Blog Home  Home Feed your aggregator (RSS 2.0)  
Eric Malamisura's Blog
Geek Ramblings
 
 Thursday, April 24, 2008

So I have been working on a project where we are taking the existing data access code which is in pretty bad shape and we are trying to redesign and redo most of it using LINQ to SQL.  So after some research and reading from Scott Gu, LINQ in Action, and MSDN White Paper about how to use LINQ to SQL I noticed most of the examples they provide are overly immature.  They play like everything is great, and just drop it all in your code behind, or have a single layer of responsibility and it will all just work like magic!!  This sounded to good to be true, and after my experience it really is too good to be true!

 

So in this series of articles I am going to go through all of the things that we encountered.  The first thing you will run into with LINQ to SQL is that the DataContext is stateful and this is particularly concerning if you are planning to use it in a stateless environment like ASP.NET.  You can download the source code to the Northwind sample project that follows the design that is being discussed in this series here.

 

Below is a list of the issues we encountered while using LINQ to SQL and how we dealt with these issues.

  • DataContext Lifetime Management (Part 1)
  • Entity Persistence (Part 2)
  • Data Concurrency (Part 3)
  • N-Tier Design (Part 4)

DataContext Lifetime Management

So how do you manage the lifecycle of the DataContext?  There is a pretty good article (LINQ to SQL DataContext Lifetime Management) about this on Rick Strahl's blog and I won't cover what he already has in his article, I would recommend going to go read it and take from what you want.  I borrowed multiple concepts from his article and adapted them to my needs.

 

The solution we chose was keep the HttpContext around per a HTTP request, this was achieved by writing an HttpModule and creating an instance of this DataContext and storing in the HttpContext.Current.Items collection.  Now using this method you will have access to the same DataContext per a page request essentially, so you can do all of your work against a single DataContext and then submit it.

 

Below is the code that was used for storing the DataContext in a web application:

 

   10 public class WebContextStorage : IContextStorage

   11     {

   12         public WebContextStorage()

   13         {

   14 

   15         }

   16 

   17         #region IContextStorage Members

   18 

   19         public T GetItemContext<T>(string key)

   20         {

   21             Dictionary<string, object> storage = GetDictionaryFromStorage();

   22             return (storage == null) ? default(T) : (storage.ContainsKey(key) ? (T)storage[key] : default(T));

   23         }

   24 

   25         public void SetItemContext<T>(string key, T item)

   26         {

   27             Dictionary<string, object> storage = GetDictionaryFromStorage();

   28 

   29             if (storage == null)

   30             {

   31                 storage = new Dictionary<string, object>();

   32                 HttpContext.Current.Items.Add("storage", storage);

   33             }

   34 

   35             storage[key] = item;

   36         }

   37 

   38         private static Dictionary<string, object> GetDictionaryFromStorage()

   39         {

   40            return (Dictionary<string, object>)HttpContext.Current.Items["storage"];

   41         }

   42 

   43         #endregion

   44     }

 

There are other issues you need to deal with however, in our case we have Windows Services that need to be able to access our data to run various jobs, etc.  But no HttpContext exists in a WindowsService, so if you read Rick Strahl's blog you will already realize the solution to this is to use ThreadLocalStorage, we took a similar approach and used a Dictionary decorated with the [ThreadStatic] attribute, which was easier and we might be loosing some of the benefits of using ThreadLocalStorage but for now this is working just fine for us.

 

Below is the code that was used to store the DataContext in a non-web application:

 

    9 public class ThreadContextStorage : IContextStorage

   10     {

   11         [ThreadStatic]

   12         private static readonly Dictionary<string, object> _storage = new Dictionary<string, object>();

   13 

   14         public ThreadContextStorage()

   15         {

   16 

   17         }

   18 

   19         #region IContextStorage Members

   20 

   21         public T GetItemContext<T>(string key)

   22         {

   23             return (_storage.ContainsKey(key) ? (T)_storage[key] : default(T));

   24         }

   25 

   26         public void SetItemContext<T>(string key, T item)

   27         {

   28             _storage[key] = item;

   29         }

   30 

   31         #endregion

   32     }

 

So first off, we have two DataContext Lifetime Management strategies above, so we built a DataContextManager that follows the Strategy Design Pattern.  Now I don't like to get hung up on strict design pattern techniques, I always take the concept and adapt it to what works best for my situation, but I do want to point it out as a concept we used.  So we ended up writing something called the DataContextManager, the name is not really important but what it does is so below is the code:

 

   13  public class DataContextManager

   14     {

   15         private readonly IContextStorage _contextStorage;

   16 

   17         public DataContextManager(IContextStorage contextStorage)

   18         {

   19             if (contextStorage == null)

   20             {

   21                 throw new ArgumentNullException("contextStorage");

   22             }

   23 

   24             _contextStorage = contextStorage;

   25         }

   26 

   27         public T GetDataContext<T>() where T : DataContext, new()

   28         {

   29             T dataContext = _contextStorage.GetItemContext<T>(typeof(T).FullName);

   30             if (dataContext == null)

   31             {

   32                 dataContext = new T();

   33                 dataContext.Log = new DebuggerWriter();

   34                 _contextStorage.SetItemContext<T>(typeof(T).FullName, dataContext);

   35             }

   36 

   37             return dataContext;

   38         }

   39 

   40         public void Submit<T>() where T : DataContext

   41         {

   42             T dataContext = _contextStorage.GetItemContext<T>(typeof(T).FullName);

   43 

   44             if (dataContext != null)

   45             {

   46                 dataContext.SubmitChanges();

   47             }

   48         }

   49     }

 

So as you can see it is very simple, you could also add a new GetDataContextManager to pass in a connection string if needed but in our case we figured no one should need this ability since we have all of our connection strings in the configuration file we can just configured those using the LINQ to SQL designer.  If this becomes an issue we may add this ability.

 

The DataContextManager instance is stored in a ServiceLookup class we have that follows the popular design pattern by Martin Fowler called the Registry.  Essentially all it does is stores instances of all the service objects you will be using throughout your application, in our case this is built in a module on the Init so it is built once per AppDomain creation and remains static throughout.  If you remember from the lifecycle process of ASP.NET, first time a user visits the application it loads the AppDomain, when the HttpApplication is created it loads the HttpModules, and HttpHandlers, you can read about this in great detail in one of Rick Strahl's articles.

 

Below is where we define our implementation of the ServiceLookup class:

 

    8 public static class ServiceLocator

    9     {

   10         private static readonly Dictionary<string, object> _map = new Dictionary<string, object>();

   11 

   12         public static void Load<T>(T service)

   13         {

   14             Load(typeof(T).FullName, service);

   15         }

   16 

   17         public static void Load<T>(string key, T service)

   18         {

   19             _map.Add(key, service);

   20         }

   21 

   22         public static T Get<T>()

   23         {

   24             return Get<T>(typeof(T).FullName);

   25         }

   26 

   27         public static T Get<T>(string key)

   28         {

   29             return (T)_map[key];

   30         }

   31     }

 

As discussed above we want all of the objects in our ServiceLocator to be static, and maintain their lifetime throughout the lifetime of the AppDomain, this way we do not have reconstruct them multiple times.  To do this we use a HttpModule as seen below:

 

   16 public class ServiceBuilderModule : IHttpModule

   17     {

   18         public void Init(HttpApplication context)

   19         {

   20             ServiceBuilder.BuildServiceDependencies();

   21         }

   22 

   23         public void Dispose() { }

   24     }

 

The ServiceBuilder is a helper static class that simply does all the work so we can keep all of the code out of the module itself:

 

   12 public static class ServiceBuilder

   13     {

   14         public static void BuildServiceDependencies()

   15         {

   16             DataContextManager dataContextManager = new DataContextManager(new WebContextStorage());

   17             ServiceLocator.Load<DataContextManager>(dataContextManager);

   18             ServiceLocator.Load<CustomerDao>(new CustomerDao(dataContextManager));

   19         }

   20     }

 

We will cover in a later topic how the CustomerDao is used, etc. But for now the DataContextManager is what to really pay attention to since it constructs/retrieves your DataContext.  Next in order to get the DataContext all you have to do is request it from the ServiceLocator for example:

   24 public AdventureWorksLiteDataContext DataContext

   25         {

   26             get

   27             {

   28                 return _dataContextManager.GetDataContext<AdventureWorksLiteDataContext>();

   29             }

   30         }

 

So that is all for Part 1 in this series, next we will cover EntityPersistence and how we solved the problem of maintaining the state of a LINQ Entity across page posts.  If you have any questions do not hesitate to email me or preferably to comment below!

Thursday, April 24, 2008 6:04:08 PM (GMT Standard Time, UTC+00:00)  #    Comments [2]    |   | 
 Wednesday, March 12, 2008

This is nothing new but I created a bookmark for this a while back and was going through my .NET bookmarks (200+ of them) and came across this one.  This is actually pretty cool and I forgot all about it.  It lets you query a live database using LINQ, the implication of course that you are better at writing code than writing SQL queries, and most developers are much better at writing code than SQL queries.  Plus not all SQL is simple, so this will let you write those complex queries in a manner of seconds.

http://www.linqpad.net/

Oh did I forget to mention it has a killer interface?

Wednesday, March 12, 2008 2:20:48 PM (GMT Standard Time, UTC+00:00)  #    Comments [0]    |   | 
 Friday, February 15, 2008

Windows Vista SP1 is on Vista right now, albeit through a special link since it doesn't show up through the normal login page just yet.  So go here http://msdn2.microsoft.com/en-us/subscriptions/default.aspx

Click it, login using your MSDN Subscription and money money money!

Friday, February 15, 2008 12:36:36 AM (GMT Standard Time, UTC+00:00)  #    Comments [0]    | 
 Thursday, February 14, 2008

So apparently Vista SP1 Final will be available to all MSDN Subscribers by the end of this week based on a blog post on the MSDN blog.  Also you can view ALL of the changes that ended up going into Vista SP1 here.  I installed the Beta and was very impressed with the speed improvements, and even though most people claim stability problems I really didn't have any stability issues left with Vista before installing SP1.  Most of my stability issues have vanished with the latest driver updates for my hardware.  However, with SP1 there were some substantial performance increases that I noticed immediately after installing.

Thursday, February 14, 2008 1:39:42 AM (GMT Standard Time, UTC+00:00)  #    Comments [0]    | 
 Monday, February 04, 2008

So ever since I have started working with Visual Studio 2008 I find myself hitting shortcut keys that just don't exist anymore.  I have been denied my ReSharper abilities and it is really humiliating to work without them, I miss my ReSharper.  But soon, very soon in fact by February 15th the good folks at JetBrains will be opening up the Early Access Program for ReSharper 4 and not soon enough! 

Thank god! That is all I have to say, working with ReSharper substantially improves my code quality and code output. You can read all about this on ReSharper Horizons blog.

Monday, February 04, 2008 4:38:30 PM (GMT Standard Time, UTC+00:00)  #    Comments [0]    | 
 Sunday, February 03, 2008

So I have been exploring DI Frameworks lately (Dependency Injection), and there are a couple of solutions.  While I kind of like the metadata approach vs. the XML Configuration one there really isn't anything available yet.  You have ObjectBuilder, but its only a DI Framework, it doesn't include a DI Container so you are left to build your own container, which includes setting up all of your strategies, context, etc.  The CodePlex guys are apparently working on an ObjectBuilder v2 which they are trying to get vetted by the P&P team at Microsoft.

So the choices are:
Spring.NET
StructureMap
Castle Windsor Container

I have been evaluating these mostly because of a hobby project I am working on called VistaTune, I would like to use a DI Framework and the current changeset that is checked in is using Spring.NET but I am not happy with Spring.NET for a couple of reasons.  First there is no real way to control your dependency injection dynamically, its all configuration based.  Secondly there is no way to inject an instance into your container, for example during unit testing I would like to inject a mock to override the real instance that is mapped in the configuration.  Thirdly there is no DSL (Domain Specific Language), and appears to be no DSL planned for the future release.

Note: After posting this Jeremy D. Miller the developer of StructureMap updated me, the current version of StructureMap v2.0 already has DSL but the next version is being extended with even more.  See the comments section below, also visit his blog here for more information.

For those reasons, I am leaning towards using StuctureMap, which in current release has no DSL but next release will which is already in testing so its pretty much done.  It supports container injection, and it allows you to reset your injection so you can revert back to the original container before injection which is really cool.  It also supports interception, among other things that are out of the scope of my assessment.

So which DI Framework?  It is a difficult question to answer, and I really think it will be interesting to see what comes about from the current top runners at the moment.  I expect to see big things from Microsoft on ObjectBuilder, especially since they just released their PIAB (Policy Injection Application Block) which is their solution to an AOP (Aspect Oriented Programming) Framework.  I would personally rather see PIAB stay isolated by itself, I prefer small isolated Frameworks rather than huge bloated ones that you only want to use for a single thing. 

Sunday, February 03, 2008 8:14:29 AM (GMT Standard Time, UTC+00:00)  #    Comments [1]    |   | 
 Sunday, January 20, 2008

Ok, so the previous post I discussed a service locator implementation based on the Martin Fowler registry example.  After some thought and discussion with a couple of friends I decided the previous implementation was a bit too strict for it's own good.  So I have come with a way that still gives you generic goodness, it does however suffer from boxing/unboxing which may give you a slight performance penalty vs. my previous implementation but the penalty should be minimal if noticeable at all.

The new service locator implementation:

  public interface IService { }
    public class ServiceLocator
    {
        private static ServiceLocator _instance;
        private ServiceLocator() { }

        private static Dictionary<string, object> _map = new Dictionary<string, object>();

        public static ServiceLocator Instance
        {
            get
            {
                if (_instance == null)
                {
                    _instance = new ServiceLocator();
                }
                return _instance;
            }
        }

        public void Load<T>(T service) where T: IService
        {
            Load(typeof(T).FullName, service);
        }

        public void Load<T>(string key, T service) where T : IService
        {
            _map.Add(key, service);
        }

        public T Get<T>() where T : IService
        {
            return Get<T>(typeof(T).FullName);
        }

        public T Get<T>(string key) where T : IService
        {
            return (T)_map[key];
        }
    }

To use this the syntax is slightly different than before, this time everything is method based.  I still constrain the methods to the IService type, for some reason  it just gives me a warm and fuzzy feeling when I use the constraint. It is obviously up to you whether or not you want to use it, you could delete the marker interface and the constraints completely and it will work just fine.

To Get: MyService wmiService = ServiceLocator.Instance.Get<MyService>();
To Load: ServiceLocator.Instance.Load<WMIService>(new WMIService());

So both versions work, both versions are acceptable solutions but this one is more flexible than the last version.  In particular this version allows a greater degree of flexibility around unit testing and mocking frameworks.

Sunday, January 20, 2008 7:12:03 AM (GMT Standard Time, UTC+00:00)  #    Comments [0]    | 
 Wednesday, January 16, 2008

So I came across the need for a service locator but I didn't want to use a framework since my needs were relatively lightweight, and mostly based around testing.  So looking at Martin Fowlers examples of Service Locators I adopted them and came up with a .NET way of doing it using Generics.  It is of course flexible, the method I am offering obviously has a couple of constraints that I have put in place on purpose for my scenario.  One of them being I wanted only one instance of a service set in the service locator at any given time, so you could very easily adapt it to suit your own needs. 

Also the interface specified is a marker interface used to specify a constraint on the generic, the idea was that all services would be marked with this interface.

 public interface IService { }

    public class ServiceLocator<T> where T : IService
    {
        private static Dictionary<string, T> _map = new Dictionary<string, T>();
        private ServiceLocator() { }

        public static void Load(T service)
        {
           _map.Add(typeof(T).FullName, service);
        }

        public static T Get()
        {
            return _map[typeof(T).FullName];
        }
    }

An example of a service would be like follows:

public class MyService : IService
{
    //Whatever Methods
}
Then on application startup you would obviously load all of your services into the service locator:
MyService myService = new MyService(...);
//...do whatever you need to setup the service

ServiceLocator<MyService>.Load(myService);

So later on anywhere in your code you simply do a

MyService myService = ServiceLocator<MyService>.Get();

It's pretty simple but just in case...As you can see I chose to use the FullName of the type as the key in map, you could always choose to do it the way Martin Fowler did where you simply expose the instance of each service type directly but for me I kind of liked this method.

 

Wednesday, January 16, 2008 5:15:04 AM (GMT Standard Time, UTC+00:00)  #    Comments [0]    | 
Copyright © 2008 Eric Malamisura. All rights reserved.
DasBlog 'Portal' theme by Johnny Hughes.