[Disclaimer: This post had a point. ChannelFactory vs. Proxy vs. Hand Rolled Proxy. That point got lost along the way.]

The “application” that I am working on is more of a platform then a single application. Our solution contains over 150 projects and is organized into what we call, sub-systems. A sub-system is basically a stand-alone application. sub-systems contain web, shared, service, configuration and testing projects. They can be deployed independently and are designed to be one piece in a larger SOA. These sub-systems are broken down into 5 categories of work. Web (MVC/Silverlight/API), Services, Data (Entities/NHibernate), Queuing, and Support.

For the last years worth of development, with the exception of the silverlight clients talking back to the web server via service reference generated proxy classes, we’ve been using a Service Factory/Locator pattern. Currently the development team is considering the merits of this implementation and possible alternatives.

Personally; I really like our current design, it follows the basic tenants of SOA and provides many benefits. The benefits out-weight the drawbacks and the drawbacks SHOULD be easy to overcome.

Cross Sub-System Communication

Every sub-system has a “shared” library project that houses it’s interfaces. This library is designed to hold only items that need to be shared between sub-systems. Service contracts, request and response messages and data contracts are typical items that should be in the “shared” library project in the sub-system.

When SS-A wants to call a service on SS-B, SS-A adds a project reference to the SS-B’s shared project. Next, SS-A uses our ServiceFactory class to locate and/or create the channel and service client. Now SS-A can work with SS-B’s methods as if they were local and without the need for a service reference or generated proxy class.

Screen-shot here.

We use this pattern because we have control of both the client and server code.

1. Separating the web and app tier.
2. No need to update ref/proxy when service changed.
3. Compile time checks.
4. Testable – Not that we are doing it.
5. Less code to maintain.

Problems With Current Implementation:

1. Can’t mock ServiceFactory – I shouldn’t say “can’t”, I just don’t think we’ve tried. Are tests are shit right now anyway.
2. The Shared project in every sub-system has become a dumping ground for implementations that don’t fit into our current project structure. For example, validators, extensions, repositories, and on and on are in these shared projects.

Shared Sub-System Bloat

At first this does not sound like a very big deal until you start to work with it. Every class added to the shared projects could potentially have a dependency on another sub-system. When this happens the entire platform starts to break down into a spider-web of dependency hell.

No really! This is the fucking problem! When I add a reference to a shared project I, by-proxy, add a reference to up to 40 other projects! The shared project in each sub-system should not have ANY references. It should only contain interfaces and/or contracts.

“The shared project in each sub-system should not have ANY references”

Serving & Consuming SOA

There are 2 main sub-systems that act as the services in our SOA. They reside on our application servers, have access to the database, don’t (shouldn’t) communicate with each other and provide a bulk of the business logic.

The consumers of these sub-systems include a web application, console applications, windows services, silverlight web hosts. These systems use the ServiceFactory class to communicate with the application servers.

F’ing up the program

The silverlight systems have another hop as they have to send the data down to the client. Our original design was to use RIA Services to emulate being on the server and hopefully write less code. However; since we moved to visual studio 2010 and couldn’t get .NET 4.0 on the servers our team decided to remove RIA.

Now we have a situation where the Silverlight clients need communicate back to their web servers. We used “Right Click-> Add Service Reference” to generate the proxy and a service locator to create the client.

Knee-Jerk Reactions

Our dev team, for some crazy reason, is suggesting that having a hand generated service proxy will be easier to mock then one generated via “Add Service Reference”. Also, they are adding “ServiceFactory” like code to the silverlight clients. I don’t know why we can’t use our current implementation by using linked classes in a silverlight class library and more importantly: How this is better in any way?

Proxy Class Fail

I feel that the champions for hand rolled proxies are looking at the problem idiosyncratically. The extra work required to keep the client and server code in sync (for both Silverlight and inside the platform) would FAR out weight the benefit of hand rolling proxies.

I just don’t think they realize the amount of use the ServiceFactory gets throughout the platform.

Testing WCF

I am far from a testing expert but I am pretty damn good at the googles.

1. http://mokosh.co.uk/post/2009/04/19/prism-2-wpf-and-silverlight-services/ – (we use prism)
2. http://houseofbilz.com/archives/2008/11/18/testing-wcf-service-apps-part-0-of-4/ – (we use RhinoMock)
3. http://stackoverflow.com/questions/2509316/best-way-to-mock-wcf-client-proxy – (client proxy)

Governance

SOA is a complex technology theory that can’t be bought. It requires buy-in from a team, developers and someone to watch over the implementation. Every developer thinks they understand SOA, when asked to implement it they create services and then complain that they don’t need SOA.

When I spent Christmas even developing across 4 layers just to get a fucking coupon onto a page. I bitched but I didn’t break the existing convention. I spent endless hours trying to understand how to fit MVC and SOA into the same project. During my development vietnam, SOA was a pain-in-the-ass, but now that we have a “stable” codebase and have 20/20 I can see how SOA will save our ass when the time comes. 

Coup d’état

A bunch of words go here. They aren’t important. Just know that when implementing SOA the development staff will constantly say things like “this would be so much easer if I could call the database”. SOA is not for developers; it’s for the project and the organization (much like TFS).

SOA is not for the developers; it’s for the project and the organization.

Concluding

I think that Silverlight clients should use “Add Service Reference” to generate proxies for a few reasons:

1. Right click “Update Service Reference” is much easier then keeping client and server code in sync.
2. We are in pretty rapid development.
3. We can mock the service.

Silverlight clients could, not for us, not now, use nant to keep the proxies up-to-date.

All service communication from the web to app tier should use the ServiceFactory pattern.

Console applications, windows services don’t use a consistent pattern. We need to resolve that.

Explicitly Covering My Ass

If any of my co-workers happen to read this: I wrote most of the code I’m suggesting needs to be fixed. It’s how I (we) spent my (our) holidays.

I’m just putting some thoughts down so I can understand the situation. I’ve rewrote this a few times when I grasped certain concepts/implementations/reactions/reasons/etc.

I think this sums it up. Now I need to get that one last question answered.

Why check in proxies for sub-systems that we have control over the client and server?

No related posts.

Azure WCF development

There is a patch you must install that creates a local “load balancer” that fixes an issue with requests to and from AppFabric services. You’ll also have to update the behavior.

http://code.msdn.microsoft.com/wcfazure/Wiki/View.aspx?title=KnownIssues

Azure Initialization

After publishing your services might hang, if they don’t start in around an hour, try stopping and restarting them.

http://blog.toddysm.com/2010/01/windows-azure-deployment-stuck-in-initializing-busy-stopping-why.html

Azure Instances

• You can’t set the number of running instances for any service to 0. If you do you get a Dr. Watson exception thrown in the Azure UI….Really; Dr. Watson!
• You can’t have more then 5 web roles, again cryptic Dr. Watson error message

Azure Upgrade

Upgrade means upload new code, when you do this you can’t change the configuration. If you need to add/remove/rename roles you will have to delete and redeploy.

WP7 Service References

You can only have one service reference in the current CTP.

Upgrading WP7 CTP to April

I tried it; following all the correct instructions the application was running but would bomb randomly. Recreating the project from scratch (copy/paste the code) fixed the issue.

WP7 Panorama

By far the best and most stable panorama: http://phone.codeplex.com/

Related posts:

1. ASP.NET MVC in Action Personal Notes Keep using a custom controller Learn and Use Ajax...

Quick tip:

If you want to put your custom controls in a separate assembly you need to include the layout (generic.xml) in that assembly.

Related posts:

1. Thoughts on Silverlight In a recent post Jamie Kosoy, who I believe is...
2. AutoMapper.Silverlight I could, and will soon, write a long ass post...
3. Silverlight, Actions and Prism When using Prism for composition in Silverlight how do...

While my solution compiles, 80 projects, I started working on ASP.NET MVC CodeTemplates for Spark View Engine. They are available in my GitHub repo. Feel free to fork and extend!

http://github.com/detroitpro/spark/tree/master/src/CodeTemplates/

Related posts:

1. Spark Binaries For ASP.NET MVC RC2 THIS IS NOT AN OFFICAL RELEASE…YET I spent the evening...
2. ASP.NET MVC Spark Support ASP.NET MVC 2 supports Spark! What? hear me out. The...
3. Spark with VB.NET Project Demo #2 Outline: Status of using spark in a VB.NET project Bugs...

Finding talented developers is always a challenge and in the current economic climate many developers are holding off moving to a new position. As such, contract houses are submitting any available candidates for every position. This practice is not only another reason to avoid contracting shops but is rather taxing on the company looking to fill a position.

The man hours spent screening a single applicant can quickly balloon over 8 man hours. My current client recently felt this pain and decided to require the consulting houses to have their candidates fill out a questionnaire. In an environment with many different technologies and developer background experiences coming up with a questionnaire that covers technical, theoretical as well as passion is a daunting task.

Below is one of the recent questionnaires I filled out that I thought balanced technical and theoretical questions well. Some of the details have been redacted. And oh’ Be warned, I have no idea if these answers are correct

What is reflection?
reflection is a way you can interrogate (inspect and/or invoke) objects in a running application.

How do you inherit from a class in C#? (give a code sample)
public class GreatClass:Baseclass {}

When you inherit a protected class-level variable, who is it available to?
The derived class and of course the base class.

Describe each accessibility modifier (public, private, etc).
Public: can be accessed by anyone
Private: can only be accessed from inside the base class
Protected: can be accessed by derived and base class
Internal: like public by also must be in the same assembly
Protected Internal: like private by also must be in the same assembly

How is method overriding different from overloading?
overriding changes the functionality of a base method, overloading creates another method with different inputs/attributes (post in MVC)

What does the keyword virtual mean in the method definition?
virtual methods are methods that the derived class must override.

How can you overload a method?
By changing the type or number of parameters.

What’s the difference between an interface and abstract class?
An abstract class is something that I don’t often use. Really; C# does not support multiple inheritance so we often use interfaces to define our entities and hierarchies. In c# we can inherit from one abstract class and implement multiple interfaces.

What is the default accessibility for a class?
Private

Can you implement multiple interfaces?
Yes

What is a delegate?
The C# version of AddressOf

Code a short and simple delegate example. Code the same sample with a lambda expression.

   1:  public delegate int myCalc(int x, int y);

   2:   

   3:  public class doMath {

   4:      public string Add(int x, int y){ Console.Wr...etc }

   5:      public string Sub(int x, int y){ Console.Wr...etc }

   6:  } 

   7:   

   8:  public static void Main(){

   9:      myCalc adder = new myCalc(doMath.Add);

  10:      adder(1,1);

  11:  } 

  12:   

  13:  public static void Main(){

  14:      myCalc adder = (doMath.Add);

  15:      adder(1,1);

  16:  } 

  17:   

  18:   

Strings are immutable, what does this mean?

The contents will not change once created, methods like toLower will not effect the string, only the output.

Code an enumeration for a status where options are: Queued, In Process, Finished

enum Status { Queued, In Process, Finished };

Can C# have multiple catch blocks?

Sure

What are design patterns?

Examples of how people smarter then me make my job easier.

Describe two design patterns you have used in a project.

Singleton! – Really; it’s was a good use…for a WCF Service to cache database results and serve those results to all web clients.

I’m not sure if MVC counts but It what I miss right now…MVVM is fine for large SL projects and the Flex stuff I am doing…well…as some of the Headspring guys on twitter might now, I’m not happy with it.

What are the most important aspects of an effective software development team?

Communication, planning, direction, understanding the problem domain (++DDD)

What is your experience with test-driven development? What are its limitations? What are its strengths?

I’m a fan and I use it where I’ve been able to convince a project manager that it fits into a project plan. When I have to take over a project without unit tests I feel lost. TDD, like design patterns, are something I love that makes my job easier.

What is your experience with Extreme Programming? What are its limitations? What are its strengths?

Not enough; the last 3 years I’ve been stuck in a big-design-up-front, not even waterfall, we spent a year on documents. 5 people, 1 year. sick. I was in a XP environment in 2001-2003.

Will this code compile? What will be its output?

Sample code was provided.

Write the PRECISE output of this code as it would appear in the console.

Sample code was provided

No related posts.

ASP.NET MVC 2 supports Spark!

What? hear me out. The release of ASP.NET MVC 2 includes an amazing hidden gem. You see the “Add View/Controller” dialog now takes into account a T4 directive “output extension”.

Getting started with Custom CodeTemplates

To get started with Custom CodeTemplates check out this excellent article by K. Scott Allen

Getting started with Custom CodeTemplates and Spark

1. At the top of your templates add the output extension as shown below.
2. Modify the template to use spark.

No related posts.

ASP.NET MVC 2 was RTM’ed today and with the help of Jay Harris – @jayharris we upgraded our rather large codebase to not only ASP.NET MVC 2 RTM but also a new drop of the Spark View Engine built against the RTM.

Since the release of Spark View Engine 1.1 RC a few weeks ago there have been some minor bugs/enhancements identified but no blockers. The update to RTM was nothing more then changing the assembly reference, build number and running the build script.

These changes are not yet reflected in on the GitHub source. I’m sure there will be a pull request or Lou – @loudej will update master pretty quickly.

Head over to CodePlex to download the new release and of course thanks to Lou for building an amazing tool.

Related posts:

1. Spark Binaries For ASP.NET MVC RC2 THIS IS NOT AN OFFICAL RELEASE…YET I spent the evening...
2. ASP.NET MVC Spark Support ASP.NET MVC 2 supports Spark! What? hear me out. The...
3. Spark with VB.NET Project Demo #2 Outline: Status of using spark in a VB.NET project Bugs...

try:
If you annotate your methods with the WebGet annotation, when a client adds your service as a service ref, there will be no client configuration information.

catch:
To put that another way, if you use the WebGet annotation it’s assumed the calling party is going to use rest.

finally:
if you use the WebGet annotation on your WCF operations they can be called as soap operations

No related posts.

THIS IS NOT AN OFFICAL RELEASE…YET

I spent the evening updating the spark source to support ASP.NET MVC RC/RC2. Well, actually I simply applied the changes that were posted to either the mailing list or codeplex.

All the tests pass:

Download it here:

Binaries: Spark-1.0.1.0-release.zip

Source: Spark-1.0.1.0-source.zip

This post, hopefully, won’t be up long.

No related posts.

Ok, that might be going a little too far but it sounds like the barrier I wrote about a few weeks ago might be going away.

Earlier this evening Phil Haack posted this tweet:

Excited to hear this I asked a leading implementation question about honoring the output extension directive in T4 templates.

Us old-timer remember the almost year before ASP.NET MVC 1.0, when we had not tooling support inside visual studio. Right before 1.0 was released we got right-click view/controller support and it made a world of difference. I honestly believe that without the GUI integration, that pushed us to use strong-typed views, MVC would have taken longer to get into less progressive shops.

I’m really excited about MVC2 …but it still needs a logo.

No related posts.