According to Wikipedia,
Service Component Architecture (SCA) is a software technology used for composing applications that follow Service Oriented Architecture (SOA) principles. It is a development model that comes with many advantages including:
Separation of business logic from the details of its
service implementation.
Supports services in a multitude of languages including C++, Java,
COBOL, and PHP as well as XML, BPEL, and XSLT
The ability to seamlessly work with various communications constructs
including One-Way, Asynchronous, Call-Return, and Notification.
The ability to "bind" to legacy components or services, accessed
normally by technologies such as Web Services, EJB, JMS, JCA, RMI,
RPC, CORBA and others.
The ability to declare (outside of business logic) the Quality of
Service requirements, such as Security, Transactions and the use
of Reliable Messaging
Data could be represented in Service Data Objects
I would add,
Loose Coupling between the different modules (components).
I did implement a simple software using SCA technology, and I could see the power of SCA heterogeneity and its platform independence, with help of Tuscany Tutorial.
Today, I am looking at another model that seems a bit related. It is the Declarative Services Component Model (DS), which is a component model that simplifies the creation of components that publish and/or reference OSGi Services. In DS, an OSGI bundle seems to be wrapped as a component by adding an XML component declaration file to the bundle resources. The XML file generally contains declarations of the bundle services and references, something SIMILAR to SCA composite file. Here's an example of such file:
<?xml version="1.0" encoding="UTF-8"?>
<scr:component xmlns:scr="http://www.osgi.org/xmlns/scr/v1.1.0" name="it.eng.test.ds.consumer">
<implementation class="it.eng.test.ds.consumer.Consumer"/>
<reference bind="bindHappy" cardinality="0..1" interface="it.eng.test.ds.happy.IHappy" name="IHappy" policy="dynamic" unbind="unbindHappy"/>
<reference bind="bindSad" cardinality="0..1" interface="it.eng.test.ds.sad.ISad" name="ISad" policy="dynamic" unbind="unbindSad"/>
</scr:component>
My questions is: Is there any kind of relationship between SCA and DS?
Can DS achieve the SCA heterogeneity, and its component isolation? For example, Can DS provide services or references to/from different platforms like SCA components? Can a DS component be independent (isolated) in the same sense an SCA component is isolated?
DS and SCA are complimentary things. One is not a substitute for the other. You can use DS to build OSGi services. These services can be using in an OSGi framework. SCA can be used to describe a larger SOA design across multiple nodes. OSGi can be an implementation type for SCA components. So use DS for OSGi services when using OSGi as an implementation type for SCA.
I think both models serve a different purpose though they are both service based. DS is part of the OSGi framework and while it is possible to use remote services, but mainly ds is constrained to java/OSGi. There are frameworks like Apache Wicket which provide some kind of integration for declarative services in a web environment.
DS is a powerful framework for java/OSGi. One of the main distictions to most other frameworks I know is the dynamic side of OSGi. Services can come and go any time. Related to the concurrency converter service (in your linked example): In OSGi you could exhange the implementation at runtime, mock it for test purposes etc. An OSGi bundle (mainly a java project with meta information) may provider 0...n service components). Mostly best practice is to separate the service definition from the implementions into distinct bundles (unlike in the example you linked). To cut it short: Isolation would be better with ds, heterogeneity (support for different technological platforms) is not aimed at by ds (of course it could be achieved in one way or another if you make some effort). Hope that helps.
As B.J. mentioned in his answer, DS can be used as an SCA component implementation type. Similarly, Spring beans, POJOs or a BPEL process can be used as SCA component implementation types.
OSGi bundles aren't an implementation type but a packaging mechanism supported by SCA. In particular, the SCA Java POJO specification uses OSGi for Java artifact modularity. SCA provides additional modularity mechanisms such as composites (see http://java.dzone.com/articles/service-composition-modularity).
The SCA Java specification also outlines how dynamic wiring of services works but does not require SCA runtimes to support that capability. I don't know if Tuscany supports dynamic wiring but Fabric3 (www.fabric3.org) does. For example, Fabric3 supports dynamic service reference injection (addition, removal, update). This is particularly useful with multiplicity references (i.e. references that are collections of wired services). Dynamic wiring works for both local as well as distributed services.
Related
I've recently been given a project to work on that involves writing a web application. I've never done Java EE before. A lot of resources on the web are dated and I'm having trouble figuring out what the current differences are between the various standards and Java technologies.
Originally I thought I really needed EJB 3.1 due to dependency injection, JPA, session management, and web services. I started experimenting with Glassfish but was told that we had to write the thing in Tomcat. So I've been trying to figure out what I need as well as what and how to put into Tomcat to get there. I've begun to question whether I even need EJB at all.
I want to use JFS, I think, for the MVC architecture. In learning about that I've run into both ManagedBeans and CDI, which according to some renders the former obsolete and also seems to provide all the dependency injection stuff I want to enable unit testing. I've also come to realize that I can get JPA outside of EJB in the form of Hibernate and maybe a few others. Additionally it seems that web services, which I don't know that I need anyway, come in the form of another standard I can't think of the name right now and this also can be installed independently.
My major issue here is session management and state. It seems to me that all which remains for EJB to do is provide #Stateless/#Stateful and #Local/#Remote. However, as I understand it, some of this is already exists in the form of session management in the servlet container...but I don't know how much or what the major differences are that I need to account for in order to decide if I need these things at all.
So my question is, what are the basic, essential differences that I need to know in order to decide if EJB is worth looking at or if I have enough in the form of other libraries and technologies? I've been all over google and Usenet and have not been able to find this information anywhere.
Just thought of another bit. As I understand it, the #Stateful bean annotation provides me thread-safe state saving. I'm probably not going to directly use threads, but I know Java does so behind the scenes a lot and suspect EE especially so. When I need to keep state I don't want to be dealing with threads if this already provides it.
ManagedBean
Java EE 6 has three different ways of defining beans that are managed in one way or another:
#javax.faces.bean.ManagedBean
JSF 2.0 introduced this annotation declaring managed beans in faces-config.xml. This annotation is used to access a bean from the Expression Language.
#javax.inject.Named
In an EE 6 container CDI this annotation is a built-in qualifier types to provide a name to a bean, making it accessible through EL.
#javax.annotation.ManagedBean
This annotation attempts to generalize JSF managed beans for use elsewhere in Java EE.
If you deploy in an EE 6 container (If you use Tomcat or another servlet container, you can also get CDI by adding the Weld jar to your web app), then there is really no reason to use #javax.faces.bean.ManagedBean. Just use #javax.inject.Namedand start taking advantage of CDI sevices.
CDI
One of the objectives of CDI specification is to bring together the Web tier and the transactional services, making easy for developers to use EJB along with JSF in web applications of the Java EE platform.
With CDI you have the following services among others : well-defined lifecycle contexts(influenced by seam 2 and conversation scope), Dependency injection, loose coupling facilities like interceptors, decorators and events and portable extensions, which allows third-party frameworks to integrate in the Java EE 6 environment like SEAM 3 extensions
Managed beans and EJB Services
First of all CDI applies to any managed bean. Some managed beans are EJBs. When we need to use the EJB services in a managed bean, we just add a #Stateless, #Stateful or #Singleton annotation. IMO they act as complementary technologies allowing you to have a flexible and gradual development just only adding some annotations.
So, when should we use a session bean instead of a plain managed bean?
When you need some EJB features that CDI is missing : declarative transactions, concurrency management, pooling, remote or web service invocation ,timers and asynchronous method invokation.
Of course you could also get all aspects using third party libraries - but this would introduce additional complexity to your project. In terms of functionality IMHO EJB are the place to implement :
Business logic, allowing you to have a cleaning separation of busniness logic and web tier logic (implemented by "JSF backing beans" which are CDI managed beans but no EJB)
Functionality which makes most sense for components which are entrypoints to the application (endpoints for remote invocations delivered via RMI or HTTP)
Finally if you need EJB services then you need an Aplication Server (eg. GlassFish or Jboss AS) if you only need CDI services you need a Servlet Container(e.g Tomcat) plus CDI libraries.
Do you need the features provided by EJBs, i.e. security and transaction management? If the answer is yes, EJBs might be a good option.
If the answer is no, and you only need dependency injection, CDI could then be a good option.
You can also get similar capabilities with other 3rd party products, like Spring (dependency injection, Spring security, etc), but deciding whether you use one (EJB) or the other (e.g. Spring) is in many of the cases a matter of previous skillset.
In my opinion, if there are no previous constraints, going Java spec compliant is a good investment.
I would suggest you to start with the CDI and proceed to the EJB's (which is really adding one annotation on top of your POJO) if the requirements needs them (just as it was told - transactionality, web services, JMX, timers, EJB asynchronous invocations).
It's quite reasonable to develop an application in which your entry point is an EJB which encompasses your call in the transaction and allows you to define multiple entry points. The EJB then invokes the CDI beans with the business logic in them.
It's also worth noticing that the TomEE is a certified Java EE 6 Web Profile developed on top of the Apache Tomcat.
For example, Play-framework supports RESTful services like this: RESTful on Play! framework
How does this compare to something like Jax-RS Jersey implementation? Does a framework like Play run circles around Jersey because of all it's cool bells and whistles, and it does REST too?
Developer productivity is important, but so is a proper implementation. Perhaps using an MVC framework for REST only services is 'wrong'?
Note, only RESTful services, no UI components at all.
Even though it's not "wrong" to use an MVC framework for RESTful services, there are some pros and cons versus using a JAX-RS implementation.
(Disclaimer: I have only used Jersey and Play! for fun, and not on production-grade systems, so I have tailored my comments more generally to MVC vs. JAX-RS. Keep in mind that these are broad generalizations.)
MVC frameworks--at least the ones that are considered developer friendly and "slick"--typically save you from having to build a persistence layer (the model part). Most also simplify "routing" requests using either scaffolding via convention or some form of configuration. The downsides are that you have to conform to some conventions for your controllers and usually have to write a view for each resource (or build layers of abstractions to avoid rewriting the same code).
JAX-RS excels at defining the routing (using Java annotations) as well as eliminating any restrictions on the service class. In my experience, that has greatly reduced the amount of boilerplate code and developer overhead. Jersey and Apache CXF also handle the XML or JSON serialization using JAXB annotations, which eliminates the need to figure out the view in an MVC context. The downside here is that you have to figure out your own ORM or persistence layer, which could be good or bad depending on whether you're building on top of existing data or creating a greenfield system (or using something other than an JPA/RDBMS e.g. NoSQL data store).
My own personal comment: Play! is a really cool framework, but I'd choose CXF (or Jersey) over an MVC framework any day for building out a RESTful service. In my experience, this frees up the developer to focus on the logic needed for the service, and opens up options for different database approaches. Right tool for the right job.
As a rule of thumb: For Scala, use Play. For Java, use Jersey.
You can use Jersey/Scala and Play/Java; I've done both. It works. It isn't bad. But unless you have a particular reason to do that, I wouldn't mix ecosystems. Java and Scala are interoperable but they have different ecosystems, I would avoid adding Java-isms if you are using Scala or Scala-isms and dependencies if you are running straight Java.
Jersey and Play are generally close for REST services. Neither really has any killer features over the other.
Jersey defines URL mappings in annotations, Play defines them in a service wide route file. And they bundle or have varying quality of integration with different libraries for things like XML, JSON, database, testing, mocking, dependency injection libraries and app server deployment.
The Java world has JMS, Spring, JUnit, jdbi/hibernate/jpa, Jetty/Grizzly. The Scala world has Akka, specs2/ScalaTest, Anorm/slick. Jersey is a better fit for the first world, Scala for the second. You can definitely cross that, but it will be a little less elegant and might require more glue coding.
JAX-RS is a standard and implementations can be created by different vendors. Jersey is one such implementation. The other frameworks may make use of JAX-RS but are not standards. So it is not a one-to-one comparison.
I have never heard of Play before but it does look interesting, more akin to Rails and Django than Jersey. What I like about Jersey is that it can be integrated into existing Java web applications by simply adding the JARs and declaring some things in the web.xml. What I find confusing about Jersey and JAX-RS is the routing.
Play seems to make routing easier, however, correct me if I'm wrong, seems like it is an all-or-nothing framework and cannot be used alongside other servlets in the same web application.
I'm learning how to develop with OSGi in recent days, but facing a lot of problems.
I don't know how to design my client-server system based OSGi with Equinox as framework. Should I use socket connections to implement this or use a distributed model?
I'm also confused whether I need to import some third party packages to support my programming on socket, jdbc, swing, etc, or there're existing packages offered by Equinox or other OSGi open source projects?
First time to ask questions here, can anybody help me?
just take a look at OSGi Remote services implementations (more details in OSGi spec):
Web services (Apache CXF: distributed OSGi) - http://cxf.apache.org/distributed-osgi.html
Eclipse Communication Framework - http://www.eclipse.org/ecf/
Cheers,
Dmytro
The HTTP model is probably the simplest to implement, as opposed to protocols based on low-level socket handling and binary protocols.
The first question I'd ask is what kind of information your clients and server need to exchange. If you can live with HTTP (and in many cases you can), you can just use servlets on the server-side, http client libraries on the client side and a RESTful communications model.
Servlets are easy to implement in OSGi using the HttpService, see for example http://www.osgilook.com/2009/09/08/osgi-http-service-registering-servlets-on-the-fly/
Working with sockets is not more or less complicated in OSGi than it is in 'plain' Java. You design your application in much the same way as you would normally, but gain the benefits of modularity and services by using OSGi.
You could, for instance, decouple connection acceptance and request handling by introducing something like a WorkerFactory, or maybe reuse services on both the client and the server; those are the things that OSGi is good at.
OSGi as a 'technology' (as far as that notion holds water) does not mandate any specific interaction method. The compendium specification mentions Remote Services (a.k.a. Distributed OSGi), but it also contains an HTTP service specification is Servlets are your thing.
In short, just pick the interaction solution that best matches your scenario, and let OSGi help you with the services and modularity.
My question is very simple, my intention is to generate a repository with your responses so it could serve to the community when selecting frameworks for developing enterprise general purpose applications.
This could apply very well for general purpose languages such as C++, C# or Java.
What Framework do you recommend for generating Layered Architectures?
Based on you experience why do you prefer the usage of some Framework versus your own architecture?
How long do you believe your selected Framework will stay as a preferred option in the software development industry?
This is indeed an overly general question, especially since there are so many interpretations of the very word framework, and within the world of frameworks many different kinds for different tasks. Nevertheless, I'll give it a shot for Java.
Java
Java EE
The default overall enterprise framework of Java is called Java EE. Java EE strongly emphasis a layered architecture. It's a quite large framework and learning every aspect of it can take some time. It supports several types of applications. Extremely small and simple ones may only use JSP files with some scriptlets, while larger ones may use much more.
Java EE doesn't really enforce you to use all parts of it, but you pick and choose what you like.
Top down it consists of the following parts:
Web layer
For the web layer Java EE primarily defines a component and MVC based Web Framework called JSF - JavaServer Faces. JSF utilizes an XML based view description language (templating language) called Facelets. Pages are created by defining templates and letting template clients provide content for them, including other facelets and finally placing components and general markup on them.
JSF provides a well defined life-cyle for doing all the things that every web app should do: converting request values, validating them, calling out to business logic (the model) and finally delegating to a (Facelets) view for rendering.
For a more elaborate description look up some of the articles by BalusC here, e.g. What are the main disadvantages of Java Server Faces 2.0?
Business layer
The business layer in the Java EE framework is represented by a light-weight business component framework called EJB - Enterprise JavaBeans. EJBs are supposed to contain the pure business logic of an application. Among others EJBs take care of transactions, concurrency and when needed remoting.
An ordinary Java class becomes an EJB by applying the #Stateless annotation. By default, every method of that bean is then automatically transactional. Meaning, if the method is called and no transaction is active one is started, otherwise one is joined. If needed this behavior can be tuned or even disabled. In the majority of cases transactions will be transparent to the programmer, but if needed there is an explicit API in Java EE to manage them manually. This is the JTA API - Java Transaction API.
Methods on an EJB can easily be made to execute asynchronous by using the #Asynchronous annotation.
Java EE explicitly supports layering via the concept of a separate module specifically for EJBs. This isolates those beans and prevents them from accessing their higher layer. See this Packaging EJB in JavaEE 6 WAR vs EAR for a more elaborate explanation.
Persistence layer
For persistence the Java EE framework comes with a standard ORM framework called JPA - Java Persistence API. This is based on annotating plain java classes with the #Entity annotation and a property or field on them with #Id. Optionally (if needed) further information can be specified via annotations on how objects and object relations map to a relational database.
JPA heavily emphasizes slim entities. This means the entities themselves are as much as possible POJOs that can be easily send to other layers and even remote clients. An entity in Java EE typically does not take care of its own persistence (i.e. it does not hold any references to DB connections and such). Instead, a separate class called the EntityManager is provided to work with entities.
The most convenient way of working with this EntityManager is from within an EJB bean, which makes obtaining an instance and the handling of transactions a breeze. However, using JPA in any other layer, even outside the framework (e.g. in Java SE) is supported as well.
These are the most important services related to the traditional layers in a typical enterprise app, but the Java EE framework supports a great many additional services. Some of which are:
Messaging
Messaging is directly supported in the Java EE framework via the JMS API - Java Messaging Service. This allows business code to send messages to so-called queues and topics. Various parts of the application or even remote applications can listen to such a queue or topic.
The EJB component framework even has a type of bean that is specifically tailored for messaging; the message driven bean which has a onMessage method that is automatically invoked when a new message for the queue or topic that the bean is listening to comes in.
Next to JMS, Java EE also provides an event-bus, which is a simple light-weight alternative to full blown messaging. This is provided via the CDI API, which is a comprehensive API that among others provides scopes for the web layer and takes care of dependency injections. Being a rather new API it currently partially overlaps with EJB and the so-called managed beans from JSF.
Remoting
Java EE provides a lot of options for remoting out of the box. EJBs can be exposed to external code willing and able to communicate via a binary protocol by merely letting them implement a remote interface.
If binary communication is not an option, Java EE also provides various web service implementations. This is done via among others JAX-WS (web services, soap) and JAX-RS (Rest).
Scheduling
For scheduling periodic or timed jobs, Java EE offers a simple timer API. This API supports CRON-like timers using natural language, as well as timers for delayed execution of code or follow up checks.
This part of Java EE is usable but as mentioned fairly basic.
There are quite some more things in Java EE, but I think this about covers the most important things.
Spring
An alternative enterprise framework for Java is Spring. This is a proprietary, though fully open source framework.
Just as the Java EE framework, the Spring framework contains a web framework (called Spring MVC), a business component framework (simply called Spring, or Core Spring Framework) and a web services stack (called Spring Web Services).
Although many parts of the Java EE framework can be used standalone, Spring puts more emphasis on building up your own stack than Java EE does.
The choice of Java EE vs Spring is often a religiously influenced one. Technically both frameworks offer a similar programming model and a comparable amount of features. Java EE may be seen as slightly more light-weight (emphasis convention over configuration) and having the benefit of type-safe injections, while Spring may offer more of those smaller convenience methods that developers often need.
Additionally Spring offers a more thoroughly and directly usable security API (called Spring Security), where Java EE leaves a lot of security details open to (third party) vendors.
To specifically answer the second question:
Developing your own framework gives you the burden of having to maintain it and educating new developers in using it.
The larger your framework becomes, the more time you have to devote specifically to it and the less time you thus have to solve your actual business problem. This is okay if your business problem is the framework, but otherwise it can become a bit of a problem, even for very large companies that can dedicate a group of people to such a framework.
If you're a smaller company (say ~15 developer max) this can really become a huge burden.
Additionally, if your own framework is the kind of framework that can take advantage of third party developments (e.g. third parties can develop components for JSF), then your own framework obviously won't be able to take advantage of that.
Unless of course you open source your own framework, but this will only significantly increase the burden of supporting it. Just dumping your source code on sourceforge does not really count. You will have to actively support it. All of a sudden your framework becomes their framework with maybe 'weird' feature requests and awkward error reports for environments that you have no personal interest in.
This also assumes that your framework will actually be used by external users. Unless it's really very, very, good and you put lots of energy in it, this will probably not happen if it's simply the umpteenth Java web- or ORM framework.
Obviously, some people have to take up the job of creating new frameworks, otherwise the industry just stagnates, but if your prime concern is your business problem I would really think twice of starting your own framework.
Very vague question, I'm not really sure it's ever a good idea to "write your own" at this point for a work project (unless writing your own, IS the project). If it's a learning exercise, fine, but otherwise go use one of the libraries written by people who have been doing it far longer. If you really want to get involved, read their code, try and contribute patches etc.
For .Net there is Sharp Architecture Which is a pretty popular framework for layered applications.
Here's some of the stuff I use (I don't use Sharp Architecture)
First, the infrastructure stuff
For Dependency Injection, I use StructureMap. I use it because it's way more robust and performant than anything I would or could write, and it's very well supported within the .Net community. It also sticks to being DI, and doesn't venture out into other things that I might want to use other libs for (AOP etc). The fluent configuration is fantastic (but many .Net DI Tools have that now)
For AOP, I use Linfu Dynamic Proxy. I know a lot of people that like the code weaver variety for performance reasons, but that's always seemed a bit like premature optimization to me.
For a DataMapper, I use AutoMapper. This is one where I'm on again off again. If you can do your mappings based just on convention, then great, I'll use it. Once I have to start tweaking the configuration to do special things.... to me that starts to get into the gray area where the code might be more clear with just some left=>right wrapped in a function.
Web/UI
Asp.Net MVC. Although to be quite honest, I'm having a falling out lately and may soon be moving to FubuMvc. Asp.Net MVC seems like it has split personalities in terms of API design (dynamic over here, static over there, using blocks to render forms, but System.Actions to render other things etc). Combine that with the fact that it's not really OSS (you can't submit a patch), and to me there's a compelling reason why the community should come up with something better that's OSS.
Persistence
NHibernate, Specifically Fluent NHibernate. Sure I'd love to write my own OR/M, but at the same time I'm certain that the hordes of developers who have worked on NHibernate are way smarter than me.
Services/Distribution etc
WCF for Synchronous calls
NServiceBus for Messaging and most async calls.
Most of this stuff is OSS, so how long will it be around, well, I would imagine a good long while.
This question doesn't work very well. Selecting frameworks is difficult, and very context specific. For each selection process you might end up with a simple shortlist and a simple list of questions to answer, but those lists do not transfer well to other selections.
The number of parameters and the parameter sensitivity influencing a decision is very large, and at enterprise level a lot of them are not technical.
Currently, there are no frameworks available that are ready to support these near-term enterprise needs:
the switch for most of the workforce from pc to tablet and phone;
the switch from web client and rdbms to p2p/disconnected based storage and distribution
I am trying to get my head around OSGi Services. The main question I keep asking myself is: What's the benefit of using services instead of working with bundles and their exported packages?
As far as I know it seems the concept of Late Binding has something to do with it. Bundle dependencies are wired together at bundle start, so they are pretty fixed I guess. But with services it seems to be almost the same. A bundle starts and registers services or binds to services. Of course services can come and go whenever they want and you have to keep track of these chances. But the core idea doesn't seem that different to me.
Another aspect to this seems to be that services are more flexible. There could be many implementations for one specific Interface. On the other hand there can be a lot of different implementations for a specific exported package too.
In another text I read that the disadvantage of using exported packages is that they make the application more fragile than services. The author wrote that if you remove one bundle from the dependency graph other dependencies would no longer be met, thus possibly causing a domino effect on the whole graph. But couldn't the same happen if a service would go offline? To me it looks like service dependencies are no better than bundle dependencies.
So far I could not find a blog post, book or presentation that could clearly describe why services are better than just exposing functionality by exporting and importing packages.
To sum my questions up:
What are the key benefits of using OSGi Services that make them superior to exporting and importing packages?
Addition
I have tried to gather further information about this issue and come up with some kind of comparison between plain export/import of packages and services. Maybe this will help us to find a satisfying answer.
Start/Stop/Update
Both, bundles (hence packages) and services, can be started and stopped. In addition to that they can be kind of updated. Services are also tied to the bundle life cycle itself. But in this case I just mean if you can start and stop services or bundles (so that the exported packages "disappear").
Tracking of changes
ServiceTracker and BundleTracker make it possible to track and react to changes in the availability of bundles and services.
Specific dependencies to other bundles or services.
If you want to use an exported package you have to import it.
Import-Package: net.jens.helloworld
Would net.jens.helloworld provide a service I would also need to import the package in order to get the interface.
So in both cases their would be some sort of "tight coupling" to a more or less specific package.
Ability to have more than one implementation
Specific packages can be exported by more than one bundle. There could be a package net.jens.twitterclient which is exported by bundle A and bundle B. The same applies to services. The interface net.jens.twitterclient.TwitterService could be published by bundle A and B.
To sum this up here a short comparison (Exported packages/services):
YES/YES
YES/YES
YES/YES
YES/YES
So there is no difference.
Additionally it seems that services add more complexity and introduce another layer of dependencies (see image below).
alt text http://img688.imageshack.us/img688/4421/bundleservicecomparison.png
So if there is no real difference between exported packages and services what is the benefit of using services?
My explanation:
The use of services seems more complex. But services themselves seem to be more lightweight. It should be a difference (in terms of performance and resources) if you start/stop a whole bundle or if you just start and stop a specific service.
From a architectural standpoint I also guess that bundles could be viewed as foundation of the application. A foundation shouldn't change often in terms of starting and stopping bundles. The functionality is provided by services of this packages in some kind of dynamic layer above the "bundle layer". This "service layer" could be subject to frequent changes. For example the service for querying a database is unregistered if the database is going offline.
What's your opinion? Am I starting to get the whole point of services or am I still thinking the wrong way? Are there things I am missing that would make services far more attractive over exported packages?
Its quite simple:
Bundles are just providing classes you can use. Using Imports/Exports you can shield visibility and avoid (for example) versioning conflicts.
Services are instances of classes that satisfy a certain contract (interfaces).
So, when using Services you don't have to care about the origin of a implementation nor of implementation details. They may even change while you are using a certain service.
When you just want to rely on the Bundle Layer of OSGi, you easily introduce crosscutting dependencies to concrete implementations which you usually never want. (read below about DI)
This is not an OSGi thing only - just good practice.
In non OSGi worlds you may use Dependency Injection (DI) frameworks like Guice, Spring or similar. OSGi has the Service Layer built into the framework and lets higher level frameworks (Spring, Guice) use this layer. - so in the end you usually dont use the OSGi Service API directly but DI adapters from user friendly frameworks (Spring-->Spring DM,Guice-->Peaberry etc).
HTH,
Toni
I'd recommend purchasing this book. It does an excellent job explaining services and walking through the construction of a non-trivial application that makes use of OSGi Services.
http://equinoxosgi.org/
My company routinely builds 100+ bundle applications using services. The primary benefits we gain from using services are:
1) Loose coupling of producer/consumer implementation
2) Hot swappable service providers
3) Cleaner application architecture
When you start with OSGi, it is always easier to start with an export-package approach it certainly feels more java-like. But when your application starts growing and you need a bit of dynamicity, services are the way to go.
Export-package only does the resolution on startup, whereas services is an on-going resolution (which you may want or not). From a support point of view having services can be very scary (Is it deterministic? How can I replicate problems?), but it is also very powerful.
Peter Kriens explains why he thinks that Services are a paradigm shift in the same way OO was in its time. see µServices and Duct Tape.
In all my OSGi experience I haven't had yet the occasion to implement complex services (i.e. more than one layer), and certainly annotations seem the way to go. You can also use Spring dynamic module to ease the pain of dealing with service trackers. (and many other options like iPOJO, and Blueprint)
Lets consider the two following scenarios:
Bundle A offers a service which is an arithmetic addition add(x,y) return x+y. To achieve this, it exports "mathOpe package" with "IAddition interface", and registers a service within the service registry. Bundles B, C, D, ... consume this service.
Bundle A exports "mathOpe package", where we found a class Addition exposing an operation (x+y)<--add(x,y). Bundles B, C, D, ... import the package mathOpe.
Comparison of scenario 1 vs. scenario 2:
Just one implementation instance vs. many instances (Feel free to make it static!)
Dynamic service management start, stop, update vs. no management, the consumer is owning the implementation (the "service")
Flexible (we can imagine a remote service over a network) vs. not flexible
... among others.
PS: I am not an OSGI expert nor a Java one, this answer shows only my understanding of the phenomena :)
I think this excellent article could answer a lot of your questions:OSGi, and How It Got That Way.
The main advantage of using a service instead of the implementation class is that the bundle offering the service will do the initialization of the class itself.
The bundle that uses the service does not need to know anything about how the service is initialized.
If you do not use a service you will always have to call kind of a factory to create the service instance. This factory will leak details of the service that should remain private.