iDatamining.org

I am looking for projects to work on
Please contact with me at yiyu.jia@iDataMining.org!

Friday, April 29, 2011

Remove and reconfigure SSHD setting after running Zend Studio 8.0 repair installation.

I happened to run Zend Studio 8.0 installer to "repair" my zend studio 8.0 for IBMi. However, I got below error message after I click the folder under Sftp files and My Home directory in the Remote Systems View window.
ProxyHTTP:java.ioException: proxy error: Forbidden

It is strange because a dialogue was pop up and asked me to input password. So, that means studio can see the sshd server running on i5/OS.

I tried many ways. But, the final solution is simple: deleting the exsiting SSHD server setting from the Remote Systems window and create a new one. Then, a new key is generated for remote sshd server and connection works just well now.

Sunday, April 24, 2011

Using Abstract class and Interface in ExtJS 4.x programming.

I was driven to think about how to implement Interface and Abstract class when I review a project using Ext JS as front GUI library. It is good to see they try to do OOP and design the GUI component as reusable.

However, I saw that there was no clear idea about Interface and Abstract class design in the code. I say this because I saw some empty function in parent classes. Obviously, the programmer of parent classes wish subclass programmer to implement those empty functions. But, I did not see any code there to force subclass programmer to implement the function. The way to solve this is very simple, we only need to add a throw statement in parent class.

Ext.define('jia.blog.ParentClass', {
    greeting : function(){
                  throw "Unimplemented method.";
               }
}

Ext.define('jia.blog.ChildClass', {
    extend: 'jia.blog.ParentClass',

    greeting : function(){
                  alert('I implemented a abstract method');
               }
}

About using Interface in Ext JS, I think the new feature mixins introduced in Ext JS 4.0 can be good candidate as it is invented to solve multiple inherent. In Java, we know that one class can only extends from one and only one parent class. But, it can implement multiple Interfaces. So, for my opinion, I will declare my interface as mixins as below,

//a mixins class to be used as Interface
Ext.define('jia.blog.TechnicalManager', {
    level : 'manager',
    meetingWithPM : function() {
        throw "unimplemented method";
    }
});

//a mixins class to be used as Interface
Ext.define('jia.blog.Mother', {
    sex : 'female',
    feedChild : function() {
        throw "unimplemented method";
    }
});

Ext.define('jia.blog.WorkingMom', {
    
    mixins: {
        inWork: 'jia.blog.TechnicalManager',
        atHome: 'jia.blog.Mother'
    }

    meetingWithPM : function() {
        alert('I am meeting with project manager');
    }

    feedChild : function() {
        alert('I am feeding my kids');
    }

});


Of course, you can put "abstract" method in mixins too. I did not do it here just because I want to show how I use mixins as Interface in OOP manner. Actually, Javascript has no class at all. Class here should be called as Ext JS class as Ext JS 4.0 create the infrastructure of Class and Object. extend and mixins are both preprocessors in ExtJS 4.0. I list default preprocessors and postprocessors as below.

  • default preprocessors (defined in Class.js):
    'extend', 'statics', 'inheritableStatics', 'mixins', 'config'
  • default postprocessors (defined in ClassManager.js):
    'alias', 'singleton', 'alternateClassName'

Although ExtJS tries to implement a Class infrastructure for OOP, it is not a sophisticated OOP environment after all. For example, the throw statement we used above does not really force subclass developer to implement methods. Subclass developer will not know until he run the code. This happens there is no something like ExtJS compiler or interpreter. But, we can apply certain OO design concept with it. Also, we need to be careful not to make same properties or function names in different preprocessors or postprocessors as, obviously, ExtJS class functions will deal with these processors in sequence.

Monday, April 18, 2011

A modular architecture based approach for Single Page Web application development.

I happened to see two discussions on ExtJS forum (here and here) about how to organize large Single Page Application. ExtJS 4.0 introduce a MVC framework, which is similar idea as I mentioned about one year ago in my blog "We don't need MVC framework on the server side today". Implementing a MVC framework on browser side is good idea for organizing team work on a big Single Page Application. But, one obvious cost is that you are going to stick with it if you use a specific MVC framework, for example this new MVC from ExtJS 4.0. Also, it brings in extra learning curve for any new MVC framework.

MVC is a well known term for application developer. I know MVC when Struts is released more than ten years ago. However, MVC has been invented much earlier. Later, I was aware that Swing uses MVC. MFC use MVC too. I think a concept need to be cleared when we have so many things implemented in MVC pattern.

For my opinion, all these MVC implementations can be classified into two groups: desktop application MVC and Web application MVC. For example, Swing and MFC are desktop application MVC. Struts, Spring, Zend Framework are MVC implementation for Web application. Or from an other point of view, we can say desktop application MVC implementation is on the GUI component design level. And those MVC framework for Web application is on the application architecture design level. The selection of design pattern will affect how to organize team work on a big application. ExtJS 4.0 MVC is a Struts style MVC framework for pure client side programming. We say Javascript library like ExtJS let us be able to easily create a Single Page Application looks like desktop application. Now, ExtJS 4.0 introduce a Struts style MVC as its application framework. Will it really make code easier to main and better team work organization? I am afraid it is not true.

Back to those discussion in forums, those great detailed discussion are about how to organize a big single page web application. People all talk about MVC and intend to implement their own MVC framework. The interesting thing is, if we will treat our web application as Single Page Application, which kind MVC we are going to follow, the MVC for desktop application or MVF framework for Web application?

Let's have a look at MFC's document/viewer architecture diagram (linked from microsoft site). MFC declares it is an implement of MVC but the document contains both Controller and Model. I think Swing is designed in similar way. But Sun Microsystem names it as modified MVC.


Figure 1


For MVC framework for the Web application, I do not put a diagram here as it is well known to Web developers. Those framework always have a clear definition of Model, View, and Controller. The difference is that some controllers use configuration file to link View and Model, for instance Struts; Others use name conventions to link Model and View, for example, Zend Framework.

Now, we are making a Web application. And, we are developing a Web2.0 application, which looks and behave more like a desktop application instead of traditional multiple pages Web application. So, which type of MVC we should follow? For my opinion, it is good to follow the MVC design pattern employed in desktop application development. Since we all say modern Web2.0 application looks more and more like desktop application. What is the reason for us not to use methodology that have been successfully used for decades?

Before go to further discussion, I want to say something about Ajax library first. When Ajax became hot in industry in the past years, we saw many Ajax products in the market. They all declare that they are Ajax. But, according to my understand, these Ajax libraries can be divided into two groups: pure Javascript library running in browser only and those framework generates GUI widgets on server side. Personally, I do not like generating GUI widgets on the server side as it mixes many things together. It is messy and not flexible. I like libraries such as ExtJS, JQuery, and Dojo etc. I like them because using these pure Javascript libraries allow us implement a simpler and more scalable architecture than using one specific server side Ajax library can do. On the following picture, I present two ways of assigning tasks to Web developers. With pure Javascript Ajax library, we can assign in way (A). However, if you choose a server side Ajax library, you can only assign job in way (B). I like (A) as it means that I have real professional programmers who can focus on area, which they are professional and enjoying. And, method (A) implies that Javascript programmers do not need to care about what language server side uses. In the other words, server side programming language can be easily changed.


Figure 2


OK, now, suppose we select a pure client side Javascript Ajax library (group of awesome GUI widgets and more) for our Web 2.0 application (most likely an enterprise Web application instead of a public web site), where we will put our MVC? Well, I think we will have more than just one MVC here. Since I treat my Web application as a desktop application, I will go to MVC style employed by MFC or Swing. So, where will the MVC be in the architecture? Let's have a look at the following diagram.


Figure 3


As we can see, I present two different architecture designs in above picture. The whole application is composed by multiple modules. Each module may employee MVC inside. The MVC used here is something like modified MVC implemented in Swing and MFC Document/View architecture. For example, in ExtJS, don't you think the Grid widgets and Store can regarded as View and Document in MFC or View and Modal in Swing? But, in the application scope, can we use MVC too? Certainly we can. ExtJS 4.0 MVC is exactly the one. But, I believe that using modular programming concept will make things simpler than using MVC concept on this level for a "desktop application". I made my conclusion as described in above diagram. There are two ways to design the Web applications into modules. (A) represents a way to allow each module has its own interface and protocol to communicate with server side application for data. (B) represents a way to have a central proxy in Web browser to serve as a broker for two way data transfer. I prefer design (A) as it delivers an architecture design giving a clear ownership of module developing. That is, tasks can be assign to developers and they can work parallel.

In Figure 3, design (A) actually presents a Matrix modularization. It contains not only client side GUI function/feature module but also server side data service provider. Investigate Figure 3 with Figure 2, I will say that (B) in Figure 2 divided Web application into matrix modules. But, it assigns one matrix modules to one / one group of developers. I think the best way will be modularize the Web application in a matrix modularization way and assign tasks based on strategy presented in (A) of Figure 2. Modern Ajax libraries like Dojo, ExtJS and data delivering/representation techniques like RESTful Web service lhas make this kind design to be simpler and more scalable than others can do.

About the main application in the diagram. Its functions are more about initializing and destroying each module during their life cycle. It looks a little bit like Controller in MVC architecture. However, it is not because it mainly does two tasks,

  1. Load/unload modules. Looking into discussing threads mentioned at the begin, those functions like dynamically load sub application (module) should be implemented in this main application (entry application). 
  2. Support a message/event passing architecture. A event publish-subscribe architecture is good for connect each module because it make interface among module simple and flexible. 

So, according to my understanding, for a pure Javascript framework like ExtJS4.0, it actually used MVC on its widgets level already. It really makes thing more complicate when ExtJS 4.0 introduce another level of MVC design. Using modular design pattern for large Single Page Application is more nature way. It make the system design simple and task assign in team work easier and clearer ownership of coding.

Tuesday, April 12, 2011

IBM i5/OS V5R4 prestart job and database connection tuning.

Database connection tuning is an important part of application performance tuning, particularly, when applications uses connection pool. I noticed that some iSeries user does not care about this too much as I can not get answer when I ask them what is the limitation of database connection iSeries can support. The answer may not be simple. Or it can be as simple as no limitation as the default maximum number of QSQSRVR job and QZDASOINIT job is *NOMAX. This wont be good answer as the number of job should be optimized to a number, which fits your applications requirement.

On IBM i5/OS, there are two jobs used by database connection. They are SQL server mode prestarted job QSQSRVR and Database server prestart job QZDASOINIT job. QSQSRVR job is used when Database connection is coming through Command Level Interface (CLI). For example, PHP DB2 extension uses CLI to connect to DB2. Therefore, its corresponding job is QSQSRVR. Native iSeries JDBC driver (type 2) is using QSQSRVR job too. QZDASOINIT job is used when the connection is from ODBC or type 4 JDBC. Therefore, access from jt400.jar goes through QZDASOINIT job. When the type 4 JDBC is configured to use SSL, QZDASSINIT job is used instead of QZDASOINIT.

Both QSQSRVR job and QZDASOINIT job have their deamon job, which creates them. QSQSRVR job's deamon job is QYPSJSVR. Command WRKJOB JOB(QYPSJSVR/QYPSJSVR) can be used to do further investigate. For QSQSRVR job itself, we can use command WRKSBSD SBSD(QSYS/QSYSWRK) or command WRKJOB JOB(QSQSRVR) to do deeper investigation. For QZDASOINIT job, its deamon job is QZDASRVSD. We can use command WRKJOB JOB(QZDASRVSD) or WRKSBSD SBSD(QSYS/QSERVER) work with it further. Command CHGPJE can be used to change the configuration of prestart jobs. Also, don't forget that iNavigator is a convenience tool for us to investigate too: Work Management -> Server Jobs . The database server prestart jobs (QZDAINIT, QZDASOINIT, and QZDASSINIT) by default are shipped to run in subsystem QSERVER QZDAINIT) and QUSRWRK (QZDASOINIT and QZDASSINIT).

Also, there is another prestart job called QRWTSRVR, which is used by DB2 Connect driver. DB2 Connect uses DRDA protocol that is used to reduce the cost and complexity of accessing data in different DB2 supporting DRDA. In fact, in Work Management menu of iNavigator, you can see that iNavigator uses QRWTSRVR job. However, its license fee is expensive for enterprise edition. QRWTSRVR job typically run in QSYSWRK subsystem.

I made a diagram to present these three ways to connect to DB2 for iSeries. It is helpful for having a clear concept how the database connection is connect to DB2 and what is under those licensed and unlicensed program indeed. For example, people always ask why IBM gives type 4 JDBC driver for free and charge DB2 Connect. From this post, we can see jt400.jar, php db2 extension, and DB2 Connect use different protocol connect to DB2 for iSeries and different server jobs serve them. Furthermore, different protocols are designed for different usage. For example, DRDA is designed to support topology in complicate distributed network environment.


And, in Zend server for IBMi, there is a package called i5 Toolkit. It can be used to connect to i5/OS and access data too as DB2 for i5/OS is part of OS. A daemon called i5_COMD server is shipped together with Zend Server. i5 toolkit client communicate with it-COMD server for accessing i5 Data object. However, I feel it emulates 5250 terminal protocol and uses Remot Command Server prestart job QZRCSRVS. I am not able to discuss it here. I will spend some time to dig it later.

Here are some links from IBM for prestarted job on i5/OS V5R4 and checking job log of the appropriate "Host Server" and a PTF for improved management of QSQSRVR jobs.

Thursday, April 7, 2011

SQL naming must be used when calling user defined function in SQL Stored procedure on UDB DB2 for i5/OS

Today, I got a small tip, which let me aware that when SQL naming must be used when calling a Java User Defined Function in SQL stored procedure.

I wrote a UDF in Java for DB2 for i5/OS. A dummy code sample like below,

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.SQLException;

/**
* @author Yiyu Jia
*/
public class DummyUDF {

 boolean debugtableOn = true;
 public static String helloJia(String name)throws SQLException, Exception {
  //Connection con = DriverManager.getConnection("jdbc:default:connection");
  return "hello" + name;
 }
}

Then, we register it as DB2 user defined function as below,
CREATE FUNCTION YOURSCHEMA.DummyUDF ( 
NAME VARCHAR(500) 
) 
RETURNS VARCHAR(1000)   
LANGUAGE JAVA 
SPECIFIC YOURSCHEMA.DummyUDF 
NOT DETERMINISTIC 
MODIFIES SQL DATA  
RETURNS NULL ON NULL INPUT 
EXTERNAL NAME 'DummyUDF.helloJia' 
PARAMETER STYLE JAVA ;

However, when I call it in either iNavigator or SQL stored procedure, I was reported error message, which says
SQL State: 42704
Vendor Code: -204
Message: [SQL0204] DummyUDF in *LIBL type *N not found. ...


The statement used to call DummyUDF is as below,
values(YOURSCHEMA/DummyUDF('Yiyu'));
Since my iNavigator is configured to use system naming. I use "/" to link the schema name and function name. However, this wont work!After research and trying, I found it is extremely simply to solve this problem. That is, using SQL naming instead of System naming to call UDF though iNavigator is set to use system naming already. Below code works,
values(YOURSCHEMA.DummyUDF('Yiyu'));