized void handleEvent(StreamWindowEvent<Tuple> event) throws Exception {

        switch (event.getType()) {
        case INSERTION:
            for (Tuple tuple : event.getTuples()) {
                String myString = tuple.getString("myString");
                stringList.add(myString);
                tupleCount++;
            }
            break;
        case EVICTION:
            if (tupleCount != 0) {
                OutputTuple tuple = output.newTuple();
                java.util.Collections.sort(stringList);
                
                //get first alphabetical String
                String firstString = stringList.get(0);
                tuple.setString("myString", firstString);
                
                output.submit(tuple);
                output.punctuate(Punctuation.WINDOW_MARKER);
                stringList.clear();
                tupleCount = 0;
            }
            break;
        case FINAL:
            // handle final mark
            break;
        default:
            break;
        }
        
        
    }

}

Defining Custom Metrics

A metric represents a measurement of an element in either an operator or a processing element. A metric’s value is an signed 64-bit integer value, represented as a primitive long in the Java Operator API. A metric enables the operator keep track of current state, and allows the end-user or external systems to monitor the operator performance based on metric values.

Two types of metrics are provided by the SPL language runtime:

  • System metrics - predefined and maintained by the SPL runtime (this includes things like number of tuples processed on a port, tuple flow rate, etc.)
  • Custom metrics - created and maintained by the operator.

There are three kinds of metrics:

  • Counter indicates that this metric represents a count of occurrence of some event.
  • Gauge indicates a value that is continuously variable with time.
  • Time indicates a metric that represents a point in time.

Live feed of the system metrics and operator custom metrics can be viewed in Streams Studio and the Streams Console. You may retrieve the latest metrics using the Streams REST APIs. This allows you to monitor your Streams domain, and your applications using your favorite monitoring tools. Refer to this documentation about the Metrics REST APIs.

For more information about metrics, refer to the Metrics Javadoc.

This section will show you how to add your a custom metric to your Java operator.

Custom Metric Example

Below are steps to add a custom operator metric. In this example, we will try to create a metric that counts the number of characters the Java primitive operator StringToCaps has processed.

  1. Import the Metric library, com.ibm.streams.operator.Metrics:

    import com.ibm.streams.operator.metrics.Metric;
    
  2. Add a private field of type Metric. In our example, we added the numCharacters field in the StringToCaps class:

    private Metric numCharacters;
    
  3. Create a setter method for the Metric field. Add the @CustomMetric annotation to the setter method.

    @CustomMetric(name = "numCharacters", kind = Metric.Kind.COUNTER,
        description = "Count of the number of characters processed."))
    public void setNumCharacters(Metric runtimeMetric){
        numCharacters = runtimeMetric;
    }
    

    The @CustomMetric annotation let you define the name and description of the metric. In addition, you can specify the metric kind: COUNTER, GUAGE OR TIME. Optionally, specify if the metric should be registered with the platform’s MBean server, using the mxbean property. The default value is false.

    The setter method is called before initialization similar to the way @Parameter sets are done.

  4. In the process() method, increment the numCharacters metric by the length of myString right before the outgoing tuple is submitted:

    @Override
    public final void process(StreamingInput<Tuple> inputStream, Tuple tuple)
            throws Exception {
    
    	// Create a new tuple for output port 0
        StreamingOutput<OutputTuple> outStream = getOutput(0);
        OutputTuple outTuple = outStream.newTuple();
            
        // Copy incoming attributes to output tuple if they're in the output schema
        outTuple.assign(tuple);
            
        String myString = tuple.getString("myString");
        myString = myString.toUpperCase();
        if (reverseString){
        	myString = Reverse.reverse(myString);
        }
        outTuple.setString("myString", myString);
        numCharacters.incrementValue(myString.length());
        // Submit new tuple to output port 0
        outStream.submit(outTuple);
    }
    
  5. Once you save and build your toolkit and test application, submit your application. In the instance graph view in Studio, you will be able to hover over the StringToCaps operator and see a live update of numCharacters.

    Streams Studio

Problem Determination and Debugging

Debugging your Java operator is similar to debugging normal Java.

Here is the path that most Streams developers take to determine if their operator is working:

  1. Go to the instance graph in Streams Studio and see if your operator is healthy. If there is a series of three unhealthy operators, it is typically the fault of the middle one (it’s crash makes the connections to the operators on either side unhealthy).
  2. If your operator is unhealthy, try looking at standard out for that PE: right click on your operator -> Show Log -> Show PE Console
  3. If you still don’t have the information you need, increase the level of logging (this can be set during application launch) and look through the Operator Trace: right click on your operator -> Show Log -> Show Operator Trace

You can also use the Eclipse Java Debugger by following these instructions.

SPL to Java Type Mapping

It’s not always obvious which SPL types map to which Java types. It’s important to get this mapping right when you are defining parameters, reading from input tuples, and writing to output tuples. Follow this link for a comprehensive table of type mapping.

Improving Performance

As we mentioned at the beginning of this guide, your performance will depend on the efficiency of your process(…) or produceTuples(…) methods (in the case of a windowed operator, it will be in your window handler).

Here are some things to keep in mind:

  • Do not print anything to standard out.
  • Avoid excessive logging in your process method. If necessary, make sure that your logging is protected by an if statement:
    if (trace.isInfoEnabled())
       trace.log(TraceLevel.INFO, "StateHandler close");
  • Minimize the copying of variables.
  • Avoid type conversion (e.g. byte[] to String).
  • Use multiple threads when possible.
  • Only use locks when necessary.

Next steps

This covers the basics of writing a Java primitive operator. To learn more about the details, refer to the knowledge center.

We will continue to improve this development guide. If you have any feedback, please click on the Feedback button at the top and let us know what you think!