WADE USER GUIDE

WA D E U S E R G U I D E

USAGE RESTRICTED ACCORDING TO LICENSE AGREEMENT.

Last update: 5-December-2013 - WADE 3.3

Authors: Giovanni Caire (Telecom Italia S.p.A)

Copyright (C) 2010 Telecom Italia

Copyright (C) 2011 Telecom Italia

WADE (Workflows and Agents Development framework) is a software platform that facilitates the
development of distributed multi agent applications where agent tasks can be defined according to
the workflow metaphor. From a different point of view WADE can be seen as an agent oriented
Workflow Engine suitable to be used for business process orchestration.
This document gives an overview of the WADE platform, presents its architecture and main
functionalities, and outlines the conceptual model underlying it. Two major aspects of the platform
are described: the execution model based on the workflow metaphor and the configuration and
control features designed to hide the complexity of the distribution. WADE is based on JADE, a
popular open source agent-oriented middleware and inherits from it a distributed topology with
peer-to-peer networking and a software component architecture based on the agent paradigm.

The WADE library is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation, version 2.1 of the License.
This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty
of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
details.
You should have received a copy of the GNU Lesser General Public License along with this library; if not, write to the Free Software
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.

Java and HotJava are trademarks of Sun Microsystems, Inc., and refer to Sun's Java programming language and HotJava browser
technologies. The Workflows and Agents DEvelopment Framework WADE is not sponsored by or affiliated with SUN
Microsystems, Inc."

1
WADE USER GUIDE

TABLE OF CONTENTS

1 INTRODUCTION .......................................................................................................... 4

1.1 Workflows ............................................................................................................................... 4

1.2 Scope........................................................................................................................................ 5

1.3 Approach................................................................................................................................. 6

1.4 Managing the complexity of the distribution ...................................................................... 7

1.5 The WADE Suite .................................................................................................................... 8

1.6 Download and Installation .................................................................................................... 9

2 WADE-BASED APPLICATIONS ............................................................................... 10

2.1 Architecture .......................................................................................................................... 10

2.2 Application activation and termination ............................................................................. 11

2.3 Getting started with the platform ....................................................................................... 13

2.3.1 WADE Distribution structure overview ............................................................................ 13
2.3.2 Starting up the platform ..................................................................................................... 15
2.3.3 The Management Console ................................................................................................. 15

2.4 Agent types and roles ........................................................................................................... 17

2.4.1 Using environment variables ............................................................................................. 18

2.5 WADE Agents ...................................................................................................................... 19

2.6 Running WADE-based applications .................................................................................. 20

2.7 Hot deployment of workflows ............................................................................................. 22

3 WORKFLOWS ........................................................................................................... 23

3.1 The workflow meta-model................................................................................................... 23

3.2 Workflow elements implementation................................................................................... 25

3.2.1 The WorkflowBehaviour class........................................................................................... 25
3.2.2 Data Fields and Formal Parameters ................................................................................... 27
3.2.3 BuildingBlock invocation .................................................................................................. 27
3.2.4 Layout information ............................................................................................................ 28

3.3 Synchronization Activities ................................................................................................... 29

3.3.1 Events ................................................................................................................................. 29
3.3.2 Incoming Web Services ..................................................................................................... 30
3.3.3 Long-Running workflows .................................................................................................. 32
3.3.4 The Persistence Add-On .................................................................................................... 33

2
WADE USER GUIDE

3.4 Workflow inheritance .......................................................................................................... 33

4 EXECUTING WORKFLOWS ..................................................................................... 35

4.1 Launching workflows from code: the DispatchingCapabilities class .............................. 36

4.2 Using the LauncherAgent to launch workflows ................................................................ 37

5 ADVANCED FEATURES ........................................................................................... 39

5.1 Agent Pools ........................................................................................................................... 39

5.1.1 Agent allocation policies .................................................................................................... 40

5.2 Agent and containers fault tolerance ................................................................................. 40

3
WADE USER GUIDE

1 Introduction
WADE (Workflows and Agents Development Environment) is a domain independent software
platform built on top of JADE (http://jade.tilab.com), a popular open source middleware conceived
to facilitate the development of distributed applications based on the agent-oriented paradigm.
Readers that are not familiar with JADE, are encouraged to look at the JADE Administration
Tutorial (http://jade.tilab.com/doc/tutorials/JADEAdmin/index.html) and, if they have programming
skills, at the JADE Programming Tutorial for Beginners
(http://jade.tilab.com/doc/tutorials/JADEProgramming-Tutorial-for-beginners.pdf) before going
through the advanced sections of this document.
As depicted in Figure 1, JADE provides a distributed runtime environment, the agent and behaviour
(a task performed by an agent) abstractions, peer to peer communication between agents and basic
agent lifecycle management and discovery mechanisms. WADE adds to JADE the support for the
execution of tasks defined according to the workflow metaphor and a number of mechanisms and
components that help managing the complexity of the distribution both in terms of administration
and fault management.

In principle WADE supports “notepad-programming” in the sense that there is no hidden stuff that
developers can’t control. However, especially considering that one of the main advantages of the
workflow approach is the possibility of representing processes in a friendly graphical form, WADE
comes with a development environment called WOLF that facilitates the creation of WADE-based
application. WOLF is an Eclipse (www.eclipse.org) plug-in and as a consequence allows WADE
developers to exploit the full power of the Eclipse IDE plus additional WADE-specific features.
This document does not discuss WOLF. Interested readers should refer to the WADE Development
Tutorial (http://jade.tilab.com/wade/doc/tutorial/WADE-Tutorial.pdf) for details about how to
install and use it.

- Application specific
Application features
WOLF
- Workflow development
WADE - Administration and fault
Eclipse management
- Distributed runtime
- Agent and behaviour
JADE - Communication
- Discovery

Figure 1. The WADE platform

1.1 Workflows
A workflow is the formal definition of a process in terms of activities to be executed, relations
between them, criteria that specify the activation and termination and additional information such as

4
WADE USER GUIDE

the participants, the software tools to be invoked, required inputs and expected outputs and internal
data manipulated during the execution.
The key aspect of the workflow metaphor is the fact that the execution steps as well as their
sequencing are made explicit. This makes it possible to give a graphical representation of a
process defined as a workflow. Such representation is clearly extremely more intuitive with respect
to a piece of software code and in general is understandable by domain experts as well as by
programmers. Domain experts can therefore validate system logics directly and not only on
documents that most of the time are not perfectly up to date. In some cases they could even
contribute to the actual development of the system without the need for programming skill.
Another important characteristic is that, being the execution steps explicitly identified, the
workflow engine (i.e. a system able to automatically execute a process defined as a workflow) can
trace them. This makes it possible to create automatic mechanisms to facilitate system monitoring.
Typically a workflow engine embeds ready-made mechanisms to trace activities and to create
reports on them. Additionally, when processes have to be executed within the scope of a
transaction, semi-automatic rollback procedures can be activated in case of unexpected fault.
Finally, since workflows are fully self-documented, workflow-based development releases the
development team of the burden of keeping documentation aligned each time design choices must
be revisited to face implementation details or evolving requirements.

1.2 Scope
If on the one hand the workflow metaphor has a number of advantages as discussed in previous
section, it is in general not suitable to deal with low level operations such as data management and
transformation, mathematical computations and so on. Software code is typically more powerful
and efficient to implement them. As a consequence, nowadays the workflow metaphor is mostly
used in BPM contexts where a workflow represents a business process and orchestrates a number of
existing systems typically (but not necessarily) accessible by means of Web Services-based
interfaces.
WADE tries to break this barrier and brings the workflow approach down to the level of system
internal logics too. Therefore, as depicted in Figure 2, it can be fruitfully adopted both as an
Orchestrator to coordinate existing systems and as a Software Development Framework to create
brand new applications that imply the execution of possibly long and complex tasks.

5
WADE USER GUIDE

Orchestration context System development context

WADE is used as a Workflow Engine and workflows WADE is used as an agent-oriented Development
implement processes that coordinate different systems. Framework and workflows implement some of the
WADE is seen as a “black box” that executes agent tasks. In some cases workflows may not even be
workflows. Agents inside WADE are not even seen. used at all.

System 1
System
System 2
WADE
WADE Agent
WADE Agent

System 3 WADE WADE

Agent Agent
System 4

Figure 2 WADE application contexts

Furthermore, considering the distributed nature of the underlying agent-oriented JADE layer,
WADE can provide a good value added whenever scalability is an issue.

1.3 Approach
First of all it should be noticed that WADE does not include a single powerful workflow engine as
the majority of BPM oriented tools do. On the contrary it provides an ad hoc type of Agent
(Workflow-Engine Agent) that embeds a micro-workflow engine. As a consequence, besides
normal tasks (JADE behaviours), each Workflow-Engine agent active in a WADE-based
application is able to execute workflows represented according to a WADE specific formalism.
The second important point to highlight is that, in order to allow developers to exploit the workflow
metaphor to define system internal logics and, at the same time, to give them the same power of a
software programming language and a comparable execution efficiency, the WADE workflow
representation formalism is based on the Java language. That is, a workflow that can be executed
by WADE Workflow-Engine agents is expressed as a Java class with a well defined structure
(detailed in chapter 3). As such WADE workflows can be edited, refactored, debugged and in
general managed as all Java classes and can include all pieces of code (methods, fields of whatever
types, inner classes, references to external classes and so on) needed to implement the process
details. In addition, of course, the execution flow they specify can be presented and modified in a
friendly, graphical way. More in details WOLF (the development environment for WADE based
applications) is an Eclipse plugin and allows developers to work with a graphical view (suitable to
manage the process flow) and a code view (the usual Eclipse Java editor suitable to define
execution details) that are kept in synch.
Therefore the WADE workflow engine embedded into Workflow-Engine agents is not an
interpreter of a workflow description language, but executes compiled Java code. This on the one
hand makes it extremely performant, but on the other hand requires the necessary workflow classes
to be available when an agent is requested to execute a workflow. For this reason WADE uses ad
hoc Java class loaders to allow deploying new/modified workflows that become immediately
executable without the need to turn the system down.

6
WADE USER GUIDE

Another important characteristic of WADE workflows is that they can be extended. That is, it is
possible to create new workflows by extending existing ones and just defining the differences. This
feature will be detailed in section 3.4.
Finally it must be noticed that WADE does not impose that all system logics are defined as
workflows. Developers are free to exploit the workflow metaphor to describe those tasks for which
they think it is appropriate and use normal JADE behaviours (or other purely Java patterns)
elsewhere. In certain cases one could even decide to create a WADE based application that does not
use workflows at all.
How to define and execute workflows is described in details in chapters 3 and 4.

1.4 Managing the complexity of the distribution

WADE inherits from JADE a distributed runtime composted of several “Containers” that can run
on different hosts and can contain each one a number of agents. Even if this is not strictly
mandatory, most of the time a container corresponds to a JVM. The set of active containers is called
a Platform. As depicted in Figure 3 a special container exists in the platform called “Main
Container”. The Main Container must be the first one to start and all other containers (typically
called peripheral containers) register to it at bootstrap time. Furthermore the Main Container holds
two special agents.
The AMS (Agent Management System) that represents the authority in the platform, i.e. it is the
only agent that can activate platform management actions such as creating/killing other agents,
killing containers and shutting down the platform. Normal agents wishing to perform such actions
must request them to the AMS.
The DF (Directory Facilitator) that implements the yellow pages service by means of which agents
can advertise their services and find other agents offering services they need.
Further details on main JADE architectural components can be found in the JADE administration
Tutorial (http://jade.tilab.com/doc/tutorials/JADEAdmin/index.html).

A1

AMS DF

Main container

Is registered
A4 Is registered with A2 A3
with

Container 2
Platform Container 1

Network

Figure 3. JADE distributed architecture

7
WADE USER GUIDE

Distribution is an important characteristic, especially for applications that are expected to support
heavy loads, since a distributed application can be deployed on highly scalable hardware
architectures such as blades. It is clear however that administering a distributed application is more
complex than administering a monolithic application unless proper tools are made available.
Furthermore the probability that a host crashes increases proportionally with the number of hosts
the application is distributed on and proper recovery mechanisms are necessary to ensure the
application can survive. Clustering systems are often used for those purposes, but they are typically
quite expensive and difficult to configure.
WADE faces these problems by providing a number of mechanisms that help the administrator in
 Installing the application
 Configuring it (tuning parameters).
 Activating/deactivating the application spreading components (i.e. containers and agents)
across available hosts according to specific needs.
 Monitoring runtime events and critical conditions such as disk and memory consumption.
 Deploying new/modified system logics at runtime without system down.
 Automatically recovering from host, container and agent faults.
The main mechanisms that WADE provides to hide the complexity of the distribution are described
in chapters 2 and 5 that focus on the administration and fault tolerance aspects.

1.5 The WADE Suite

Besides the actual WADE platform and the Workflow Graphical Editor WOLF, the WADE Suite
includes additional components briefly outlined in this section.
Web Administration Console – This is a Web Application that can be installed in whatever Web
Container such as Apache Tomcat and allows administering and monitoring a Wade-based
application, deploying, launching and controlling workflows, creating reports on executed or
running workflows and investigating problems.
WadeServices – This is a Web Application that can be installed in whatever Web Container such as
Apache Tomcat and allows Wade-based applications to automatically expose Web Services. When
creating workflows it is possible to define steps where the execution flow suspends until a given
Web Service operation is invoked (typically by an external system). This feature is described in
section 3.3.2.
Persistence Add-On – This additional module allows persisting the state of running workflows on a
relational DB and is mandatory to execute long-running workflows, i.e. workflows that survive to a
system shutdown and restart.
Figure 4 depicts a typical scenario for a Wade-based application

8
WADE USER GUIDE

Administrator User
WOLF
Developer
Web Admin Application
Console Console

Install, Start/Stop and Application specific

monitor the System
user interaction
Deploy workflows

Application
specific
Application Specific Components persistent
Wade data
Services
Web Service
Invocations
WADE Persistence
Add-On Workflow
execution
state
System
Figure 4. Wade based application scenario

1.6 Download and Installation

All components of the WADE Suite can be downloaded from the WADE download page of the
WADE web site (http://jade.tilab.com/wade) in both bynary and source form.
In order to install them it is sufficient to unzip the bynary distribution packages as described below.
WADE – Unzip the WADE distribution package somewhere on your disk. The structure that is
produced is described in section 2.3.
Persistence Add-On – Unzip the Persistence Add-On distribution package in the same directory
where WADE was unzipped.
Wolf – As all Eclipse plug-ins the Wolf bynary distribution package must be unzipped in the home
directory of your Eclipse installation.
Web Administration Console – Unzip the Web Administration Console bynary package in the
webapps directory of your servlet container
WadeServices – Unzip the WadeServices bynary package in the webapps directory of your servlet
container

9
WADE USER GUIDE

2 WADE-based applications
2.1 Architecture
This section presents the components that (together with the basic components provided by JADE
and summarized in 1.4) make up the WADE architecture. As depicted in Figure 5, from the
architectural point of view, a WADE-based application appears to be quite similar to a purely JADE
based application. It runs on a set of hosts each one holding one or more JADE containers and, as
usual, the Main Container (hosting the JADE AMS and DF) must be activated first with other
containers registering to it at bootstrap time. Application specific agents are distributed across the
containers according to some criteria that depend on the application requirements. WADE specific
components are the Boot Daemon processes (one per host) each one responsible for activating
containers in the local host, the Configuration Agent (CFA) always running in the Main Container
and responsible for interacting with the boot daemons and controlling the application life cycle and
the Controller Agents (one per container) responsible for supervising activities in the local
container and for all the fault tolerance mechanisms provided by WADE.
Of course the last fundamental component of the WADE platform are the Workflow Engine Agents
i.e. the agents able to execute tasks defined as workflows as will be described in section 3.
Workflow Engine Agents can be used as they are and in certain cases all the logics of an application
can be embedded in workflows that will be executed by “naked” Workflow Engine agents. In a
more general case an application can extend the base WorkflowEngineAgent class in order to
provide domain specific features. How many Workflow-Engine agents to use and where (in which
containers) to deploy them depends on the application requirements. In general some of the
application specific agents will be (possibly extended) Workflow-Engine agents and will therefore
be able to execute workflows, while others will be “normal” agents running usual JADE
behaviours. It should be noticed that a WADE-based application may even not use Workflow
Engine agents at all and just exploit the administration and fault tolerance features. In that case we
end up with a WADE-based application that does not use workflows.

10
WADE USER GUIDE

Boot Daemon Boot Daemon Boot Daemon

CA WSMA
CA CFA RAA AMS DF
Container-4
Container-1 Main-Container

CA ESA

Container-5
CA CA

Container-2 Container-3
CA

Application Container-6
specific agents

HOST1 HOST2 HOST3

Figure 5. The WADE Architecture

Figure 5 also shows another 3 WADE specific agents:

Runtime Allocation Agent (RAA) active on the Main Container. The role of this agent will be
discussed in chapter 5 when presenting Agent Pools and Fault Tolerance.
Workflow Status Manager Agent (WSMA). This agent can be activated on whatever container and
is responsible to trace workflow executions and to persist their state.
Event System Agent (ESA). This agent can be allocated in whatever container and implements the
WADE Event System that allows a workflow to suspend waiting for an event.

In many cases the application may involve other non-agent components such as a Database or a
Web Interface running on the same or on separate hosts. How to manage these components is
outside the scope of this document.

2.2 Application activation and termination

First of all it has to be highlighted that, though a single WADE platform can host several
applications, unlike the J2EE architecture, such applications have dedicated processes. More in
details the Boot Daemons are shared among all applications hosted by a given WADE platform. On
the other hand each application will have its own Main Container and peripheral containers.

When the application is down only the boot daemons (one per host) and the Main Container (with
the JADE AMS, DF and the WADE CFA) are up and running. In this status it is possible to request
the CFA to import an application configuration (ImportConfiguration1 action of the
Configuration Ontology). An application configuration is a file that specifies, according to an XML

1
Look at the Javadoc of the classes included in the com.tilab.wade.cfa.ontology for details about the parameters to be
specified in the actions of the Configuration Ontology

11
WADE USER GUIDE

based format, which hosts are involved, which containers must be executed in each host and which
agents must be activated in each container. The snippet below shows an example of an application
configuration file involving a single host (the local host) with two containers (called
Administration-Node and Execution-Node); the first one holds the Workflow Status Manager Agent
and the Event System Agent while the second one holds two Workflow Engine agents.

<?xml version="1.0" encoding="UTF-8"?>

<platform description="This is a sample configuration" name="Sample">

<hosts>
<host name="localhost">
<containers>
<container name="Administration-Node">
<agents>
<agent name="wsma" type="Workflow Status Manager Agent"/>
<agent name="esa" type="Event System Agent"/>
</agents>
</container>

<container name="Execution-Node">
<agents>
<agent name="performer1" type="Workflow Engine Agent"/>
<agent name="performer2" type="Workflow Engine Agent"/>
</agents>
</container>
</containers>
</host>
</hosts>

</platform>

Each application configuration has a name and an optional description. Application configurations
available for import are kept in files called <configuration-name>.xml and stored by
default in the ./cfg/configuration directory.

For each agent, an application configuration specifies the name, the type (agent types will be
described in section 2.3) and possibly some arguments as will be described in section 2.5.
An application configuration also supports the specification of container profiles. Each
ContainerProfile includes either a set of JADE parameters (type=”JADE”) that tune the
behaviour of the container itself or a set of JAVA parameters (type=”JAVA”) that tune the
behaviour of the JVM hosting the container.
For instance if we want to create a container in a JVM using 512MB of heap we could define in our
application configuration file a container-profile of type JAVA as below

<containerProfile name="Heap512M" type="JAVA">

<properties>
<property key="Xmx512m"/>
</properties>
</containerProfile>

In order to activate the application it is necessary to request the CFA to perform the
StartupPlatform action of the Configuration Ontology. When serving this action the CFA
reads the currently imported application configuration (also called the target configuration) and:
 For each host requests the associated boot daemon to launch the specified containers (with
the JADE and JVM configuration options defined by the JADE and JAVA container profiles
if any) each one with a Controller Agent on top.

12
WADE USER GUIDE

 For each container requests the associated Controller Agent to activate the specified agents.
The target configuration is kept internally by the CFA (by default it is stored in a file called
_target.xml) and users are not expected to act on it directly.
Once the application is active, depending on the application needs, the containers and agents
distribution may change (e.g. a new container can be launched with some agents on top to face an
unexpected workload). Therefore the running configuration may differ from the target
configuration. By requesting the CFA to perform the SaveConfiguration action of the
Configuration Ontology, it is possible to take a snapshot of the running configuration and make it
become the target configuration. In this way if the application is stopped and restarted its
distribution of agents and containers will not change.
In order to tear down the application it is necessary to request the CFA to perform the
ShutdownPlatform action of the Configuration Ontology. This has the effect of killing all
containers, but the Main Container.
At any time it is possible to request the CFA to export the target configuration
(ExportConfiguration action of the Configuration Ontology) to an application configuration
XML file. Figure 6 summarizes the main actions of the Configuration Ontology and their effect on
application configurations.

Application
configurations
Main Container available for import

Container-1
CFA
Export
Configuration

Container-2
Startup Import
Platform Configuration

Container-3
SaveConfiguration Target
Running
configuration
configuration

Figure 6. Managing application configurations

2.3 Getting started with the platform

Having described the architecture of WADE-based applications and the main management actions,
in this section we describe how to launch the WADE platform and start playing with it.

2.3.1 WADE Distribution structure overview

WADE does not require any installation procedure a part from un-zipping the distribution zip file.
Having done that somewhere on your disk, this is how the resulting directory tree should look like.

13
WADE USER GUIDE

wadeSuite/
|
|---wade/
|---build.xml
|---...
|---bin
|---startMain.bat
|---startBootDaemon.bat
|---startMain.sh
|---startBootDaemon.sh
|---cfg/
|---main.properties
|---types.xml
|---...
|---configuration/
|---sample.xml
|---...
|---lib/
|---wade.jar
|---wadeInterface.jar
|---jade.jar
|---...
|---src/
|---deploy/
|---log/
|---projects/
|---add-ons/

The build.xml file allows recompiling WADE and recreating the libraries using Apache ANT.
The bin directory includes .bat and .sh startup scripts (for Windows and Linux respectively)
that can be used to start the Main Container and the BootDaemon process.

The cfg directory includes all configuration stuff. In particular

 the main.properties file is the JADE property file that includes all configurations
related to the Main Container
 the types.xml file define agent types and roles as will be described in section 2.4
 the configuration subdirectory (as described in section 2.2) is used to store application
configurations. A sample application configuration (actually that presented in 2.2) is
provided in the sample.xml file.

The lib directory includes the WADE libraries wade.jar and wadeInterface.jar and all
libraries used by WADE including JADE.

The src directory includes all WADE sources.

The deploy directory is where deployed workflows will be stored.

The log directory is where log files produced by the Main Container and the local BootDaemon
will be stored. In particular the BootDaemon log file contains all logs produced by the containers
launched by the BootDaemon prefixed with the container name.
14
WADE USER GUIDE

The projects directory is used to keep property files describing the location and structure of
WADE-based applications that have to be launched using the WADE startup scripts as will be
detailed in section 2.6.

2.3.2 Starting up the platform

In order to start the WADE platform up on a single host it is sufficient to
 double-click on the startMain.bat file or execute the startMain.sh script depending
on the underlying operating system. This activates the Main Container.
 double-click on the startBootDaemon.bat file or execute the startBootDaemon.sh
script depending on the underlying operating system. This activates the BootDaemon on the
local host.
To try WADE on more than one host, WADE must be installed on all hosts and the BootDaemon
must be activated on each of them.

2.3.3 The Management Console

In general we expect real world applications to provide their own management console (possibly
developed as a web application) properly customized according to the addressed domain and the
application requirements. In order to start familiarizing with the platform however, WADE comes
with a minimal management console that allows performing basic administration actions such as
starting and stopping an application, importing and exporting configurations and deploying
new/modified workflows without tearing the application down (this feature will be described in
section 2.7). It has to be noticed the WADE suite also include a powerful web based administration
console that provides a lot of additional features, but how to install and use it is outside the scope of
this document.
The basic WADE Management Console is implemented as an agent
(com.tilab.wade.cfa.ManagementAgent) and is activated by default on the Main
Container. In order not to activate it, it is sufficient to remove the ManagementAgent from the
list of agents to be activated at Main Container bootstrap in the cfg/main.properties file.
Having started the Main Container and the BootDaemon as described in 2.3.2 the WADE
Management Console should appear showing that the status of the application is DOWN (see
Figure 7).

15
WADE USER GUIDE

Figure 7. The basic WADE Management Console

By clicking on the Import Configuration button (or selecting the Configurations 

Import Configuration menu item) it is possible to select and import the sample
application configuration.

Figure 8. Importing a configuration

Having imported a configuration it is now possible to activate the application (note that at this point
this is just a dummy application with no logics as we don’t have any application specific code or

16
WADE USER GUIDE

workflow) by clicking on the Startup button (or selecting the Platform  Startup menu
item). Figure 9 shows how the JADE RMA should look like once the application has started.

Figure 9. Running the sample configuration

2.4 Agent types and roles

As mentioned a WADE-based application is composed of a set of agents that are instances of one or
more agent types. As described in 2.1 WADE natively provides some agent types (e.g. the
Configuration Agent, the Controller Agent, the Workflow Engine Agent and so on). An
application often adds other domain specific agent types either extending the WADE base types
(Primarily the Workflow Engine Agent) or directly extending their common base class
WadeAgentImpl. In general we recommend WADE developers not to extend the
jade.core.Agent class of the JADE library directly. As will be described in section 2.5, in
facts, the WadeAgentImpl class provides a number of ready made mechanisms that are used to
manage in a uniform way several aspects such as DF registration, agent attributes inspection,
application termination and fault recovery.
In JADE an agent type just corresponds to a class that (directly or indirectly) extends
jade.core.Agent. WADE explicitly introduces the notion of agent type. More in details an
agent type has a name, a corresponding class and possibly a set of properties that will apply to all
agents of that type.
The agent types involved in a WADE-based application are defined in a file called types.xml
located in the wade/cfg directory.
In order to show how the types.xml file looks like we provide below the default types.xml
file defining the WADE base types.

<platform>
<agentRoles>
<agentRole description="Administrator"/>
<agentRole description="Workflow Executor"/>

17
WADE USER GUIDE

</agentRoles>

<agentTypes>
<agentType description="Workflow Engine Agent"
className="com.tilab.wade.performer.WorkflowEngineAgent"
role="Workflow Executor">
</agentType>

<agentType description="Control Agent"

className="com.tilab.wade.ca.ControllerAgent"
role="Administrator">
<properties>
<property name="autorestart" value="true"/>
<property name="threadNumberThreshold" value="300"/>
<property name="class-loader-root" value="./deploy"/>
</properties>
</agentType>

<agentType description = "Configuration Agent"

className="com.tilab.wade.cfa.ConfigurationAgent"
role="Administrator">
<properties>
<property name="configurationsPath" value="./cfg/configuration"/>
</properties>
</agentType>

<agentType description = "Runtime Allocator Agent"

className="com.tilab.wade.raa.RuntimeAllocatorAgent"
role="Administrator"/>
<agentType description = "Workflow Status Manager Agent"
className="com.tilab.wade.wsma.WorkflowStatusManagerAgent"
role="Administrator"/>
<agentType description = "Event System Agent"
className="com.tilab.wade.esa.EventSystemAgent"
role="Administrator"/>
</agentTypes>
</platform>

From the snippet above it can be noticed that the types.xml file also allows the definition of
agent roles (the WADE default types.xml file defines 2 agent roles: Administrator and
Workflow Executor). An agent role can be seen as a macro-type and many agent types can refer to
the same agent role. As for agent types, also for an agent role it is possible to specify properties.
These properties will be inherited by all agent types that refer to that agent role.
The agent type and agent role abstractions provide the following advantages:
 Application configuration files refer to agent types instead of agent class names that are
typically less friendly to understand and remember.
 All configuration parameters that must be shared by all agents of a given type/role can be
specified once for all by means of agent type/role properties.
 WADE provides a number of utility methods (provided by the DFUtils class) to easily
discover agents at runtime on the basis of their type/role.

2.4.1 Using environment variables

When specifying values for properties of agent types and roles it is possible to refer to environment
variables using the syntax ${variable-name}.

18
WADE USER GUIDE

For instance the following snippet sets the directory where the CFA searches for application
configurations to be imported/exported to the applicationConfigurations directory under
the home directory of a project pointed to by the project-home environment variable

...
<agentType description = "Configuration Agent”
className="com.tilab.wade.cfa.ConfigurationAgent"
role="Administrator">
<property name="configurationsPath"
value="${project-home}/applicationConfigurations"/>
</agentType>
...

2.5 WADE Agents

In order for the management and fault tolerance mechanisms provided by WADE to work properly,
agents in a WADE-based application must follow a set of rules regarding for instance how to
register with the DF. These rules are automatically managed by the
com.tilab.wade.commons.WadeAgentImpl class. Therefore WADE developers are
recommended to extend it rather than the jade.core.Agent class directly.
The WadeAgentImpl class already implements the setup() method of the Agent class.
agents extending WadeAgentImpl should redefine the agentSpecificSetup() method
(rather than setup()) to perform their application specific initialization operations.
More in details the WadeAgentImpl class manages the following aspects.

Arguments passed to agents at startup time. As described in 2.2, agents that make up a WADE
based application are specified in the application configuration XML file. That file also allows
specifying startup arguments to be passed to each agent instance as exemplified in the snippet
below.

...
<container name="Execution-Node">
<agents>
<agent name="john" type="Book Buyer Agent">
<parameters>
<parameter key=”TARGET-BOOK”>
<value>The Lord of the Rings</value>
</parameter>
<parameter key=”MAX-PRICE”>
<value>50</value>
</parameter>
</parameters>
</agent>
</agents>
</container>
...

The WadeAgentImpl class makes available such arguments to subclasses in the form of a
protecetd Map (called arguments). For instance, refering to the snippet above, the Book
Buyer Agent can retrieve the target book to buy by just doing

String targetBook = (String) arguments.get(“TARGET-BOOK”);

Furthermore methods to access agent arguments as String, boolean, int and so on (managing all
necessary format conversions) are directly provided as exemplified below.

19
WADE USER GUIDE

int maxPrice = getIntArgument(“MAX_PRICE”, 30); // Default: 30

DF registration. A WADE agent extending WadeAgentImpl automatically registers with the DF

providing a DFAgentDescription that allows the Control Agents to restart it in case of
unexpected failure (either of the agent itself or of the container/host the agent lives in). Such
DFAgentDescription includes, among others, the agent type, role, class name, location, host
and all startup arguments received by the agent. Thanks to this registration it is possible to search
agents on the basis of their type and role. The com.tilab.wade.utils.DFUtils class
provides the searchAnyByType(), searchAllByType(), searchAnyByRole() and
searchAllByRole() utility methods to do that very easily.

Agent attributes runtime inspection and management. The WadeAgentImpl class provides an
embedded mechanism that allows exposing some agent attributes and make them readable and
possibly modifiable at runtime with minimal effort. In order to make an attribute readable at
runtime it is sufficient to have a getter method annotated by means of the @AttributeGetter
annotation as exemplified below.

@AttributeGetter(name=”The name of the book I’m trying to buy”)

public String getTargetBook() {
return targetBook;
}

In order to make an attribute not only readable, but also modifiable it is sufficient to add a setter
method annotated by means of the @AttributeSetter annotation as below.

@AttributeSetter
public void setTargetBook(String tb) {
targetBook = tb;
}

Runtime attribute inspection/modification can be done by requesting an agent extending

WadeAgentImpl to perform the GetAgentAttributes and SetAgentAttribute actions
of the WadeManagementOntology. The WOLF development environment provides a ready
made GUI to inspect and modify agent attributes.

2.6 Running WADE-based applications

In section 2.3.2 we showed how to use the WADE startup scripts to activate the WADE platform. In
general, however, a WADE-based application is not only composed of workflows that can be
executed on a “naked” WADE platform, but also includes a number of classes (e.g. application
specific agents) and possibly makes reference to some external libraries. In order for the WADE
startup scripts to include application specific jar files into the classpath, it is of course possible to
put them all (including external libraries) into the lib directory of the WADE installation.
Similarly it is possible to modify the configuration files included in the cfg directory of the
WADE installation to reflect application specific configurations. In general however we
recommend to keep all application specific stuff separate from the WADE installation as
described in the followings. In this way it is possible to deal with more than one WADE based
application at the same time without mixing libraries and configuration files.

20
WADE USER GUIDE

In order to do that it is sufficient to put a simple property file named <application-

name>.properties into the projects/ directory. This property file is mainly intended to
specify the project-home property that must point to where the project (i.e. the application) is
located on the disk. At this point the application Main Container can be launched by typing
startMain <application-name>
A similar command line can be used in Linux.
Figure 10 provides an example showing how the project property file and the types.xml file of a
Book Trading WADE based application can look like. With reference to that example the book
trading Main Container can be launched by typing
startMain booktrading

C/ project-home=C:\bookTr
|---wade/ ...
|---...
|---bin/
|---startMain.bat
|---startBootDaemon.bat
|---...
|---cfg/ ...
|---lib/ <agentType description="Control Agent"
|---wade.jar WADE ...>
<properties>
|---jade.jar libraries only ...
|---... <property name="class-loader-root"
|---log/ value="${project-home}/deploy"/>
</properties>
|---projects/ </agentType>
|---booktrading.properties
|---... <agentType description="Configuration Agent"
...>
<properties>
| <property name="configurationsPath"
|---... value="${project-home}/cfg/configuration"/>
</properties>
| </agentType>
|---bookTr/ ...
|---src/
|---workflow/ Book Trading
|---src workflow sources
(if any)
|---classes/
|---cfg/
|---main.properties
|---types.xml Book Trading
|---... project
|---configuration/ configurations
|---bookTrading.xml
|---lib/
|---bookTrading.jar Book Trading
|---... project libraries
|---log/
|---deploy/
|---... Book Trading project workflows
|
|---...

Figure 10. Project property file

The meaning of the class-loader-root property of the Control Agent type is described
in more details in next section.
Finally it should be noticed that, while the Main Container is actually part of the application and
therefore must be launched specifying the project name as described above, the BootDaemon is
fully independent from the application and will always be launched by simply typing
startBootdaemon

21
WADE USER GUIDE

2.7 Hot deployment of workflows

As mentioned in 1.3 WADE workflows are implemented as Java code. In order to support
workflow hot deployment (i.e. the possibility of adding new workflows and modifying existing
ones without the need to tearing the system down) an ad-hoc ClassLoader is used to load them.
Moreover, since agents executing workflows are typically scattered across different containers, for
each container there is an instance of this ClassLoader managed by the Control Agent controlling
the container itself. More in details the property class-loader-root of the Control Agent type
defines the directory (<project-home>/deploy by default) where jar files containing
workflows must be placed.
As shown in Figure 11, jar files containing new/modified workflows can be deployed at runtime by
means of the WADE Management Console.

Figure 11. Workflow hot deployment using the WADE Management Console

NOTE: It should be noticed that, due to the Java ClassLoader delegation model, if the class
implementing a workflow is included in the application classpath it is not loaded by the WADE
Control Agent ClassLoader, but by the Java System ClassLoader. As a consequence it will not be
possible to update it by means of the WADE workflow hot deployment mechanism.

22
WADE USER GUIDE

3 Workflows
In this section we go in more details in the most relevant feature that WADE provides i.e. the
support for defining system logics according to the workflow metaphor. More in details in this
chapter we present how workflows are implemented in WADE, while in next chapter we will
describe how to execute them. Instead this guide does not provide a section focusing on how to use
Wolf to create workflows. Readers can refer to the WADE Tutorial available on line in the WADE
site (http://jade.tilab.com/wade/doc/tutorial/WADE-Tutorial.pdf) for a step-by-step description
about the Wolf Workflow Editor and the other relevant features that Wolf provides.

3.1 The workflow meta-model

As anticipated the approach adopted by WADE to support workflow development is to provide a
workflow view over a Java class (with a proper structure) that can therefore be managed exploiting
the full power of the Eclipse IDE. As a consequence WADE does not use internally any standard
workflow definition language, but a proprietary formalism based on the Java language. However, in
order not to re-invent the wheel and to facilitate import/export operations, WADE adopts a
workflow meta-model derived from that defined in the XPDL standard specified by the Workflow
Management Consortium (www.wfmc.org). The main elements that compose this meta-model are
described hereafter.
A task that is being described is called a Process. A process is composed of a set of Activities
each one corresponding to the execution of given operations. A process defines a single Start
Activity (specifying the execution entry point) and one or more End Activity (specifying the
execution termination points). Each non-end activity has one or more outgoing Transitions,
possibly associated to a condition, leading to another activity in the process. Once the execution of
the operations included in a given activity is terminated, the conditions of all outgoing transitions
are evaluated. As soon as a condition holds the corresponding transition is fired and the execution
flow goes on with the operations included in the destination activity.
A process can have one or more Formal Parameters defining the type of required inputs and
expected outputs. At process invocation time proper values must be provided for input parameters
and, at the end of the execution, the values produced as output parameters are returned to the
requester.
The main types of activity supported by WADE are described below.

Execution activities
 Code activities. The operations included in a code activity are specified directly by a
piece of Java code embedded in the workflow process definition.
 Tool activities. The operations included in a tool activity consist in invoking one or
more external tool generically identified as Applications. Applications are
computational entities defined outside the workflow process and wrapped by a
uniform interface. Similarly to processes, applications have formal parameters
defining required inputs and expected outputs. In general we expect applications to be
provided to e.g. interact with an external system or appliance or to execute a ready
made computation, but developers are free to use the Application abstraction as they
like.
 Subflow activities. The operations included in a subflow activity consist in the
invocation of another workflow process. The execution of the subflow takes place in a
separate computational space and (as will be described in section Errore. L'origine
riferimento non è stata trovata.) can be even carried out by a different performer

23
WADE USER GUIDE

agent (possibly running in a remote host) with respect to that executing the main
process.
 WebService activities. The operations included in a Web Servicve activity consist in
invoking a web service.

Synchronization activities
 WaitEvent activities. When the execution flow reaches a WaitEvent activity, the
workflow suspends until a given event happens. Events are further discussed in
section 3.3.1.
 WaitWebService activities. When the execution flow reaches a WaitWebService
activity the workflow suspends until a given operation of a previously exposed Web
Service is invoked.
 SubflowJoin. By default subflows are executed synchronously, i.e. the main workflow
blocks until the subflow completes and then goes on. Alternatively it is possible to
execute a subflow asynchronously, that is the main workflow proceeds just after
launching the subflow. When the execution flow reaches a Subflow Join activity the
workflow blocks until a previously launched asynchronous subflow completes.

Finally the process makes reference to a set of Data Fields each one having a name, a type
and possibly an initial value. Data fields can be referenced wherever in the process e.g. in the
conditions associated to the transitions, as actual parameters for application, subflow and web
service invocations and in the pieces of code triggered by code activities.
Figure 12 shows an example summarizing some of the elements of the WADE meta-model.

Process P

1) Application X
produces value 10 Data Fields
and stores it in data
filed “e” a “Hallo world”
Activity A1
b 25
(Tool)
Appl. X 3.25
c
d
e 10
2) This
transition is
otherwise e <= b fired
f
Activity A2 ...
(Code) Activity A3
(Subflow)
e = e+10;
f = 0;
3) A3 consist in the
execution of another
workflow process
defined outside P

Figure 12. Main elements in the XPDL meta-model

24
WADE USER GUIDE

3.2 Workflow elements implementation

Having described the main elements that make up a workflow process, in this section we show the
structure of the Java code that actually implements them. It is important to note that, even if they
could, we expect developers not to write the pieces of code that are described in this section
directly, but to use the graphical workflow editor provided by WOLF to do that. However
understanding the Java-based workflow definition formalism adopted by WADE may be useful to
prevent unexpected behaviours and to investigate problems.

3.2.1 The WorkflowBehaviour class

Each workflow process is implemented by a Java class that extends (directly or indirectly) the
com.tilab.wade.performer.WorkflowBehaviour base class. The
WorkflowBehaviour class on its turn extends the JADE FSMBehaviour class and provides
on top of it an API consistent with the meta-model presented in 3.1. In particular the
registerActivity() and registerTransition() methods allow adding activities and
transitions to the process. Even if in principle this is not strictly necessary it is highly recommended
to register all process activities in a private void method called defineActivities()
and all transitions in a private void method called defineTransitions(). This is
because the workflow graphical editor included in Wolf searches for these methods to detect
process activities and transitions to show.
The registerActivity() method gets the behaviour implementing the registered activity as
an argument. More in details there are different classes corresponding to the different types of
activity presented in 3.1. These are CodeExecutionBehaviour to be used to register a code
activity, ToolExecutionBehaviour to be used to register a tool activity,
SubflowDelegationBehaviour to be used to register a subflow activity,
WebServiceInvocationBehaviour to be used to register a WebService activity and so on.
The actual operations to be performed in a registered activity (no matter of its type) are specified in
a void method of the workflow class. That method has a name constructed by adding the
“execute” prefix to the name of the corresponding activity. Each activity behaviour is just
responsible for invoking the related method when the activity is visited.
Similarly the registerTransition() method gets a Transition object as an argument. In
case the transition has an associated condition, this is implemented by a boolean method of the
workflow class. The Transition object is just responsible for invoking that method when the
transition condition must be evaluated. The name of the method implementing a condition is
constructed by adding the “check” prefix to the condition name.
Figure 13 summarizes the above correspondences.

25
WADE USER GUIDE

package coffee.processes;
...

public class CoffeeProcess extends WorkflowBehaviour {

public static final String GRIND_ACTIVITY = “Grind”; COFFEE PROCESS
public static final String PREPARE_ACTIVITY = “Prepare”;
public static final String POUR_ACTIVITY = “Pour”;
public static final String ERROR_ACTIVITY = “Error”;
...
private void defineActivities() {
CodeExecutionBehaviour ceb = new CodeExecutionBehaviour(GRIND_ACTIVITY, this);
registerActivity(ceb, INITIAL);

ToolExecutionBehaviour teb = new ToolExecutionBehaviour(PREPARE_ACTIVITY, this);

teb.addApplication(“coffee.applications.PrepareCoffee”);
registerActivity(teb);
Grind

SubflowDelegationBehaviour sdb = new SubflowDelegationBehaviour(POUR_ACTIVITY, this);

sdb.setSubflow(“coffee.processes.PourCoffee”);
registerActivity(sdb, FINAL);
NotEnoughCoffee
RouteActivityBehaviour rab = new RouteActivityBehaviour(ERROR, this);
rab.setError();
registerActivity(rab, FINAL); Prepare
}

private void defineTransitions() {

registerTransition(new Transition(“NotEnoughCoffee”, this), Error
GRIND_ACTIVITY, ERROR_ACTIVITY);
registerTransition(new Transition(), GRIND_ACTIVITY, PREPARE_ACTIVITY);
registerTransition(new Transition(), PREPARE_ACTIVITY, POUR_ACTIVITY);
}
Pour
...

protected void executeGrind() throws Exception {

// Code to grind coffee
}

protected void executePrepare(ApplicationList appls) throws Exception {

Application prepareCoffee = appls.next();
// Fill actual input parameters of application PrepareCoffee
...
executeApplication(prepareCoffee);
// Store actual output parameters of application PrepareCoffee
...
}

protected void executePour(Subflow sblf) {

// Fill actual input parameters of process PourCoffee
...
executeSubflow(prepareCoffee);
// Store actual output parameters of process PourCoffee
...
}

protected boolean checkNotEnoughCoffee() {

// Return true if we don’t have enough coffee
}
}

Figure 13. Workflow class structure

In certain cases the execution of an activity cannot be fully carried out within a single method. For
instance when the execution flow reaches a WaitWebService activity, before suspending, it is often
necessary to execute some logics to precisely define wich invocation of the indicated web service
operation will actually resume the workflow. For example the workflow must be resumed as soon
as the notifyCoffeeReady web service operation will be invoked with parameter orderId
equals to some value. Of course as soon as the workflow will be resumed it is often necessary to
execute some other logics to e.g. process the parameters of the web service operation that has just
been invoked. As a consequence for these types of activity there is not a single “execute” method,

26
WADE USER GUIDE

but a “before” method and an “after” method that will be invoked before and after the execution of
the activity. For instance, given the WaitForCoffee WaitWebService activity the workflow class
will have the methods

protecetd void beforeWaitForCoffee(…) {

…
}
and
protecetd void afterWaitForCoffee(…) {
…
}

3.2.2 Data Fields and Formal Parameters

Data Fields of a workflow process are implemented as fields of the workflow class. For instance,
with reference to the coffee workflow shown in Figure 13, we could have a data field containing the
amount of ground coffee in grams produced by the GRIND activity. This is implemented as an int
field of the CoffeProcess class.
Workflow formal parameters need to be accessed by the workflow in a similar way to data fields.
Therefore they are implemented, like data fields, as fields of the workflow class. In order to let
WADE know that they are formal parameters, however, they must be annotated by means of the
FormalParameter annotation included in the com.tilab.wade.performer package. This
annotation has several attributes indicating for instance the mode (i.e. INPUT or OUTPUT) or
whether the parameter is mandatory (default) or optional. For instance the coffee process can take
an input formal parameters nCups indicating the number of cups to be prepared. This would be
implemented as an int field annotated as below

public class CoffeeProcess extends WorkflowBehaviour {

...
// Formal parameters
@FormalParameter(mode=FormalParameter.INPUT)
private int nCups;
...
// Data fields
private int groundCoffeeGr;
private int coffeeGrPerCup = 10;
...
protected boolean checkNotEnoughCoffee() {
return groundCoffeeGr < coffeeGrPerCup * nCups;
}
}

3.2.3 BuildingBlock invocation

As mentioned, for each registered activity there is a corresponding method in the workflow class.
While for code activities this method actually specifies the operations to be performed in the
activity, for tool, subflow, WebService and other types of activity, the corresponding method just
specifies the invocation of the indicated element. Such elements therefore can be seen as building
blocks that are composed by the workflow to implement a complex process. The invocation of a
building block (no matter if it is a subflow, a web service or an application) is composed of three
steps:
1. fill actual input parameters.
2. invoke the indicated building block.
3. extract actual output parameters and store them in suitable workflow data fields for later
processing

27
WADE USER GUIDE

In the case of a tool activity, where it is possible to invoke more than one application, these steps
are repeated for each application to be invoked. Therefore, unlike methods corresponding to Code
activities that take no arguments, the methods corresponding to building block invocations take an
argument representing the building block(s) to invoke.
With reference to the coffee example, let’s assume the PrepareCoffee application has the
following formal parameters:
coffeeType (INPUT) of type String indicating the type of coffee to produce (e.g. “Espresso”,
“American” ...)
cupsNumber (INPUT) of type int indicating the number of cups to prepare
coffeeJug (OUTPUT) of type CoffeeJug (a domain specific class) indicating the jug
containing the prepared coffee.
Here is how the executePrepare() method corresponding to the PREPARE activity would look
like.

public class CoffeeProcess extends WorkflowBehaviour {

...
// Formal parameters
@FormalParameter(mode=FormalParameter.INPUT, index=0)
private int nCups;
...
// Data fields
private int groundCoffeeGr;
private int coffeeGrPerCup = 10;
private CoffeeJug jug;
...
protected void executePrepare(ApplicationList appls) throws Exception {
Application prepareCoffee = appls.next();
// Fill input parameters
prepareCoffee.fill(“coffeeType”, “Espresso”);
prepareCoffee.fill(“cupsNumber”, nCups);

// This actually invokes the application execution

executeApplication(prepareCoffee);

// Store output parameters

jug = (CoffeeJug) prepareCoffee.extract(“coffeeJug”);
}
...
}

The Application class provides overloaded fill() methods to deal with all types of object
without the need of casting. On the other hand when storing output parameters into a data field a
proper cast to the data field type and possibly a primitive conversion (e.g. from Integer to int)
must be done explicitly.
Handling actual parameters of a subflow or web service is done exactly the same way since the
Subflow class and the WebService class provides identical fill() and extract()
methods.

3.2.4 Layout information

All information related to the appearance of a workflow as shown in the graphical editor provided
by Wolf (e.g. the fact that the box representing a given activity is placed in a given position within
the screen, or that the arrow representing a given transition has some routing points) are specified
by means of proper annotations associated to the workflow class. Clearly Wolf fully manages all
these annotations and developers do not need to care or know anything about them.

28
WADE USER GUIDE

3.3 Synchronization Activities

3.3.1 Events
As mentioned WADE allows defining synchronization activities where the execution flow blocks
until a given event happens. These are called Wait Event Activities.
The first step to use Wait Event Activities is to define one or more “custom” event types. This is
done in the types.xml (already mentioned in 2.4 when describing agent types and roles) file as
exemplified in the snippet below.

…
<customEventTypes>
<EventType description="CoffeeReady">
<parameters>
<Parameter name="requestId" type="java.lang.String" />
<Parameter name="cupsNumber" type="java.lang.Integer" />
</parameters>
</EventType>
</customEventTypes>
…

Each EventType has a description (that acts as the event type identifier) plus a set of Parameters
each one with a name and a type (the fully qualified class name to be used for values of that
parameter).
When adding a Wait Event Activity in a workflow, the developer is requested to specify the type of
the expected event (one of those included in the customEventTypes section of the
types.xml file) and, optionally an expression that precisely identifies which specific event the
workflow intends to receive. For instance, with reference to the snippet above, a workflow that
previously activated the preparation of a coffee can block until a CoffeeReady event is received
with parameter requestId equals to a given value.
As usual it is possible to store the information (parameters) carried by the received event as it
happens for e.g. subflow or web service output parameters. This is done in the afterXXX()
method as exemplified below.

private int nCups;

…
public void afterWaitForCoffee(GenericEvent ev) {
nCups = (int) ev.extract(“cupsNumber”);
}

As briefly mentioned in 2.1, internally events are managed by the Event System Agent. However,
unless they have specific requirements, developers do not have to care about it. The only important
information they need to know is how to submit events to a WADE based system. This is done by
means of the EventChannel API included in the com.tilab.wade.event package as
exemplified by the snippet below where we show the code that could be used by the module in
charge of actually preparing the coffee to submit a CoffeeReady event.

…
private EventChannel channel = new EventChannel();
…
private submitCoffeeReady(String id) {
GenericEvent ev = new GenericEvent(“CoffeeReady”);
ev.fill(“requestId”, id);
try {
channel.submitEvent(ev);
}

29
WADE USER GUIDE

catch (Exception e) {
e.printStackTrace();
}
}

Events can be submitted by both external components/systems and internal modules (agents). As a
consequence the EventChannel class provides two constructors. In the former case the
constructor with no parameters must be used (as in the snippet above) that internally uses the
default JadeGateway to connect to the WADE-based world (see the JADE documentation for
details about how to initialize the JadeGateway). In the latter case the constructor that takes an
Agent instance as parameter must be used.

The Web Administration Console included in the WADE Suite provides a section to manually
submit events to the system.

3.3.2 Incoming Web Services

In this section we describe how to block a workflow until a given Web Service operation is invoked
(typically by an external system). This is done including in the workflow a Wait Web Service
Activity. The invocation of a Web Service operation can be considered as a “predefined” type of
event and, as a matter of facts, a Wait Web Service activity is very similar to a Wait Event Activity
described in previous section. The only difference is that, the developer is not requested to select a
custom event type among those defined in the types.xml file, but to select a Web Service
operation among those exposed by the system as will be described later in this section. A part from
that, the underlying mechanism is exactly the same and therefore, as soon as the indicated WS
operation is invoked, the workflow is resumed and receives, in its afterXXX() method, a
GenericEvent whose parameters reflect those of the invoked operation.

The process to make a WADE based system expose a Web Service so that workflows can block
waiting for its operations to be invoked is very simple.
The first step is to import the WSDL of the Web Service to be exposed. This is done within Wolf:
the developer is requested to indicate whether the WSDL describes a Web Service to be invoked
(Import as client) or to be exposed (Import as server). In this section we focus on
the second option. The Import as server operation generates a new jar file in the
wadeServices directory. This jar file includes all the classes necessary to expose the Web
Service and to manage operation parameters within workflows. As soon as a WSDL has been
imported as server its operations become available for selection when including a Wait Web Service
activity in a workflow.
The second step is to expose the Web Service so that it becomes available for invocation. This
requires the WadeServices component as depicted in Figure 14. The WadeServices module is a web
application (.war file) that can be deployed in whatever Servlet container such as Apache Tomcat.
The actual Web Service exposure operation can be done from Wolf as described hereafter.
 Connect Wolf to the running WADE-based application that must expose the Web Service.
 Be sure the WadeServices module is installed in your servlet container and the servlet
container is up and running.
 Be sure the WadeServices module is connected to the WADE based application. This can be
done by checking if the WadeServices Gateway Container (WSEG-Container) appears
in the Wolf AgentTree view as depicted in Figure 15. If this is not the case, connect the
WadeServices module by means of the Start button of the WadeServices administration
web page (http://localhost:8080/wadeServices)
 In the Eclipse Package Explorer right click on the service jar file and select the WADE
Tools  Expose menu item as depicted in Figure 16.
30
WADE USER GUIDE

WSDL

1) Import WSDL

WOLF External
2) Deploy WebService
System
Developer

3a) Invocation
External
WadeServices
Exposed System
Web Service
Tomcat
Wade Based Application
External
System
3b) Resume workflow

Figure 14. Exposing Web Services

The WadeService Gateway Container

Figure 15. The WadeServices Gateway Container in the Wolf AgentTree view

31
WADE USER GUIDE

Figure 16. Exposing a Web Service using Wolf

3.3.3 Long-Running workflows

With the term “Long-running” workflows we mean workflows that survive to the system shutdown
and restart. That is if the system is shut down while a long running workflow is running, as soon as
the system is restarted, the workflow is automatically resumed from the point where it was
interrupted. In order to make this possible, when executing a long-running workflow, the WADE
Workflow Engine automatically saves the workflow state after each step of execution. This is done
by serializing the WorkflowBehaviour object implementing the workflow and storing the
serialized bytes on a relational DataBased.
Clearly long-running workflows are less efficient with respect to normal “short-running” ones.
Therefore, depending on the characteristics and requirements of a given process to be implemented,
developers can choose whether to implement it as a short-running or as a long-running workflow.
Even though this is not mandatory, workflows that include Wait Event or Wait WebService
activities (i.e. workflows that can block for possibly long amount of times) are typically declared as
long-running.

In order to create a long-running workflow it is sufficient to

1) Declare the workflow as long-running by redefining the isLongRunning() method of the
WorkflowBehaviour class as below

public boolean isLongRunning() {

return true;
}

2) Be sure that the workflow class is Serializable

Furthermore, as mentioned, in order to properly execute long-running workflows, the Persistence

Add-On must have been added to the WADE installation (this simply means unzipping the
Persistence Add-On distribution package in the same directory where the main WADE distribution
package was unzipped).

32
WADE USER GUIDE

3.3.4 The Persistence Add-On

The Persistence Add-On allows WADE components (agents) to persist information on a relational
Data Base by means of Hibernate. Currently the only component that exploits the Persistence Add-
On is the Workflow Status Manager Agent, i.e. the agent in charge of tracing the state of
workflows. However we expect to have other agents exploiting it in future versions.

By default the DB related configurations are specified in the wadePersistence.cfg.xml file

included in the wadeSuite/wade/add-ons/persistence/cfg directory. This is a standard
Hibernate configuration file where it is possible to define the Hibernate dialect (indicating which
DB to use), and the connection driver, URL, username and password (if any).
Of course it is possible to tell WADE to use a different Hibernate configuration file by specifying
the property hibernateConfigurationFile in the global properties section of the
types.xml file as exemplified below.

<platform>
<properties>
<Property name="hibernateConfigurationFile" value="myPath/myFile.cfg.xml"/>
…
</properties>

<agentTypes>
…

This allows for instance to make WADE internal information and application specific information
to be stored in the same DB without duplicating the same configurations in two different files.

3.4 Workflow inheritance

Considering the landscape of workflow management tools, one of the most characterizing features
of WADE is certainly the possibility of creating new workflows by extending existing ones. In facts
while the workflow approach is intrinsically procedural, inheritance is a typical characteristic of
Object Orientation.
The simplest extension mechanism is based on Java class inheritance: being WADE workflows
Java classes, it is possible to redefine the methods associated to the process activities thus
modifying the related operations according to the needs of the extended workflow. Similarly it is
possible to redefine the methods that implement transition conditions.
Beside that basic extension mechanism, however, WADE allows modifying the very flow of the
parent process. In order to support that, the WorkflowBehaviour class provides the
deregisterActivity() and deregisterTransition() methods that, in conjunction
with the registerActivity() and registerTransition() methods described in section
3.2.1, allows modifying the process execution flow at will. For instance we could create an
extended version of the CoffeeProcess presented in section 3.2.1 as shown in Figure 17. The
Sweet Coffee Process adds the AddSugar activity and the related transitions. Moreover it removes
the transition from the Prepare activity to the Pour activity. The defineActivities() and
defineTransitions() methods of the SweetCoffeeProcess class reflect these
modifications.

33
WADE USER GUIDE

package coffee.processes;
...
SWEET COFFEE PROCESS
public class SweetCoffeeProcess extends CoffeeProcess {
public static final String ADD_SUGAR_ACTIVITY = “AddSugar”;
...
private void defineActivities() {
CodeExecutionBehaviour ceb = new CodeExecutionBehaviour(ADD_SUGAR_ACTIVITY, this);
registerActivity(ceb, INITIAL);
}
Grind
private void defineTransitions() {
deregisterTransition(PREPARE_ACTIVITY, POUR_ACTIVITY, false);
registerTransition(new Transition(), PREPARE_ACTIVITY, ADD_SUGAR_ACTIVITY); NotEnoughCoffee
registerTransition(new Transition(), ADD_SUGAR_ACTIVITY, POUR_ACTIVITY);
}

... Prepare Error

protected void executeAddSugar() throws Exception {
// Code to add sugar
}
... AddSugar
} NEW

Pour

Figure 17. Workflow extension

The changeActivityOrder() method allows changing the order of an activity: for instance
an activity that is FINAL in the parent workflow and must become INTERMEDIATE in an
extended workflow that adds an additional step at the end.
Finally all layout information can be changed at will.

34
WADE USER GUIDE

4 Executing workflows
As already mentioned WADE does not provide a single Workflow Engine. Rather it embeds “micro
workflow engines” into agents giving them the ability to execute tasks defined according to the
workflow metaphor. As a consequence, launching a workflow means requesting an agent to execute
it. The interaction between the requester agent and the executor agent follwos a FIPA-request
protocol as depicted in Figure 18.

Requester Executor
Agent Agent

REQUEST Load Workflow and

(ExecuteWorkflow) initialize input parameters

AGREE
(Execution-id) Execute Workflow

INFORM
(Result)

Figure 18. Workflow execution request interaction

The ExecuteWorkflow action, defined in the WorkflowManagementOntology (as all

other workflow execution related concepts), mainly includes the WorkflowDescriptor
specifying the workflow to execute, the actual parameters and optionally (but specifying it is a
recommended practice) a session-id that must be unique and that will be propagated to any subflow
triggered by the workflow that is been launched.
If the workflow class is correctly loaded and instantiated, the executor agent generates a unique
execution-id that can be used later on to act on the workflow (e.g. to kill it). This
execution-id is returned to the requester within the content slot of an AGREE message.
Note the difference between the session-id and the execution-id: the former is choosen by
the requester and will be preserved in subflows; the latter is choosen by the executor and changes
across subflows (i.e. if a subflow is launched, this will have a different execution-id).
Once the workflow has been executed the executor agent sends back an INFORM holding the list of
actual output parameters. In case of execution failure, instead of an INFORM message, a FAILURE
message will be sent back holding the failure reason.

35
WADE USER GUIDE

4.1 Launching workflows from code: the DispatchingCapabilities

class
In order to request an agent to execute a workflow from your code it is of course possible to create
the ExecuteWorkflow object and to implement a FIPA-request protocol with the intended
executor as described in previous section. However WADE provides a ready-made class, called
DispatchingCapabilities, that simplifies that. The code snippet below, taken (and
simplified) from the WorkflowTracer example included among the WADE examples distribution,
shows how to use it.

// Create a DispatchingCapabilities instance and link it to the local agent

DispatchingCapabilities dc = new DispatchingCapabilities();
dc.init(this);

...

try {
// Create the WorkflowDescriptor specifying the workflow to execute
// and the parameters
WorkflowDescriptor wd = new WorkflowDescriptor("tracer.ThreeStepWorkflow");
Map<String, Object> parameters = new HashMap<String, Object>();
parameters.put("message", "Hello World!");
wd.setParametersMap(parameters);

// Launch the workflow

dc.launchWorkflow(executor, wd, new WorkflowResultListener() {

public void handleAssignedId(AID executor, String executionId) {

// The executor correctly loaded the workflow and started the execution.
}

public void handleExecutionCompleted(List results,

AID executor,
String executionId) {
// Workflow execution successfully completed
System.out.println("Workflow completed successfully");
// Get output parameters
Map<String, Object> params = ElementDescriptor.paramListToMap(results);
int length = (Integer) params.get("messageLength");
System.out.println("Output parameter messageLength = "+length);
}

public void handleExecutionError(ExecutionError er,

AID executor,
String executionId) {
// Workflow execution failed.
}

public void handleLoadError(String reason) {

// Executor failed/refused to load the workflow
}

public void handleNotificationError(AID executor, String executionId) {

// Generic communication error. This should never happen
}

}, null);
}
catch (Exception e) {

36
WADE USER GUIDE

e.printStackTrace();
}

From the code snippet above one can notice that:

 The fully qualified name of the workflow class must be indicated to specify the workflow to
launch.
 Actual parameters can be filled in form of a Map and then set to the
WorkflowDescriptor
 The launchWorkflow() method of the DispatchingCapabilities class can be
used to trigger the execution of the indicated workflow on a given
WorkflowEngineAgent.
 A WorkflowResultListener can be specified to be notified, among others, about the
workflow execution completion.
 Workflow results (i.e. output parameters), if any, are provided as a List of Parameter
objects. The paramListToMap() static method can be used to convert it into a Map thus
simplifying the retrieval of output parameter values.

4.2 Using the LauncherAgent to launch workflows

In order to try the execution of workflows, WADE provides a simple utility agent called Workflow
Launcher (com.tilab.wade.tools.launcher.LauncherAgent class) that allows
launching workflows and tracing their execution. Again the Web Administration Console provides
a more powerful interface to launch and control workflows, but it is not discussed in this document.
The Workflow Launcher Agent can be started exactly as all other agents (e.g. using the JADE
RMA or specifying it among the agents to be activated at bootstrap time in the
main.properties file - see section 2.3.1). Alternatively the WADE Management Console
presented in section 2.3.3, allows starting the Workflow Launcher Agent by just clicking the
Workflow Launcher button.

37
WADE USER GUIDE

Figure 19. The Workflow Launcher Agent GUI

The GUI of the Workflow Launcher Agent, depicted in Figure 19, allows specifying, among others,
the workflow to launch, the agent that will be requested to execute it and the actual parameters (if
any).

38
WADE USER GUIDE

5 Advanced features
This section describes some advanced features that belong to the set of mechanisms provided by
WADE to manage the complexity of the distribution, but were not presented in chapter 2, to make
that chapter easier to read and understand.

5.1 Agent Pools

In many cases, especially when an application has to serve a heavy load of requests, it is useful to
have a set of agents of the same type to distribute the requests on. Since in many cases such sets can
count tens, hundreds or even thousands of agents, specifying them all one by one in the application
configuration file, as described in section 2.2, can be really annoying.
In order to deal with these situations in a friendly way, WADE supports the notion of Agent Pool.
Agent Pools are defined in application configuration files as exemplified below

<platform name="Test-pool">

<hosts>
...
</hosts>

<agentPools>
<agentPool name="testPool" size="10" type="Workflow Engine Agent">
<parameters>
<parameter key="testParameter">
<value>test value</value>
</parameter>
</parameters>
</agentPool>
</agentPools>

</platform>

The above snippet indicates that 10 agents of type Workflow Engine Agent will have to be
launched. Such agents will be called <agent-pool-name>-n and each of them will receive an
argument testParameter whose value is “test value”.
It should be noticed that the specification of an Agent Pool is very similar to the specification of an
Agent. A part from the size attribute, however, Agent Pools are not associated to a specific host and
container. WADE will take care of starting the agents of a pool in the available containers
according to “agent allocation policies”. More in details, when executing the StartupPlatform
action (see section 2.2) the Configuration Agent first creates all containers and agents explicitly
indicated in the target configuration and then, for each agent pool (if any), delegates the creation of
agents to the Runtime Allocator Agent (RAA). The latter, already briefly mentioned in section 2.1,
manages a set of agent allocation policies and, whenever an agent must be created and there is no
explicit indication about the container where to start it (as in the case of agents in a pool), it is
requested to create that agent according to its policies.

39
WADE USER GUIDE

5.1.1 Agent allocation policies

Agent allocation policies managed by the RAA are specified in an XML file called raa.xml and
located in the cfg/raa/ directory. The snippet below provides an example of how the raa.xml
file looks like.

<allocationRules>
<allocationRule>
<!-- Match all agents of type Foo with argument arg1 set to Bar -->
<condition>type==Foo;arguments.arg1==Bar</condition>
<!-- Allocate them round-robin across all containers with Java-Profile P -->
<policy name="com.tilab.wade.raa.policies.JavaProfileBasedRRPolicy">
<properties>
<property name="javaProfile" value="P"/>
</properties>
</allocationRule>

<allocationRule>
<!-- Match all agents -->
<condition>name==*</condition>
<!-- Allocate them round-robin across all containers -->
<policy name="com.tilab.wade.raa.policies.RoundRobinPolicy"/>
</allocationRule>
</allocationRules>

When the RAA is requested to create an agent somewhere in the platform, it scans the rules defined
in the raa.xml file. Each rule is composed of a condition and an allocation policy. When
processing a given rule the RAA matches the agent to be started against the condition. If the match
is OK then the RAA uses the allocation policy to select a container where to start the agent. If the
policy returns an existing container and the agent creation succeeds, the process stops. Otherwise
the RAA moves to the next rule.
The snippet above for instance would result in starting all agents of type Foo with the argument
arg1 set to Bar in containers with ContainerProfile (of type JAVA) P. All other agents would be
allocated round-robin across all available containers.
An allocation policy is a Java class implementing the
com.tilab.wade.raa.AgentAllocationPolicy interface or extending the
com.tilab.wade.raa.policies.BaseAgentAllocationPolicy base class.
Developers interested in defining their allocation policy according to application specific
requirements just need to provide new policy classes (see the javadoc of the
AgentAllocationPolicy interface for details). However WADE already provides a set of
ready-made policies (such as the RoundRobinPolicy and the
JavaProfileBasedRRPolicy mentioned in the snippet above) that can be used to cover many
realistic situations. Details on how such policies work and how to configure them can be found in
the related javadoc.

5.2 Agent and containers fault tolerance

Fault tolerance is a very important issue especially when dealing with real-world application. Using
a distributed approach (as typically happens when building an application on top of JADE/WADE)
provides a mean to deal with unexpected HW faults.
JADE already includes suitable fault tolerance mechanisms (such as the Main Container Replication
Service described in chapter 9 of the book “Developing Multi Agent Systems with JADE”) that
allows the platform to survive to a fault of a container or host. Such mechanisms however work at
the platform level and not at the application level. That is they ensure that the mechanisms provided

40
WADE USER GUIDE

by the platform, such as message delivery and agent creation/destruction, continue to work even
after an unexpected fault of one of the hosts where a JADE-based application is running. Of course
if an agent implementing a given application specific piece of functionality F was running on the
crashed host, the application (even if its remaining agents will still be able to exchange messages
and exploit other platform services) may not work anymore as functionality F is no longer available.

In order to support fault tolerance at the application level, WADE provides an additional
mechanism that enable automatically restarting all agents that suddenly disappear due to HW or SW
faults. This mechanism is implemented by the Control Agents. As described in section 2.1, each
container (but the Main Container) holds a Control Agent that is responsible for supervising the
resources in the local container. Furthermore Control Agents coordinate themselves so that a single
leader is elected.
The “autorestart” mechanism works as follows.
 If an agent suddenly dies (this may happen if there is a software bug in the agent’s code that
causes an uncaught exception), the Control Agent of the local container automatically
restarts it.
 If an entire container suddenly dies (e.g. because the container process is killed), of course
the local Control Agent dies too. The leader Control Agent then takes care of restarting the
whole container (this operation is actually performed by the BootDaemon on the host where
the container was active upon a request from the leader Control Agent) with all its agents.
 If the dead container included the leader Control Agent, other Control Agents elect a new
leader that takes care of restarting the dead container.
 If the recreation of the whole container fails (this is always the case when a container
disappeared due to a fault of the underlying HW) the leader Control Agent restarts all agents
that where living in the dead container through the Runtime Allocator Agent described in
section 5.1. The latter will then recreate all dead agents according to its agent allocation
policies.

It should be noticed that the autorestart mechanism provided by WADE is only responsible for
restarting the dead agents passing them the same arguments they were launched with. Application
developers are responsible for implementing application specific mechanisms to restore the internal
state of the agents after a fault/restart. In order to facilitate that, the WadeAgentImpl class (that
as mentioned in section 2.5 should be extended by all application agents) provides the
getRestarted() method that can be used in the agentSpecificSetup() method to
distinguish between a normal startup and a restart after a crash..

41