Difference between revisions of "SMILA/Documentation/HowTo/Howto integrate a component in SMILA"

Revision as of 04:55, 30 September 2008

This page summarizes the different types and complexity levels for the integration of components in SMILA.

Introduction

Due to its architecture SMILA allows for the easy integration of third-party components into its framework. Actually there are three different possible integration scenarios available that are depicted in the following table.

Integrating services in BPEL	Integrating agents and crawlers	Integrating alternative implementations for SMILA core components
This is propably the most frequently used integration scenario. It allows for the integration or exchange of functionality (services, 3rd party software, etc.) used to process records in the workflow engine.	Agents and crawlers unlock new data sources, providing additional input to SMILA	Offers experienced developers the potential to exchange existing implementations of SMILA core components with their own implementations.
The above figures exemplary demonstrate at which levels in the SMILA architecture an integration of new components is applicable. However, for simplicity reason, we restricted the above figures to the index processing chain while completely ignoring the search processing chain that also offers similar integration options but is currently not in the focus of this page.

Integrating services in BPEL

There are several options on how to integrate new functionality in SMILA BPEL workflows. In SMILA we call these workflows pipelines.

• Simple: as the standard BPEL functionality to invoke webservices. Not all data of SMILA Records is usable in this option.
• Default: the recommended way to integrate additional functionality in SMILA. Two interfaces allow for easy Java implementations of so called Pipelets and ProcessingServices.
• Advanced: an advanced alternative for integrating ProcessingServices that do not run in the same OSGi runtime as the BPEL workflow but in another OSGI runtime that may even run on a remote machine.

Simple: Integrating web services

The simplest way of integrating additional functionality in SMILA is to call a webservice. This is a standard BPEL workflow engine functionality independent of SMILA. However, there are some limitations concerning the input and result data to/from webservices: In SMILA the workflow object (a DOM object) that enters the BPEL workflow contains by default only the Record IDs. Records and the data contained therin are NOT accessible from a BPEL workflow! The BPEL workflow can only access and use the values contained in the BPEL workflow object. It is possible to change this behaviour and add additional data to the workflow object by configuring filters in the configuration file org.eclipse.smila.blackboard/RecordFilters.xml. There you can select certain Attributes and Annotations that will be copied to the workflow object and so will be accessible by the BPEL workflow. Attachments are currently NOT supported, as binary data is not reasonable in DOM! Note that you also need to include all Attributes and Annotations in the RecordFilters.xml you want to write data to.

Examples:

• A good example for this use case is the integration of [Language Weaver]. The Language Weaver Translation Server provides a webservice interface that allows a text to be translated into another language. This service could be easily used within SMILA.

Here are more detailed technical descriptions:

• How to filter and access record data in BPEL
• How to integrate the HelloWorld webservice in BPEL

Default: Integrating local SMILA pipelets or processing services

The default technique to integrate functionality or software in SMILA is to write a Pipelet or ProcessingService that runs in the same OSGi runtime as the BPEL workflow engine. Pipelets are easier to implement than ProcessingServices, as they require only standard Java knowledge. For more information about Pipelets and ProcessingServices see Pipelets and ProcessingServices. Both Pipelets and ProcessingServices have full access to Records in SMILA via the BlackboardService. So it's easily possible to read, modify and store Records. In general Pipelets and ProcessingServices follow the same (sometimes optional) logical steps (of course this depends highly on the business logic to be executed). These steps are:

• read the configuration (optional)
• read input data from Blackboard (optional)
• execute the business logic
• write result data to Blackboard (optional)

In the part of your Pipelet/Processing service that implements the business logic you are totally free to use any desired technology. Some of the posibilities include

• use of POJOs (e.g. see the various XML Processing Pipelets)
• use any local available OSGi service, even other ProcessingServices (e.g. see the AperturePipelet that uses the ApertureMimeTypeIdentifier)
• use other technologies like JNI, RMI, Corba, etc. to integrate remote or non Java components (e.g. integration of Oracle Outside In Technology)
• ...

Examples:

• Typical examples for Pipelets are the XML Processing Pipelets. These lightweight Pipelets are used for XML processing (e.g. XSL transformation). Each Pipeline uses it's own Pipelet instance.
• A good example for a ProcessingService is the LuceneIndexService. It provides functionality to index Records in Lucene indexes and can be used from multiple Pipelines in parallel.

Here are more detailed technical descriptions:

• How to write a Pipelet
• How to write a ProcessingService
• How to integrate the HelloWorld webservice as a Pipelet

Advanced: Integrating remote SMILA processing services

tbd.

Integrating agents and crawlers

SMILA's connectivity framework allows easy integration of additional datasources by providing implementations of Agents and/or Crawlers.

• How to implement an agent
• How to implement a crawler

Integrating alternative implementations for SMILA core components

SMILA's component based architecture even allows you to provide your own implementations of SMILA core components. More info comming soon ...

Examples:

• a typical example is an alternative implementation of the DeltaIndexingManager that does not store it's state in memory but in the filesystem or in a database