Tuesday, December 22, 2009

Recurring patterns in applications

I was just doodling around the architecture of my latest project when I suddenly realize that I have done the same thing over and over again for every project. I'm sure anyone of you who has successfully delivered more than ,say ,5 projects would realize that there is a pattern to almost 80-90% of stuff out there.

I started to scribble some of the recurring patterns I have seen again and again in many projects. I'm putting it here and am very greatful if anyone could add more to this. Now to verify what I'm saying is in fact recurring, I'm going to examine the presense of these patterns in three different context a) A social network ala facebook b) A workflow application c) A content management system (CMS)

1) All system must have Users

a) Social network : Yes
b) Workflow app: Yes
c) CMS: Yes

2) Users live in a Context (for example, roles, location etc.)
a) Social network: Yes. Two very obvious context: logged in logged out.
b) Workflow app: Definately yes. Swimlanes (roles) can be considered context
c) CMS: Yes. There are plain old users, there are admins, there uploaders, there are downloaders etc.

3) Given a User and a Context, there are a few Lists of Items.
a) Social network: YES. When you log in to facebook or twitter there is a list of entries/microblog
b) Workflow app: YES, you would have a list of tasks (either done, pending, kiv, delegated etc.
c) CMS: YES. You would have a list of contents

4) An Item can contain either basic data: texts, numbers, links, images, videos, file attachments or links to other Items
a) Social network: Yes. You have all the classic types of data but also replies and "retwitts" etc.
b) Workflow app: Yes, you would have tasks descriptions and attachments for example. If the item is delegated, you will also see the delegated items
c) CMS: Yes. You will see the content and if you have "folders", you can have links to other data.

5) An item is aware of whom (Users) and how (Contexts) it should be accessed
a) Social network: Yes, only users and probably their friends could access certain photos for example
b) Workflow app: It is obvious with swimlanes, certain users canonly access certain data. What is more interesting is that certain users can access part of an item while others can only access other parts (think of government forms where you have "For Office Use Only" part)
c) CMS: Yes, folders have rights much like a Unix file system.

6) A User could be in a Relationship with N other Users. We call this relationship Contact.
a) Social network: Yes, obviously
b) Workflow app: Yes. In most workflow applications I've seen, Contacts are amed 'Supervisors', 'Subordinate' or 'Team member'.
c) CMS: You can certainly share content with other people. Not formally a Contact but still uses the same principles.

7) User should be able to log in to the system and eventually log out
a,b,c) Yes, obviously

8) A user can create Item and make it accessible to other Users (either from within his Contact or not)
a,b,c) Yes, obviously. This puts the responsibility of creating an Item on the User's hand.

9) User who creates an Item is responsible for the lifecycle (CRUD) of that Item
a,b,c) Yes, obviously

10) User who has access to an Item should be able to Read and optionally Update an Item
a,b,c) Yes, obviously

11) A system should be able to listen to the lifecycle of Items. User defined filters could be defined at each lifecycle action.
a) Social network: Yes. This is similar to Facebook email alerts when people reply to your post. The reply can be considered an Item, creating a reply is an event listened to by the system.
b) Workflow app: Yes. At every stage of a workflow, a filter can be triggered (I think Adempiere works this way to update its account database).
c) CMS: Yes. Again, email alerts can be sent when items are uploaded/created/deleted

12) There are 2 types of lifecycle action 1) Rules: To validate the lifecycle action
a) Social network: Yes. An email alert should only be sent if the new item created is a response to an existing item.
b) Workflow app: Yes. A rule is always attached to every "decision path" (the exclusive choice pattern) in a workflow.
c) CMS: Yes. A rule for example that states when a non-empty forlder cannot be deleted. Or a content viewed by another user cannot be deleted.

13) Type 2) Events that could trigger lifecycle actions on other objects.
a) Social network: Yes, adding a Contact (I.E. adding another person into your Contact) triggers an event that will add you to the contact of that other person. (I add u as friend implies, you will also add me as a friend).
b) Workflow app: Yes, for example in Adempiere, a workflow that creates an Invoice triggers the modification of Account.
c) CMS: Yes. Recursively deleting a non empty forlder might delete all its contents.

14) Users can do CRUD lifecycle on their Contacts. These actions will also generate Events.
a,b,c) Yes, obviously. You can always delete and add new people to your contact.

Now why are these patterns important? Well imagine now I've created a bare bone application (call it a framework if you will) and imagine that I want to create a Social Network. All I have to do is customised that "framework" and within a short time, tada~ a social network. I have only to define what is Item in my application context (which is social network), what is a User, what Events need to be generated and what Rules need to be triggred and I'm done. Someone else might define Items as Account, Activity, Invoice, Inventory etc. They might define Events as CreditToAccount and DebitFromAccount etc. and define Rules such as "When Inventory is low, start Procument process" and bam!, the framework is now an ERP.

Now let us take this further, what if this framework define not just bare bone stuff but also things like open persistence policy (which allows you to connect your app to any database), a well thought out GUI, a caching mechanism out of the box, a clustering facility , a reporting and BI facility and some security enhancement here and there. Suddenly, I can focus on creating my latest shiny new social network or create my own ERP focusing mainly on what the application should do rather than all the nitty gritty stuff like scalability.

Of course, what is defined here is probably incomplete at best and naive at worst. There are overlaps I'm sure (like Rules and Context could be the same thing). If you do have other stuff to add, please do enlighten me. (Or if you think certain things are just superflous). Maybe creating this framework could be my next big FOSS project :)

Saturday, December 12, 2009

Weld DI and Glassfish Example

Last Thursday I gave a presentation to a few Experian staffs on Weld DI. You can find the presentation here:


Examples can be found here:


Tuesday, December 8, 2009

Message Driven POJO with Java Content and Dependency Injection API (Weld) a.k.a Look ma! No XML!

I am supposed to be presenting 'Something new and exiting in Java' to staff of Experian Malaysia. Looking around, it seems that I am frekin lucky since we have a few new and exciting things in the Java world. Java EE 6 is finally here! (Phew!~), Netbeans 6.8 RC 1 is out! (with support for Java EE 6 - you can create a session bean right in your war file - I kid you not!) and finally Java CDI (I think, JSR-299) is finally here too!.

I decide to play around with CDI implementation by JBoss (called Weld). There are other implementations out there notably CanDI by Caucho and OpenWebBeans by Apache.

Anyway, CDI allows you to do strong typing dependency injection without XML (of course, XML is still there just in case you need it). Weld can be downloaded here [http://seamframework.org/Download]. A full tutorial by JBoss teamcan be found here [http://docs.jboss.org/webbeans/reference/current/en-US/html/index.html]. Note that the tutorial was created about a year ago [on the internet scale, that is the equivalent of the time between the Triasic era and the French Revolution], so some things are not up to date. A better approach is to learn stuff off the examples in weld zip file.

Here's an example of CDI:

With CDI, we can create Java Beans with a particular scope. In this example we have a Singleton scope. This means that there will be a single instance of SomeService for the whole application.

public class SomeService{
public String doSomething(String parameters){
//do something here
return result;

To access the service, we can just inject it to a class

public class Client{
@Inject SomeService someService;

public void run(){
//use someService

Note that we do not need to invoke any constructor or factory of SomeService in Client. The lifeCycle of SomeService is managed directly by Weld.

Anyone who uses Spring before knows that Spring can also do much of the same thing. Its just that Weld can do all this while being strongly typed and xml-less (Maybe Spring does make XML optional these days, I have not checked)

Now the big finale, one use case of Spring that I find fascinating is Message Driven POJO (MDP). So, I wanted to create one using Weld.

OK, since I don't have much time, I 'cheated' a bit by using ActiveMQ libraries and by making my MDP works only with ActiveMQ... but hey, it works >:) [Maybe I should call it Message Driven ActiveMQ Producer/Consumer instead ;) ]

I based my implemetation on this [http://developers-blog.org/blog/default/2008/10/28/A-simple-ActiveMQ-example] (Thanks Gisbert) .

After much toiling (my first implemetation uses Spring libraries... a pile of fun that turn out to be), I manage to create an MDP using Weld.

An MDP looks like this:

public class Processor {

public void processRequest(RunnableMessage runnable){

public void doSomething(String string){
System.out.println("Received string="+string);

The Consumer annotation indicates that this object is ready to receive messages. The messages are in fact POJOs themselves, so when a message arrives, depending on the type of the message (or rather the type of the payload of the message), a method within the Processor object is chosen.

To send a message, I just do this:

RunnableMessage r = new RunnableMessage(); //r here is just another POJO

And processRequest will be invoked.
If I do:

activeMQService.sendMessage("If you type google into google, you will break the internet");

Then I will invoke the method doSomething instead.
Cool eh :)

Anyway, the source code with all its dependencies are here [http://www.freedrive.com/file/1045164,weldmessage.zip]. To run this, you need to download ActiiveMQ
[http://activemq.apache.org/download.html] and run it. (Unzip the file and run the executable in the bin directory). Next, just run the example.

Have fun!