Introduction

This document describes the rationale behind the architecture and design of Axis2. The Axis 2 sub components are those that were finalized at the first Axis Summit held in Colombo, Sri Lanka. The document will provide a thorough understanding of the workings of a SOAP engine that is required for a developer before examining the code in detail.

Architecture overview

Axis Java 2.0 is a SOAP engine consisting of a number of sub components. Components are identifiable with that of Axis 1.0, but the introduction of some components posses the capability of abstracting functionality to seamlessly support a number of auxiliary specifications. It is also important that the reader understands the inner workings of Axis 2 especially at component level, where dependencies and interactions between components are discussed.

Axis2 sub components

Axis2 consists of the following high level sub components, and they are described in the next section of this document.

1. Core ( Consisting of the Engine, Engine Registry, Contexts, Handler Framework and Phase Resolver)
2. Deployment
3. Axis Object Model (AXIOM)
4. WSDL Object Model.
5. Client API.

1. Core of Apache Axis2

The Big Picture

Just like Axis 1.x family Axis2 is all about processing Messages. It is important to note that Axis2 is not just about encoding of the messages from Java to XML and vise versa. Axis is a framework for the Web Service stack. All other components WS-Security, WS-RM etc of the Web Service stack fits in to Axis via the extension mechanism of the Apache Axis. Apache Axis is a framework for Web Service Stack. Axis SOAP Message processing always starts by running the information holder called MessageContext through the extension mechanism of Axis. This builds the foundation for Axis so as to give a generic framework for all the Web Service add on services.

Extension Mechanism

The extension mechanism of Axis is built with two types of components called Handlers and Phases. The Phases are placeholders for the Handlers and they can be configured via the configuration mechanism of Axis. When a web service is deployed the Phase-Resolver will order the registered Handlers inside a phase and phases will be ordered according to the configurations. When a web service is invoked the Engine will preserve the order of handlers. Once the handlers are assigned to the phases the phases are invoked in the order and in turn each Phase will invoke the contained Handlers.

The Axis contains some built in Handlers that drives the service invocation, yet the user can add new Handlers via the deployment Mechanism. The new Handlers will fit in to the Phases in the Execution Chain according to the Phase rules that is specified with each Handler.

Engine Registry

Axis starts by loading the EngineRegistry that holds the configuration of the Engine. This configuration includes the Services, Phases, order of the phases, Handlers, Phase rules for Handlers, transport and modules. It is the Deployment that creates the EngineRegistry out of the deployment mechanism. All other components of the Axis interact with the deployment via the EngineRegistry and is not aware of the exact deployment mechanism used. The default Deployment Mechanism is based on the XML configuration files, but other deployment mechanisms are also possible. For example if Axis is going to be deployed on to a device and the list of services are not going to change, one can hard code the Configuration.

Built in Handlers

Now let us focused our attention to the exact Handlers that will be invoked during execution. They are the inbuilt handlers that are explained before, the new handlers might plug in, yet following handlers are going to be there all the time.

1. TransportReceiver - Receiving the SOAP requests at the Server side or SOAP response at the Client Side. This can be a Servlet, Server socket, Maillet etc
2. TransportSender - Sending the SOAP message to Server (As a Request) or to Client (As response).
3. Sender - Initialize the outflow
4. Receiver - Handles the Message Exchange Pattern (MEP) and Sync Asynchronous Behavior of Axis. It currently supports only the InOutSyncReciver, but eventually it will support InOutAsyncReciver etc that would act accordingly.
5. Provider - Invokes the business logic of the Web Service, this can be JavaProvider, XMLProvider etc.
6. Dispatcher - decides which Web Service should be invoked at the Server Side.

These Handlers will create either an inflow or an out flow as shown in the diagram below. At the server side it will be an inflow followed by an outflow or a fault flow and in the client side out flow followed by the inflow.

Server Side Invocation

1. At the Server side the invocation is started by the Transport Receiver that listens for an incoming SOAP Message. Once the Message comes in, the TransportReceiver creates a MessageContext using the SOAP message and the information about the transport.
2. Then the TransportReceiver creates an AxisEngine and invokes the Axis Engine with the MessageContext. The Engine will invoke the Handlers that can be found without any knowledge about the Service to be invoked and then add the Dispatcher to the end of the Execution Chain and invoke the Execution Chain.
3. When the invocation reaches the Dispatcher, it will dispatch the Service and add the Service Level Handlers to itself. In other words the the Execution Chain grows itself when it reaches the Dispatcher. The Dispatcher will locate and add the appropriate Receiver based on the WSDL MEP's and the Synchronous and Asynchronous behavior.
4. Receiver will locate the Provider and invoke the Business logic of the Web Service and depending on the WSDL MEP's the outflow will be started (if required).
5. If the outflow is invoked, it will include the custom Handlers as specified by the EngineRegistry and will end with a Transport Receiver that would send the Response SOAP message back to the Client.

Client Side Invocation

1. Client Side invocation is driven by the Call object. The Call object will load the EngineRegistry (by default) .
2. When the Call is invoked the call will create a Message context with the SOAP message and the transport information provided by the user. Then the Call will start the Axis Engine using the MessageContext.
3. Axis Engine will locate the custom handlers if there are any and invoke the outflow with a transport sender at the end. When the invocation reaches the TransportSender, it will send the SOAPMessage and return.
4. Depending on the MEPs and the Sync/Asynchronous behavior the Call will start an inflow that include the custom handlers (if present), beginning with a Transport receiver with a new MessageContext. Note that the Call exhibits similar characteristics to the Sender at the Server Side.
5. Transport Receiver (TR) will wait for the input and when it reaches the TR it will fill the MessageContext out of the SOAP Message.
6. The Custom Handlers will be called with the MessageContext that is filled in and the SOAPMessage that is received at the end will return the user call

Out Line of the Each Component

• Axis Engine:-The Execution Model or the Engine in Axis terminology, addresses how the components of Apache Axis2 are executed passing control from one component to another. The Axis Execution Model is independent of specific Message Exchange Patterns or synchronous, asynchronous behavior. The Axis Engine consist of two main methods of message interchange, i.e. in-flow and out-flow respectively. In addition a fault-flow is also present for handling processing errors. Each flow encompasses a set of components, namely Handlers, Transport Senders, Transport Receivers, Receivers, Senders and Providers. These components address a specific requirement of when exchanging SOAP messages.
• Phases:- The Phases are a mechanism to specify the order of the handlers without knowing the *absolute position* of each handler in the Handler Chain. The Phase is a defined area (place holder for the handlers) in the Execution Chain. For an example say the handler developer wants (RM), the Handler to run after "Encryption and before "Transaction". But as the Handler can be used in different Web Services it is not necessary to specify the absolute position of the Handler at the Execution Chain. Phases will handle this with the following declaration.

• <handler name="RM" ..... before="Transaction" after="Encryption" ../>
  

  This will ensure that the rules are followed and hence executed appropriately (Note: support for before and after is not implemented in M1). As far as the implementation is concerned a Phase is a ordered collection of Handlers that are arranged according to the Phase rules. This can be viewed as an improved HandlerChain from Axis 1.x.
• Module:-Module is a logical collection of Handlers that act together and is more than just a Package. One such possibility is WSDL Code Generation extensions. For an example the users would like to have the ability to have Service reference to a module and have it enabled as done in J2EE.

  <service name="foo">
  
  <module ref="Authentication"/>
  
  </service>
  

  Please note that this would enable authentication for service foo, but is not included in M1.
• Engine Registry:-The Engine registry is the runtime state of the Axis engine. It contains real objects like Handlers etc. and by changing them at runtime the user can change the engine configuration. Deployment provides the implementation of Engine Registry out of the Deployment Descriptors and keeps the Engine Registry in sync with the Deployment configuration repository. In this way the Engine does not have to be aware of the deployment mechanism. The Engine would know that there is an Engine Registry and that it is maintained. By default the Engine Registry implementation is synchronized, and this is a necessary condition to support hot deployment. It is also possible not to synchronize the Engine but this would mean that hot deployment is disabled.
• Context:- Axis2 has three Contexts, Global, Session and the Message and they are the placeholders for the information in Axis2. They follow the mediator pattern and all the components interact and share the information with each other through the three contexts. The contexts enable Axis to build on loosely coupled components. Session Context is a property bag for session level information. The Global context consists of a property bag and the Engine Registry. The Message Context is the most important of the three, as it contains information about the message that is currently being processed. It is the Message Context that is passed as and when Handlers are invoked in order providing information at a message level. Among the information kept in the message context are the SOAP message, information about the transport, the references to other two contexts and a property bag. Most components of the Axis2 are stateless across the two invocations of the component. (The Engine Registry is part of the Global Context.) . The developer should store all states in the in one of the context while all contexts are accessible across the Message context which would be passed through the Handlers in the Execution Chain.

2. Deployment

There are two types of deployment in Axis2, namely, Service Deployment and Module Deployment.

2.1 Module Deployment:

Axis2 provides a J2EE like deployment (an axis archive file has to be created and dropped into the correct directory) for both modules and services. A user can deploy a module or a service as an .aar file (module1.aar, service1.aar). Module hot deployment will not be supported in M1, but will be supported in later releases. When the axis engine is started all the .aar files in the WEB-INF/Repository/modules will get deployed. But if someone wants to add a new module or service, a restarting of the engine is required after putting the new .aar file in the above-mentioned folder.

2.2 Service Deployment

One of the key improvements introduced with Axis2 is the capability to hot deploy web services. And in this M1 release only a J2EE like deployment is supported. That is, a user has to create an axis archive file that includes all the files that he/she wants, and drop that archive file into the WEB-INF/Repository/services directory. The following three features are associated with service deployment;

1. Hot Deployment is all about the ability to deploy a new web service while the axis engine is up and running.
2. Hot Un-Deployment is the ability to remove a web service (services) while the system is running.
3. Hot update is the ability to deploy a new version of an existing web service without restarting the server. It is handled as Un-Deployment followed by a Hot Deployment.

The directory structure to which modules and services should deployed are:

• WEB-INF/ modules
• WEB-INF/services

The required directory structure for a modules archive file is as follows;

m1.aar

META-INF/module.xml

lib/

classes/

The required directory structure for a service archive file is as follows;

s1.aar

META-INF/service.xml

lib/

classes/

The structure of module.xml file is as follows;

The structure of service.xml file is as follows;

The architecture of hot deployment consists of the following components;

1. The Scheduler periodically invokes the Listener to check for updates

2. If the Listener finds an update, it passes that information to the Repository

3. The Repository hands over the document to the Deployment Parser

4. Having parsed the document, the Deployment Parser returns the corresponding object

5. The Repository updates the toDeploy and toUn-Deploy list

6. The Repository informs the Listener to update the system

7. The Listener informs the Deployment Engine to do the update (both deploy and un-deploy)

2.3 Scheduler

This component itself is a thread that performs a specific task forever in a given time interval. In this case it periodically asks the Listener to listen to the file system events. Here the file system is not the entire file system, it is only the sub directories of WEB_INF/Repository where the modules and services reside.

2.4 Listener

As mentioned above the Listener listens for file system events. In order to do this it checks both the modules and services directories. Then it lists all the archive files in those two directories and compares those against the repository to check if a new service(s) has been added or if any service(s) has been modified. Then it informs DeploymentEngine to execute the required methods.

2.5 Repository

The Repository stores data about modules and services that have already been deployed. At the initialization process this loads data about all the modules and services into the module and service directories. It then deploys all those loaded modules and services. The repository stores the name of the archive file and its last modified date. It is possible to perform the following operations to the Repository;

• Add a new entry to the Repository
• Remove an entry from the Repository
• Modify an entry in the Repository.

These operations correspond to Hot Deployment, Hot Un Deployment and Hot Update.

In the add operation it checks whether the entry which is going to be added already exists in the Repository. If so it ignores the operation else it will add the entry to the Repository and add an entry to a list maintained in the DeploymentEngine (toDeploy list).

In the remove operation it directly removes the entry from the repository and adds an entry to a list maintained in the DeploymentEngine (toUnDeploy list)

In the modify operation it adds entries to both lists in the DeploymentEngine.

2.6 DeploymentEngine

The DeploymentEngine is the main component of the deployment sub component. It interacts with the axis engine and the engine registry by updating the engine registry when a new service is added or an existing service is removed. The deployment procedure is as follows;

• Check the toUnDeploy list. If it is not empty then remove the corresponding web service from the engine registry.
• Check the toDeploy list. If it is not empty then add the corresponding web service to the engine registry.
• For each and every item in the list that is passed to the DeploymentParser, create a corresponding axis service object and add it to the engine registry.

2.7 DeploymentParser

The following three types of xml document are parsed by the DeploymentParser. Its underlying parser is StAX.

• server.xml
• service.xml
• module.xml

Parsing of server.xml is done as follows;

The DeploymentEngine will create an AxisGlobal object and pass that to the DeploymentParser. Then the DeploymentParser will modify the object according to server.xml.

Parsing of service.xml is done as follows;

The DeploymentEngine will create an AxisService object and pass that to the DeploymentParser. Then it will be updated according to service.xml.

Parsing module.xml is the same as above except for it been passed an AxisModule object instead of an AxisService object.

2.8 DeploymentEngine Initialization

The Initialization process of the DeploymentEngine consists of the following steps:

• Checks whether server.xml is located in the repository directory. If it exists, then it is loaded and parsed. If not then a default server.xml is created and placed in the repository.
• Checks whether modules and services directories are available under the repository. If not they will be created.
• After server.xml is parsed, an AxisGlobal object is created.
• Load all the axis archive files to both the modules and services directories, parse them and create AxisService and AxisModule objects.
• If server.xml has references to modules, then it checks whether those modules are valid, (it is valid if it is in the modules directory),If it is not valid an exception is thrown.
  • If those modules are valid, the GlobalChains are built by the PhaseResolver (GlobalInChain , GlobalOutChain , and GlobalFaultChain) using phase rules.
• All the available services and modules will be added to the engineRegistry.
• The engineRegistry is returned.

2.9 Adding a new web service

Adding a new web service consists of the following steps;

1. Unzip the .aar file and take its service.xml and parse it using the DeploymentParser.

2. While processing the service.xml an AxisService object gets created and that object will be updated.

3. The created object is deployed to the DeploymentEngine.

4. Using the DeploymentEngine the service classLoader and the provider class will be added to the AxisService object.

5. The Handler chain will be built by resolving the phase rules.
(It should be noted here that the Handler Chain consists of the HandlerMetadata which has all the phase rules data and the actual executable handler )

6. The created object will be added to the EngineRegistry.

3. AXIOM AXIs Object Model

3.1 Introduction

AXIOM (also know as OM Object Model) is used to refer to the new and efficient XML info set model that has been developed for Axis2. Most of the XML object models used today are based on two major methods, namely;

1. DOM based where the whole document is parsed and held in the memory
2. Event based (SAX like) where the whole document is parsed at once, but no model is created in the memory and where the user has to catch the relevant events, which can not be stopped or reversed.

AXIOM gets the best from both these options and builds a better object model on top of a pull parsing methodology. It controls the phase of the parsing and builds a memory model if required from the information thats being parsed so far. That means at a given instance, AXIOM does not have a fully built object model with the rest of the information still in the stream.

The most important feature of this OM is that it is lightweight and is differed built based on StAX (JSR 173), the streaming pull parser.

The object model can be manipulated like any other object model (Such as JDOM), but underneath the objects will be created only when they are absolutely required. Hence this model is much more efficient.

AXIOM interacts with the outside world using the StAX API, that means it serializes and de-serializes using the StAX writer and StAX reader interfaces.

Since most of the data binding tools support SAX based interfaces, AXIOM comes with an adapter to be used between StAX and SAX.

3.2 High Level Architecture

AXIOM sees the XML input stream through the StAX stream reader, which is being wrapped by a builder interface provided. Current implementation has two builders, namely;

• OM Builder This will build a full XML info-set supported general XML model. But the current implementation lacks the support for Processing Instructions and DTDs.
• SOAP Builder This will build a SOAP XML specific object model. The Object model will contain SOAP-Specific objects like the SOAPEnvelope, SOAPHeader etc.

Each of these builders has the support for differed building and caching. User has the option of building the memory model or not. He can control this via setting the cache on or off.

(Since the object model is pull based, the StAX API is tightly bound to OM. To work with OM a StAX compliant parser and the API must be present in the classpath. )

The OM API works on top of the builder interface and provides for users, be they an engine developer, handler developer or anyone else. This will provide the highest flexibility as one can change builders and object model implementations completely independent to one another.

The OM has a defined set of APIs and one can implement his/her own memory model based on that. The current Axis2, comes with a linked list based implementation of those set of APIs. (There was an effort to build another OM on a table based model. Its now on hold.)

4. Handler Frame work and Phase Rules

4.1 Phase

Phase is a logically ordered collection of handlers. The handler inside a phase is ordered by phase rules. By using phase rules a user can specify where the handler should be logically placed. Some of the valid attributes associated with phase rules are listed in the following table :

* - If the before or after attribute of a handler is another handler, then the referenced handler must belong to the same phase. If the before or after attribute is a phase, there cannot be a phase attribute of that handler. If one of before or after is a handler other can not be a phase.

** - valid phase is a phase that is listed in the server.xml. If the phase name is not there then the phase name is not a valid phase.

*** - If both the PhaseFirst and PhaseLast attributes of some handler are true, then the case phase only has one handler.

N:B : If the user is going to use handlers and attributes as before, then there is no need to use PhaseFirst and PhaseLast because those are ignored by the rule engine.

4.2 Handler chain

The Handler chain is a collection of phases and handlers. The order of phases in the chain is described in server.xml. One service can have three handler chains;

1. Inflow
2. Outflow
3. Faultflow.

For each and every handler chain, handlers can come in different ways as follows:

• Corresponding flow of service.xml
• If there are any references to modules then the handlers from correspondence flow.
• Operation specific handlers corresponding flow

5 WSDL Object model

WSDL Module of Axis2 was architected with both WSDL version 1.1 and version 2.0 Component model in mind. The entire architecture of the aforesaid Module is built around an Object Model called WSDL Object Model which will be referred to as WOM here fourth.

5.1 Overview of the Axis2 WSDL Module

WSDL Module can be mainly broken down into two major functionalities:

• Service Description Functionality.
• WSDL Processing functionality(e.g. WSDL2Java, WSDL24J, Java2WSDL)

Service Description will provide an API to the Axis2 Engine that will expose the sufficient statistics about the web service that has been described in the WSDL file. WSDL Processing basically involves WSDL2Java, WSDL24J, and Java2WSDL. Following sections will give a further overview of each of the above functionalities.

5.2 WOM

WOM is engineered based on WSDL 2.0 component model, but it does not restrict its functionality to prior WSDL versions. Rather both WSDL 1.1 and WSDL 2.0 versions will be supported on top of the WOM. WOM consists of components such as Description, Interface, Service, etc. All those Components extend from one super interface called org.apache.wsdl.Component which will prove to be very useful in the implementation of the Service Desc.

WOM is a runtime representation of the WSDL file and it will provide the web service description functionality. As the following diagram illustrate the WOM will be the common Object model that will be used in bothWSDL2Java and Java2WSDL functionality. This intermediary object model ease the discussion since the WSDL processing can be broken down into the following four sub modules.

• WOM Builder.
• WSDL Emitter.
• Code Generator.
• Code Parser.

5.3 Service Description

Service Description (also known as Service Desc) is an API by which the necessary statistics will be made available to the Axis2 Engine at the runtime.

The functionality of the WOM is very much similar to the functionality expected from that of the Service Desc. Both behave as runtime description of the web service. Difference is that WOM is a clean component model that is not dependent on Axis2 or any other SOAP Engine and Service Desc is the Axis2 specific description of the web service. Thus it was necessary that the Axis Service Desc to extend the WOM to incorporate the additional Axis specific deployment information such as handlers, modules, providers, etc. The actual implementation of such extensions has been achieved using the extension capability provided by the WOM itself.

As mentioned above all the Components in the WOM extend from a super interface org.apache.wsdl.Component. Service Desc makes use of the functionality provided by org.apache.wsdl.Component to interface the WOM to behave as a Service Desc. org.apache.wsdl.Component has the following class diagram.

As the diagram illustrate the Component class provides the functionality of storing properties. Since all the WSDL Components extend from this class directly or indirectly, this functionality get inherited to all the WSDL Components. The Service Desc makes use of this functionality to store Axis2 specific properties using the WOM. In that sense the Axis2 Service Desc is a wrapper to the WOM.

Following is the Class diagram of the top level component of the description Component org.apache.axis2.description.impl.AxisService.

5.4 WSDL Processing

WSDL Processing can be identified as of operations performed on or performed using the WOM. Such definition is made possible because of the intermediary object model i.e. WOM always acts as an intermediary state. For example if WSDL2Java is considered:

WSDL2Java = WSDL-->WOM (WSDL2WOMBuilder), WOM-->Java (Code Generation Module).

The point to note is that the above allows the WSDL-->WOM (WSDL2WOMBuilder) to be identified as an independent module, not tied to the WSDL2Java operation. There are four such modules identified in the WSDL Processing Module of Axis2.

• WOM Builder.
• WSDL Emitter.
• Code Generator.
• Code Parser.

6. Client API

Types of Client programming models that Axis2 support:

1. Synchronous Invocation
2. Asynchronous Invocation using callback, with two way transport
3. Asynchronous Invocation using callbacks, transport is also one way (Addressing information is required.)

The following diagram describes all the invocations between sub components in the client side.

Call Class consist of following methods (Call API)

Callback Interface

AsyncResult 

Correlator

6.1 sendAsync Invocation

This invocation is similar to fire and forget where a request is sent and an acknowledgement is not expected. An invocation will consist of the following steps :

Code Snippet:

Sequence diagram

6.2 send Invocation

The service invocation is a void invocation. There is no return value, but there is a wait for an acknowledgment or a SOAP Fault, It consists of the following steps :

a -> call.send(SOAPEnvelope)

b -> engine.send( ..)

c -> Send the SOAP message

Code Snippet:

call.setTargetURL(URL)

call.setAction(String)

call.send(SOAPEnvelope)

Sequence diagram

6.3 sendReceive Invocation

The service method has a response and the communication happens synchronously using a bi-directional protocol. The Client hangs until the response (or fault) is returned.

a -> call.sendReceive(SOAPEnvelope)

b- > engine.send (..)

c -> Send the SOAP message

d -> Receive the response over the synchronous transport

w -> ProviderX will be called as the last step in engine.receive(..) 

e -> provider returns 

f -> Call hand over the response to the client

Code Snippet:

call.setTargetURL(URL)

call.setAction(String)

SOAPEnvelope env=call.sendReceive(SOAPEnvelope)

Sequence diagram

6.4 sendReceiveAsync Invocation

The service method has a response and the communication happens synchronously using a bi-directional protocol. Client DOES NOT hangs until the response (or fault) is returned. The Client uses a callback mechanism to retrieve the response. The Call API uses threads from a thread pool for each invocation.

a -> call.sendReceiveAsync (SOAPEnvelope, callbackObj)

p -> correlator.addCorrelationInfor(msgID,allbackObjRef)

b- > engine.send (..)

c -> Send the SOAP message

d -> Receive the response over the synchronous transport

w -> ProviderX will be called as the last step in engine.receive(..) 

q -> correlator.getCorrelationInfo(msgID)

g -> callbackObj.onComplete()

Code Snippet:

call.setTargetURL(URL)

call.setAction(String)

call.setListenerTransport(http, true) 

call.sendReceiveAsync (SOAPEnvelope, Callback)

Sequence diagram 

6.5 sendReceiveAsync Invocation with One way transport

The service method has a response and the communication happens asynchronously using a uni-directional protocol. The Client DOES NOT hang until the response (or fault) is returned. The Client uses a callback mechanism to retrieve the response. The Call API uses threads from a thread pool for each invocation.

a -> call.sendReceiveAsync (SOAPEnvelope, callbackObj)

p -> correlator.addCorrelationInfor(msgID,allbackObjRef)

b- > engine.send (..)

c -> Send the SOAP message

r -> Receive the response by the listener

s -> engine.receive(..)

w -> ProviderX will be called as the last step in engine.receive(..) 

q -> correlator.getCorrelationInfo(msgID)

g -> callbackObj.onComplete()

Code Snippet:

call.setTargetURL(URL)

call.setAction(String)

call.setListenerTransport(http, false)

call.sendReceiveAsync(SOAPEnvelope, Callback)

Sequence diagram