This document describes the steps to create a new user defined operator with ODL.

Name

First you need to choose a meaningful name for the new operator. Note, there is no other operator with the same name.

operator ODLSelect {
	//...
}

Metadata

Next you can set metadata to configure the operator with static information.

operator ODLSelect(outputMode = "INPUT", minInputPorts = 1, maxInputPorts = 1){
    //...
}

The following table shows the available metadata:

Please look at the LogicalOperator-Interface of Odysseus for more available metadata (e.g. doc, url, category).

Parameter

You need to define parameters if the operator should be configurable while building a continuous query. A parameter can be optional and consists of a data type and a name.

operator ODLSelect(outputMode = "INPUT", minInputPorts = 1, maxInputPorts = 1){   

    parameter IPredicate predicate;

    optional parameter int heartbeatrate;
	
	//...
}

Note that not every data type can be used for a parameter because each data type must have a corresponding parameter class in Odysseus. The following table shows the available data types:

Some data types have more than one corresponding parameter class. In these cases you can optionally set a parameter class as a metadata of a parameter. Please look at the Parameter-Interface of Odysseus for more available metadata.

operator ODLProject(outputMode ="MODIFIED_INPUT", minInputPorts = 1, maxInputPorts = 1){    
    
    parameter(type=ResolvedSDFAttributeParameter) SDFAttribute[] attributes;
	
	//...
}

Attributes

Besides parameters it is also possible to define attributes to store operator’s data.

operator ODLJoin(outputMode = "NEW_ELEMENT", minInputPorts = 2, maxInputPorts = 2){    
    
    ITimeIntervalSweepArea[] areas;     
    IDataMergeFunction dataMerge;
    ITransferArea transferFunction;
    IMetadataMergeFunction metadataMerge;
	
	//...
}

Validate methods

Each parameter can have a validate method. This method must return whether the value of the parameter is correct.

operator ODLProject(outputMode ="MODIFIED_INPUT", minInputPorts = 1, maxInputPorts = 1){    
    
     validate attributes {
        if (this.attributes.isEmpty) {
            error("attributes have to be set");
            return false;
        }
        return true;
    }
	
	//...
}

There is also a validate method which can be used if a parameter has dependencies to other parameters.

operator ODLProject(outputMode ="MODIFIED_INPUT", minInputPorts = 1, maxInputPorts = 1){    
    
     validate {
        return true;
    }
	
	//...
}

Event methods

These methods are automatically called when events occur. The processNext-method is the most important event method and is called whenever a new stream element is received. The type of a stream element can be Tuple, KeyValueObject or other subtypes of IStreamObject.

operator ODLSelect(outputMode = "INPUT", minInputPorts = 1, maxInputPorts = 1){   
    
    on processNext(Tuple tuple, int port) {
        if (predicate.evaluate(tuple)) {
            sendStreamElement(tuple);
        }
    }

    on processPunctuation(IPunctuation punctuation, int port) {
        sendPunctuation(punctuation);        
    } 

	//...
}

The following table shows an overview of available event methods. Note that you can call methods of AbstractLogicalOperator or AbstractPipe in event methods, e.g. to send stream elements or punctuations.

AO & PO methods

Besides validate and event methods it also possible to create norma’ methods. The ao-keyword should be used if a method needs to call a method of AbstractLogicalOperator. The po-keyword should be used if a method needs to call a method of AbstractPipe.

getOtherPort(int port) : int {
 return (port+1)%2;
 }
 
 ao getBaseTimeUnit : TimeUnit {
 SDFSchema schema = getInputSchema(0);
 SDFConstraint constraint = schema.getConstraint(SDFConstraint::BASE_TIME_UNIT);
 return constraint.value;
 }
 
 po pint(IPunctuation punctuation) {
 sendPunctuation(punctuation);
 }