Over a million developers have joined DZone.

A framework and code generator for XML Big Data

DZone's Guide to

A framework and code generator for XML Big Data

· Big Data Zone
Free Resource

Learn how you can maximize big data in the cloud with Apache Hadoop. Download this eBook now. Brought to you in partnership with Hortonworks.


Dealing with very large XML documents with complex schemata is difficult. On the one hand
because of the sheer size of the data using available technologies that are based on making
an in memory representation of the data is not straight forward, on the other when the
schema is complex, dealing with alternative technologies such as SAX or StAX is very labour
intensive and leads to software that is hard to test (inconsistencies) and maintain. In this
article we present an approach based on MDE (model driven engineering) and the SAX 2.0
API. Our design is centered on a common framework complemented by code engineering
that directly maps the underlying model (XML Schema) to Java. This approach has the
following advantages:

  • Time to market is dramatically decreased
  • Decreased cost
  • It is scalable to very large XML files (low size independent memory footprint)
  • It fully eliminates errors related to consistency
  • It presents data in a convenient JavaBeans format fully consistent with the XML Schema
  • The resulting code is highly efficient and modular
  • The resulting software is dependable
  • It is very easy to implement changes in the underlying model


Now that application development is being outsourced often applications are built in isolated
projects on an as needed basis. The problem with this approach is increased long-term risk.
With a solid foundation of common components, software becomes reusable, with the
following benefits:

  • Decreased time to market for new products
  • Decreased cost
  • Increased maintainability of code base
  • Increased dependability due to reuse

Building applications from scratch involves a tremendous effort from both offshore partners
and local staff. Often too much time is spent on technology instead of business functionality,
even when development is outsourced. One of the major reasons for this is an opportunistic
approach to software development; too often the software development process is immature resulting in poor reuse and maintainability, building similar functionality over and over again in
slightly different ways; potentially common technical components are simply rebuilt as part of
different projects.

Conflict of Requirements

Evidently the sizing requirement constrains our design. What it boils down to is:
We cannot use a model that makes an internal representation of all of the data in the
document, but on the other hand, why care for parsing the document if we are not to do
something useful with it? In other words, we’d like to have the data represented in a
convenient format (JavaBeans) and be able to process it from there without having to bother
with the technicalities of handling the XML document or the intricacies of the schema in hand.
What we have here is conflict of requirements.
On the one end, there are quite a few technological choices that allow us to make an internal
model, so we can work from the convenience of a Java model that makes sense from a
semantic point of view. On the other end we find models that throw events at us (e.g. SAX
API), or alternatively allow us to walk-through the document using either a cursor or event
model (StAX). Neither of these models really fits because they force one of the extremes
upon us. We need a way to walk the middle ground. Although lately it has become possible to
use JAXB on top of StAX allowing to partially map XML, this approach requires us to mix
low-level with high-level programming and still requires some serious design effort in order to
make our design modular and extensible.

Making a modular design

So we want our design to be modular, high-performance, intuitive, maintainable and easy to
use at the same time. We are going to use either the StAX or SAX API but we don’t want to
clutter our design with extensive control structures (making our design unreadable, poorly
maintainable and hard to test). Although the StAX API may seem easier to handle than the
SAX API, because of the pull architecture, in fact when dealing with complex XML Schemata,
it is of not much help and additionally, you may be in a situation making it hard for you to use
because of your technical environment. The SAX push architecture is fast and has low
memory footprint, there’s a large choice of implementations for this API available, but it
maybe somewhat intimidating for developers new to publish-subscribe architectures.
The important challenge clearly is: How to make a modular and maintainable design that
suits our technical requirements and is easy to use at the same time?
An important clue
can be found in the fact that the SAX 2.0 API allows us to swap content handlers on the fly as
often as we like. Robert Hustead wrote an excellent article on the matter. In
this article Robert explains how to create a modular design that eliminates complex control
structures by swapping content handlers.

Dealing with large and complex XML Schemata

In our XML parser framework we use a similar pattern that among others involves creating a
data (JavaBean) and a handler (content handler) class per schema entity (complextype) and
some other mostly private classes that glue it all together. This makes our design very
modular, but when dealing with large and complex XML schemata (such as the SEPA PAIN/PACS schemata), implementing all the JavaBeans and handler classes (or trackers as
in Robert’s design) and putting it all together by hand is still a very time-consuming and error
prone task, if not undoable. To solve this problem we use MDE (model driven engineering) to
create all of these classes directly from the model (the ISO-20022 XSDs in this case). This
way we have the benefits of efficiency, modularity and at the same time eliminate the
inconsistencies that result from the manual process of creating the classes. To give you an
idea, for SEPA/Payments alone, the number of data and handler classes exceeds 250 (the
total number of generated classes actually exceeds 1250). All of these extend the framework,
which is the engine that orchestrates it all.

Dealing with memory usage and implementing business rules

But, let’s go back to our conflict of requirements now. Remember the issue that brought us
here is dealing with our technical requirements effectively. To get the picture I’ll use an
example taken from the SEPA payments domain (ISO-20022: PAIN.001.001.03). The basic
structure of the XML file is as follows: It comprises of a header, which contains general
information about the message, followed by one or more batches, each of which may contain
many transactions. It’s the transactions that we expect many of, but according to the schema
the number of batches is also unbounded.

Type of data XML element Path
Header GrpHdr Document/CstmrCdtTrfInitn/GrpHdr
Batch PmtInf Document/CstmrCdtTrfInitn/PmtInf
Transaction CdtTrfTxInf Document/CstmrCdtTrfInitn/PmtInf/CdtTrfTxInf

To complete the picture we also have to add the types (schema types) each of these XML
elements is an instantiation of:

Element Type
GrpHdr GroupHeader32
PmtInf PaymentInstructionInformation3
CdtTrfTxInf CreditTransferTransactionInformation10

When you examine these types (PAIN.001.001.03, see: [2]), you’ll find that for instance
CreditTransferTransactionInformation10 is pretty complex. We definitely need to have a
convenient representation in Java allowing us to access its data in an intuitive way. The same
it true for the PaymentInstructionInformation3 and GroupHeader32. Now we begin to fully
understand the problem: We have to deal with complex types, but on the other hand need a
way to prevent having to store all contained objects within their parent objects if we expect a
great many of them. This is why we made the framework’s behaviour configurable through
properties: The property @process instructs the framework to send a JavaBean containing
the element in scope to the registered processor component, the property @detach tells the
framework it should not store the element inside its parent’s container, thus preserving
memory (making it short-lived).
The following picture shows the runtime configuration (properties) file for a parser that
handles PAIN.001.001.03 messages:

In plain English this configuration tells the framework to:

  • Process the GrpHdr element (there is only one)
  • Process each PmtInf element and not store it in the parent container
  • Process each CdtTrfTxInf and not store it in the parent container

At this time you may be wondering what happens to all the other XML elements: The nice
thing is, the property file overrides the defaults, which are:

@process = false and @detach = false

In other words, the other items (not being detached) are simply available through getters at
their parent level. For clarity I’ll show you a snippet (truncated) of the generated JavaDoc for

Here you see the getters for its contained elements. The use of java.util.list is consistent with the multiplicity for InstrForCdtrAgts defined in the schema. You can see the complete JavaDoc listing of the PAIN message (2009) here.

The Processor Component

Now about the business: Thus far we only talked about the framework and the Reader
Component. What we want to do with the data is implemented by the provided Processor
Component (APPLICATION PROCESSOR). The application programmer will create a specific
by implementing the required interface (DataProcessor) specified by the

The Reader Component delivers the JavaBeans to the Processor Component in accordance
with the instructions in the properties file. Following our example the processor will receive
every instance of the GrpHdr, the PmtInf complete with all of its contained elements (except
for the detached CdtTrfTxInf items) and the CdtTrfTxInf, with all of its contained elements. For flexibility the framework sends two notifications to the processor per element,
one when it first encounters the element in the XML document and the second when it
encounters its closing tag. This gives the processor maximum flexibility. Let’s have a look at
the definition of PaymentInstructionInformation3:

It immediately occurs to us it is defined as a sequence. We also find that the contained
transactions are last in the sequence. Consequently we know that when we encounter the
first transaction, the batch level information is complete and thus can be processed. Of course
this can be very convenient if for instance we want to store the data in a database (since it
enables us setting the FK relationship with the contained transactions).

Application Development with the Framework

With the LDX+ framework the team can focus entirely on the design and development of the
Processor Component. The design of the processor component is not constrained by the Reader Component; it’s up to the team to decide whether they want to implement the
interface DataProcessor synchronously or asynchronously.

There are a couple of things worth mentioning:

  • The implementation of the org.xml.sax.ErrorHandler interface, which is good practice allowing your application to respond to error conditions during parsing. This class provides an excellent place to hookup Logging.
  • There is a class called Pain001V03MessageHandler, which is the entry point for applications dealing with pain.001.001.03 type of messages. This class is generated and available as part of the framework.
  • The ProcessorException is part of the Processor Component interface. It is in fact the only exception that is allowed to leak through the interface and should used to report error conditions within the Processor Component that are unrecoverable (e.g. you may want to abort processing upon some special condition).

Now let’s have a look at the application main module. This module is optionally generated by the code generator, just requiring you to make some changes to adapt it to your case (for example add logging). In this sample we’ve used the command line arguments to get the runtime configuration file (args[1]), the XML Schema (args[2]) and the XML document (args[0]) . The inline comments
should suffice to understand the code.


When looking at validation, there are a couple of cases that we need to distinguish:
  • Reject XML message (the complete XML) when validation fails
  • Reject individual transaction when not valid but process the valid ones

In the first case we can use pre-validation as shown in the sample code. When the document
is not valid (according to the provided XML Schema) a SAXException is thrown and the
application may report it and exit.
In the latter case the XML message is partly valid but some mandatory information may be
missing. In this case pre-validation may be skipped or a validation against a relaxed schema
(defining minimum requirements) may be done. It is then up to the Processor Component to
decide what to do with individual transactions that are incomplete (e.g. log and skip).


The LDX+ Parser Framework for Java dramatically reduces development time. In fact the team
can focus entirely on the design and development of the business.
For handling large XML documents with complex schemata we clearly need a technology that
automatically maps the XML data into Java allowing us to use a convenient model but without
the overhead of loading the complete document into memory. The framework we presented
here has shown to handle large transaction files with over 1,000,000 transactions with ease
whilst providing the convenience of an internal object representation in Java. Its runtime
configurability (per application instance) allows tuning memory utilization and enabling
processing per type of element. The MDE approach automatically generates the Java classes
allowing the application programmer to access the data in a convenient and consistent
manner and enforces consistency with the model (the XML Schema).

For more information and to obtain a fully functional evaluation copy, please visit: Dijkstra ICT Consulting


Hortonworks DataFlow is an integrated platform that makes data ingestion fast, easy, and secure. Download the white paper now.  Brought to you in partnership with Hortonworks


Opinions expressed by DZone contributors are their own.

The best of DZone straight to your inbox.

Please provide a valid email address.

Thanks for subscribing!

Awesome! Check your inbox to verify your email so you can start receiving the latest in tech news and resources.

{{ parent.title || parent.header.title}}

{{ parent.tldr }}

{{ parent.urlSource.name }}