Rollup

Short Description

Rollup creates one or more output records from one or more input records.

Rollup receives potentially unsorted data through the single input port, transforms it and creates one or more output records from one or more input records.

The component can send different records to different output ports as specified by the user.

Ports

Metadata

Rollup does not propagate metadata.
Rollup has no metadata template.
Input and output metadata fields can have any data types.
Metadata on output ports can differ.
You may need a metadata for the accumulator record in rollup transformation.

Rollup Attributes

[1] One of these must specified.

Details

Rollup requires transformation. You can define the transformation using CTL (see CTL interface below) or Java (see Java Interface below).

The flow of function calls in a rollup transformation is depicted below. If any optional function (except functions for error handling) is not used, the position of unimplemented function from diagram is skipped.

If you do not define Group accumulator metadata, VoidMetadata is used in transformation functions.

CTL Interface

The transformation uses a CTL template for Rollup, implement a RecordRollup interface or inherit from a DataRecordRollup superclass. Below is a list of RecordRollup interface methods. For detailed information about this interface, see Java Interface below.

Once you have written your transformation, you can also convert it to Java language code by clicking a corresponding button at the upper right corner of the tab.

You can open the transformation definition as another tab of the graph (in addition to the Graph and Source tabs of Graph Editor) by clicking a corresponding button at the upper right corner of the tab.

CTL Templates for Rollup

function void initGroup(
      companyCustomers groupAccumulator) {
   groupAccumulator.count = 0;
   groupAccumulator.totalFreight = 0;
}

function boolean updateGroup(companyCustomers groupAccumulator) {
   groupAccumulator.count++;
   groupAccumulator.totalFreight = groupAccumulator.totalFreight + $in.0.Freight;
   return true;
}

function boolean finishGroup(companyCustomers groupAccumulator) {
   groupAccumulator.avgFreight = groupAccumulator.totalFreight / groupAccumulator.count;
   return true;
}

function integer updateTransform(integer counter, companyCustomers groupAccumulator) {
   if (counter >= Length) {
      clear(customers);
      return SKIP;
   }
   $out.0.customers = customers[counter];
   $out.0.EmployeeID = $in.0.EmployeeID;
   return ALL;
}

function integer transform(integer counter, companyCustomers groupAccumulator) {
   if (counter > 0) return SKIP;
   $out.0.ShipCountry = $in.0.ShipCountry;
   $out.0.Count = groupAccumulator.count;
   $out.0.AvgFreight = groupAccumulator.avgFreight;
   return ALL
}

function void initGroupOnError(
        string errorMessage,
        string stackTrace,
        companyCustomers groupAccumulator)
   printErr(errorMessage);
}

function boolean updateGroupOnError(string errorMessage, string stackTrace, companyCustomers groupAccumulator) {
   printErr(errorMessage);
   return true;
}

function boolean finishGroupOnError(string errorMessage, string stackTrace, <metadata name> groupAccumulator) {
   printErr(errorMessage);
   return true;
}

function integer updateTransformOnError(string errorMessage, string stackTrace, integer counter, companyCustomers groupAccumulator) {
   if (counter >= 0) {
      return SKIP;
   }
   printErr(errorMessage);
   return ALL;
}

function integer transformOnError(string errorMessage, string stackTrace, integer counter, companyCustomers groupAccumulator) {
   if (counter >= 0) {
      return SKIP;
   }
   printErr(errorMessage);
   return ALL;
}

Access to input and output fields

All of the other CTL template functions allow to access neither inputs nor outputs or groupAccumulator.

Input records or fields
Input records or fields are accessible within the initGroup(), updateGroup(), finishGroup(), initGroupOnError(), updateGroupOnError() and finishGroupOnError() functions.

They are also accessible within the updateTransform(), transform(), updateTansformOnError() and transformOnError() functions.

Output records or fields
Output records or fields are accessible within the updateTransform(), transform(), updateTansformOnError() and transformOnError() functions.

Group accumulator
Group accumulator is accessible within the initGroup(), updateGroup(), finishGroup(), initGroupOnError(), updateGroupOnError() and finishGroupOnError() functions.

It is also accessible within the updateTransform(), transform(), updateTansformOnError() and transformOnError() functions.

Capt. Eddie to ground control:

Remember that if you do not hold these rules, NPE will be thrown.

Java Interface

The transformation implements methods of the RecordRollup interface and inherits other common methods from the Transform interface. See Common Java Interfaces.
Following is the list of the RecordRollup interface methods:

• void init(Properties parameters, DataRecordMetadata inputMetadata, DataRecordMetadata accumulatorMetadata, DataRecordMetadata[] outputMetadata)
  Initializes the rollup transform. This method is called only once at the beginning of the life-cycle of the rollup transform. Any internal allocation/initialization code should be placed here.
• void initGroup(DataRecord inputRecord, DataRecord groupAccumulator)
  This method is called for the first data record in a group. Any initialization of the group accumulator should be placed here.
• void initGroupOnError(Exception exception, DataRecord inputRecord, DataRecord groupAccumulator)
  This method is called for the first data record in a group. Any initialization of the group "accumulator" should be placed here. Called only if initGroup(DataRecord, DataRecord) throws an exception.
• boolean updateGroup(DataRecord inputRecord, DataRecord groupAccumulator)
  This method is called for each data record (including the first one as well as the last one) in a group in order to update the group accumulator.
• boolean updateGroupOnError(Exception exception, DataRecord inputRecord, DataRecord groupAccumulator)
  This method is called for each data record (including the first one as well as the last one) in a group in order to update the group accumulator. Called only if updateGroup(DataRecord, DataRecord) throws an exception.
• boolean finishGroup(DataRecord inputRecord, DataRecord groupAccumulator)
  This method is called for the last data record in a group in order to finish the group processing.
• boolean finishGroupOnError(Exception exception, DataRecord inputRecord, DataRecord groupAccumulator)
  This method is called for the last data record in a group in order to finish the group processing. Called only if finishGroup(DataRecord, DataRecord) throws an exception.
• int updateTransform(int counter, DataRecord inputRecord, DataRecord groupAccumulator, DataRecord[] outputRecords)
  This method is used to generate output data records based on the input data record and the contents of the group accumulator (if it was requested). The output data record will be sent to the output when this method finishes. This method is called whenever the boolean updateGroup(DataRecord, DataRecord) method returns true. The counter argument is the number of previous calls to this method for the current group update. See Return Values of Transformations for detailed information about return values and their meaning.
• int updateTransformOnError(Exception exception, int counter, DataRecord inputRecord, DataRecord groupAccumulator, DataRecord[] outputRecords)
  This method is used to generate output data records based on the input data record and the contents of the group accumulator (if it was requested). Called only if updateTransform(int, DataRecord, DataRecord) throws an exception.
• int transform(int counter, DataRecord inputRecord, DataRecord groupAccumulator, DataRecord[] outputRecords)
  This method is used to generate output data records based on the input data record and the contents of the group accumulator (if it was requested). The output data record will be sent to the output when this method finishes. This method is called whenever the boolean finishGroup(DataRecord, DataRecord) method returns true. The counter argument is the number of previous calls to this method for the current group. See Return Values of Transformations for detailed information about return values and their meaning.
• int transformOnError(Exception exception, int counter, DataRecord inputRecord, DataRecord groupAccumulator, DataRecord[] outputRecords)
  This method is used to generate output data records based on the input data record and the contents of the group accumulator (if it was requested). Called only if transform(int, DataRecord, DataRecord) throws an exception.

Examples

Merging and updating incomplete records

You have a list of records containing name, email address and phone number. Records do not have all fields filled in. Records are sorted according to the field name.

Merge together data of records with the same name. If more records with the same name have the same field filled in, use the last one.

Alice|[email protected]|
Alice|                 |+420123456789
Alice|[email protected]|
Bob  |                 |+421212345678
Bob  |[email protected] |
Eve  |[email protected]  |+420720123456
Eve  |                 |+420720123457

Solution
Input and output metadata (updateRecord) have fields name, email and phoneNumber.

Use the attributes Group key, Group accumulator and Transform of Rollup.

//#CTL2

function void initGroup(updateRecord groupAccumulator) {
 	groupAccumulator.* = $in.0.*;
 	return;
}

function boolean updateGroup(updateRecord groupAccumulator) {
 	if (!isnull($in.0.email))
 	{
 	 	groupAccumulator.email = $in.0.email;
 	}

 	if (!isnull($in.0.phoneNumber))
 	{
 	 	groupAccumulator.phoneNumber = $in.0.phoneNumber;
 	}

 	return false;
}

function boolean finishGroup(updateRecord groupAccumulator) {
 	return true;
}

function integer updateTransform(integer counter, updateRecord groupAccumulator) {
 	raiseError("Function not implemented!");
}

function integer transform(integer counter, updateRecord groupAccumulator) {
 	if ( counter > 0 )
 	{
 	 	return SKIP;
 	}

 	$out.0.* = groupAccumulator.*;
 	return ALL;
}

Result records contains merged field values:

Alice|[email protected]|+420123456789
Bob  |[email protected] |+421212345678
Eve  |[email protected]  |+420720123457

Transforming multivalue fields to multiple records

Input records containing name, group and email have a multivalue field email. Split input stream to two data streams: one with name and group, the other one with name and email. The output records will be loaded into a database without support of multivalue fields.

Jane Green |users |[[email protected], [email protected], [email protected]]
John Smith |users |[[email protected], [email protected]]
Peter Green|users |[[email protected]]

The field name of an input record is unique.

Solution
Input metadata users has fields name, group and email. Output metadata users2 has fields name and group, output metadata emails has fields name and email.

Use the Rollup attributes Group key, Group accumulator and Transform.

//#CTL2

function void initGroup(users groupAccumulator) {
 	return;
}

function boolean updateGroup(users groupAccumulator) {
 	groupAccumulator.* = $in.0.*;
 	return true;
}

function boolean finishGroup(users groupAccumulator) {
 	return true;
}

function integer updateTransform(integer counter, users groupAccumulator) {
 	if(counter >= length(groupAccumulator.email )) {
 	 	return SKIP;
 	}
 	$out.1.name = $in.0.name;
 	$out.1.email = groupAccumulator.email[counter];
 	return 1;
}

function integer transform(integer counter, users groupAccumulator) {
 	if(counter > 0 )
 	{
 	 	return SKIP;
 	}
 	$out.0.name = $in.0.name;
 	$out.0.group = $in.0.group;
 	return 0;
}

The transformation above requires the field user to be unique.

You receive 3 records on the first output port:

Jane Green |users
John Smith |users
Peter Green|users

Six records will be send to second output port:

Jane Green |[email protected]
Jane Green |[email protected]
Jane Green |[email protected]
John Smith |[email protected]
John Smith |[email protected]
Peter Green|[email protected]

Best Practices

To process a large number of records, sort records first and than use Rollup instead of using Rollup with the Sorted input attribute set to false.

If the transformation is specified in an external file (with Transform URL), we recommend users to explicitly specify Transform source charset.

See also

Denormalizer
Normalizer
Common Properties of Components
Specific attribute types
Common Properties of Transformers