I am wondering if there is a best practice for creating a REST API with ASP.NET MVC 3? At the moment I am thinking of creating a controller for each version of the REST API. For example, so far I have:
public class V1Controller : Controller
{
public V1Controller()
{
}
public ActionResult GetUser(string userId, IUserRepository userRepostory)
{
//code to pull data and convert to JSON string
return View("Results");
}
public ActionResult GetUsersByGroup(string groupId, IUserRepository userRepostory)
{
//code to pull data and convert to JSON string
return View("Results");
}
}
Then for the views I overwrite the _ViewStart.cshtml to remove the layout and then I have Results.cshtml that just outputs the data that is formatted in the controller action, right now JSON. Having every single REST call in one controller seems like a bit too much but it is the best way I can think of so that I can keep clean separate versions of the API so that when it comes to creating version 2 of the API, I can create a V2Controller and not break the existing API to give people time to switch over to the new API.
Is there a better way to create a REST API with ASP.NET MVC 3?
I was able to find a decent solution using MVC's use of Areas.
First, I wanted to have my API follow this URL Definition:
http://[website]/[major_version]_[minor_version]/{controller}/{action}/...
I also wanted to break up the different versions in separate Project files and use the same Controller names in each version:
"../v1_0/Orders/ViewOrders/.." => "../v2_3/Orders/ViewOrders/.."
I searched around and found a workable solution with the use of MVC Areas.
I created a new project in my solution called "Api.Controllers.v1_0" and, as a test, put a SystemController.cs file in there:
using System.Web.Mvc;
namespace Api.Controllers.v1_0
{
public class SystemController : Controller
{
public ActionResult Index()
{
return new ContentResult() {Content = "VERSION 1.0"};
}
}
}
I then added a v1_0AreaRegistration.cs file:
using System.Web.Mvc;
namespace Api.Controllers.v1_0
{
public class v1_0AreaRegistration : AreaRegistration
{
public override string AreaName
{
get{ return "v1_0";}
}
public override void RegisterArea(AreaRegistrationContext context)
{
context.MapRoute(
"v1_0",
"v1_0/{controller}/{action}/{id}",
new { controller = "System", action = "Index", id = UrlParameter.Optional }
);
}
}
}
I walked through the same steps above for a "..v1_1" project with the corresponding files in there, added the projects as references into my "Api.Web" MVC project and was off and running.
If all you are returning is JSON, you do not need a view. Jusr return
new JsonResult(){Data = Data};
Look in here.
Also in terms of versioning, versions can be implemented as different controllers or as extra methods in the same controller. But without knowing why you would need versions and why your clients (which I assume are browsers) would need to know about versioning is not clear from your question.
A controller such as the one you posted in your example code should always keep that methods that you have now for instance GetUsersByGroup() with the same signature. I don't see how there could be a different version of that method.
The inputs are group and repository (which I believe comes from DI). The output is a list of users in JSON format. That's all that matters to the users of the API. What you do inside this method is no one's business.
You should think more of inputs and outputs. You shouldn't be changing the signatures of existing actions unless it is really neccessary to do so.
Think of the controller class in terms of implementing the interface. You have an interface and controller class is it's implementation (I mean you don't need to have it but just think of it in that way). You will rarely change the interface once one or several classes implement it. But you might add the methods to it. And that requires only changes in implementing classes - it does not break the functionality of the API and everyone who's using it will be able to continue using it.
Related
I have a two different projects that are actually two different websites when deployed, say WebA & WebB. The structure of both websites is the same in terms of the controllers and models they use. However, the front end is very different. They each have their own UI and own jquery plugins that work for them.
To reduce code duplication, I am proposing both those projects inherit from controllers in another referenced project[controllerDLL]. ControllerDLL is a project that will have all the logic for calling the business layer and returning json serialized model objects. WebA & WebB will each just have empty controller classes that just inherit from the base project[controllerDLL]. This way I feel the controller code is not duplicated.
Can anyone tell me a better way to achieve controller reuse other than the way I have proposed? Does asp.net mvc 4 provide any built-in way to do this better? Can I use AREAS here productively? Is DotNetNuke something I should look at? Or is my approach the best way forward? I am not looking for the how to move controllers into another project. I just want to know what my options are for achieving controller reuse.
Thanks.
Take a look at this SO question. You could keep all the views in the same project and use a custom view engine to find the views (based off a web.config setting).
For example you could have 2 folders /Views/WebA and /Views/WebB. The custom view engine could look up the web.config setting to find out which folder to use and search for the views there. This way you would not have to duplicate the controller code or move it into a separate project. It will just swap out one presentation layer for another at runtime.
You should be able to achieve this by implementing custom controller factory to instantiate the right controller class based on config settings.
You need to implement the interface System.Web.Mvc.IControllerFactory.The two methods in this interface are:
1.System.Web.Mvc.IController CreateController(System.Web.Routing.RequestContext requestContext, string controllerName) - To create the controller instance based on RequestContext and controller name.
2.void ReleaseController(System.Web.Mvc.IController controller) - Release the controller instance
Detailed information about using that interface is available
Since the difference between the two applications are the views, you don't need to have them as separate projects. you could have two sets of views and deploy the same project twice using different settings in the web config. To do this, you'll have to implement your own ViewEngine so that the controller can return the correct right views.
public class AppSettingViewEngine: RazorViewEngine
{
public AppSettingViewEngine()
{
string[] viewLocations = new string[] { };
if (ConfigurationManager.AppSettings["Site"] == "WebA")
{
viewLocations = new[] {
"~/WebB/Views/{1}/{0}.cshtml",
"~/WebB/Views/{1}/{0}.cshtml",
"~/WebB/Views/Shared/{0}.cshtml",
"~/WebB/Views/Shared/{0}.cshtml",
};
}
if (ConfigurationManager.AppSettings["Site"] == "WebB")
{
viewLocations = new[] {
"~/WebB/Views/{1}/{0}.cshtml",
"~/WebB/Views/{1}/{0}.cshtml",
"~/WebB/Views/Shared/{0}.cshtml",
"~/WebB/Views/Shared/{0}.cshtml",
};
}
else
{
//Default Settings
viewLocations = new[] {
"~/Views/{1}/{0}.cshtml",
"~/Views/{1}/{0}.cshtml",
"~/Views/Shared/{0}.cshtml",
"~/Views/Shared/{0}.cshtml",
};
}
this.PartialViewLocationFormats = viewLocations;
this.ViewLocationFormats = viewLocations;
}
}
Then you register it in your Application_Start as follows
protected void Application_Start()
{
ViewEngines.Engines.Clear();
ViewEngines.Engines.Add(new AppSettingViewEngine());
}
I would like to know what is the best practice with ServiceStack registering routes.
Define routes using Routes.Add in AppHost
Define routes using RouteAttiribute decorator on DTO's
I have read some where that doing it in AppHost is preferable because it de-couples route definition from DTO's. But most of the examples I see are using the decorator pattern.
So which is better? Are there any arguments to use one over the other?
Decorating your DTOs with route attributes has become the best practice. The release of SerivceStack's "new API" has indicated the many benefits of this but most importantly it promotes a more "succinct, typed, end-to-end client API".
By using the Route attributes in addition to marking your DTOs with one of the IReturn, IReturn<T> or IReturnVoid interfaces you can use the full features now offered by ServiceStack.
Advantages
You can dynamically discover urls to consume endpoints based on the the DTOs alone.
Your consuming service client code (if c#) becomes more strongly typed and it becomes trivial to change existing routes/DTOs
Keeps the AppHost clean and slim. I’m a believe that the less code in the AppHost the better.
code can become more testable
works alongside adding routes in AppHost (you don't have to choose just one)
and best of all it all just works!
For the given DTO you can use the ToUrl extension method to get the generate the route by discovery the attributes.
[Route("/route/{Id}/{Name}")]
public class ExampleDTO : IReturn
{
public int Id { get; set; }
public string Name { get; set; }
public string Value { get; set; }
}
var url = new ExampleDTO() { Id = 1, Name = "Test", Value = "Foo" }.ToUrl("GET");
// generates /route/1/Test?value=Foo
The ServiceClients will do all this under the hood. Your consuming code can become:
ExampleDTO response = new JsonServiceClient(“http://api.com/)
.Get(new ExampleDTO() { Id = 1, Name = "Test", Value = "Foo" });
What about decoupling?
I would argue that there is no big need to decouple the DTOs from the Routes. I want my fellow developers to know that the Request should not be changed without changing the route. I have seen a lot of runtime errors because I forget to update the route in the AppHost after changing a DTO.
I have a web solution (in VS2010) with two sub-projects:
Domain which holds the Model classes (mapped to database tables via Entity Framework) and Services which (besides other stuff) are responsible for CRUD operations
WebUI which references the Domain project
For the first pages I've created I have used the Model classes from the Domain project directly as Model in my strongly typed Views because the classes were small and I wanted to display and modify all properties.
Now I have a page which should only work with a small part of all properties of the corresponding Domain Model. I retrieve those properties by using a projection of the query result in my Service class. But I need to project into a type - and here come my questions about the solutions I can think of:
I introduce ViewModels which live in the WebUI project and expose IQueryables and the EF data context from the service to the WebUI project. Then I could directly project into those ViewModels.
If I don't want to expose IQueryables and the EF data context I put the ViewModel classes in the Domain project, then I can return the ViewModels directly as result of the queries and projections from the Service classes.
In addition to the ViewModels in the WebUI project I introduce Data transfer objects which move the data from the queries in the Service classes to the ViewModels.
Solution 1 and 2 look like the same amount of work and I am inclined to prefer solution 2 to keep all the database concerns in a separate project. But somehow it sounds wrong to have View-Models in the Domain project.
Solution 3 sounds like a lot more work since I have more classes to create and to care about the Model-DTO-ViewModel mapping. I also don't understand what would be the difference between the DTOs and the ViewModels. Aren't the ViewModels exactly the collection of the selected properties of my Model class which I want to display? Wouldn't they contain the same members as the DTOs? Why would I want to differentiate between ViewModels and DTO?
Which of these three solutions is preferable and what are the benefits and downsides? Are there other options?
Thank you for feedback in advance!
Edit (because I had perhaps a too long wall of text and have been asked for code)
Example: I have a Customer Entity ...
public class Customer
{
public int ID { get; set; }
public string Name { get; set; }
public City { get; set; }
// ... and many more properties
}
... and want to create a View which only shows (and perhaps allows to edit) the Name of customers in a list. In a Service class I extract the data I need for the View via a projection:
public class CustomerService
{
public List<SomeClass1> GetCustomerNameList()
{
using (var dbContext = new MyDbContext())
{
return dbContext.Customers
.Select(c => new SomeClass1
{
ID = c.ID,
Name = c.Name
})
.ToList();
}
}
}
Then there is a CustomerController with an action method. How should this look like?
Either this way (a) ...
public ActionResult Index()
{
List<SomeClass1> list = _service.GetCustomerNameList();
return View(list);
}
... or better this way (b):
public ActionResult Index()
{
List<SomeClass1> list = _service.GetCustomerNameList();
List<SomeClass2> newList = CreateNewList(list);
return View(newList);
}
With respect to option 3 above I'd say: SomeClass1 (lives in Domain project) is a DTO and SomeClass2 (lives in WebUI project) is a ViewModel.
I am wondering if it ever makes sense to distinguish the two classes. Why wouldn't I always choose option (a) for the controller action (because it's easier)? Are there reasons to introduce the ViewModel (SomeClass2) in addition to the DTO (SomeClass1)?
I would solve your problem by using an auto-mapping tool (like AutoMapper) to do the mapping for you. In cases where the mapping is easy (for example if all properties from one class should be mapped to properties with the same name on another class) AutoMapper will be able to do all the hook-up work for you, and you'll have to give a couple of lines of code to note that there should be a map between the two at all.
That way, you can have your entities in Domain, and a couple of view model classes in your WebUI, and somewhere (preferrably in WebUI or a sub namespace of the same) define maps between them. Your view models will in effect be DTOs, but you won't have to worry much about the conversion process between the domain and your DTO classes.
Note: I would strongly recommend against giving your Domain entities straight to the views of your MVC web UI. You don't want EF to "stick around" all the way to the front-end layer, in case you later want to use something other than EF.
introduce ViewModels which live in the
WebUI project and expose IQueryables
and the EF data context from the
service to the WebUI project. Then I
could directly project into those
ViewModels.
The trouble with this is you soon run into problems using EF trying to 'flatten' models. I encountered something similar when I had a CommentViewModel class that looked like this:
public class CommentViewModel
{
public string Content { get; set; }
public string DateCreated { get; set; }
}
The following EF4 query projection to the CommentViewModel wouldn't work as the couldn't translate the ToString() method into SQL:
var comments = from c in DbSet where c.PostId == postId
select new CommentViewModel()
{
Content = c.Content,
DateCreated = c.DateCreated.ToShortTimeString()
};
Using something like Automapper is a good choice, especially if you have a lot of conversions to make. However, you can also create your own converters that basically convert your domain model to your view model. In my case I created my own extension methods to convert my Comment domain model to my CommentViewModel like this:
public static class ViewModelConverters
{
public static CommentViewModel ToCommentViewModel(this Comment comment)
{
return new CommentViewModel()
{
Content = comment.Content,
DateCreated = comment.DateCreated.ToShortDateString()
};
}
public static IEnumerable<CommentViewModel> ToCommentViewModelList(this IEnumerable<Comment> comments)
{
List<CommentViewModel> commentModels = new List<CommentViewModel>(comments.Count());
foreach (var c in comments)
{
commentModels.Add(c.ToCommentViewModel());
}
return commentModels;
}
}
Basically what I do is perform a standard EF query to bring back a domain model and then use the extension methods to convert the results to a view model. For example, the following methods illustrate the usage:
public Comment GetComment(int commentId)
{
return CommentRepository.GetById(commentId);
}
public CommentViewModel GetCommentViewModel(int commentId)
{
return CommentRepository.GetById(commentId).ToCommentViewModel();
}
public IEnumerable<Comment> GetCommentsForPost(int postId)
{
return CommentRepository.GetCommentsForPost(postId);
}
public IEnumerable<CommentViewModel> GetCommentViewModelsForPost(int postId)
{
return CommentRepository.GetCommentsForPost(postId).ToCommentViewModelList();
}
Talking about Models, ViewModels and DTOs is confusing, personally I don't like to use these terms. I prefer to talk about Domain Entities, Domain Services, Operation Input/Result (aka DTOs). All of these types live in the Domain layer. Operations is the behavior of Entities and Services. Unless you are building a pure CRUD application the presentation layer only deals with Input/Result types, not Entities. You don't need additional ViewModel types, these are the ViewModels (in other words, the Model of the View). The View is there to translate the Operation Results to HTML, but the same Result could be serialized as XML or JSON. What you use as ViewModel is part of the domain, not the presentation layer.
I've been playing around with ASP.NET MVC for the past few weeks. I've got a simple web application with a form which contains a number of drop down lists.
The items in the drop down lists are stored in a database, and I'm using LINQ to SQL to retrieve them.
My question is - where's the appropriate place to put this code? From what I've read so far, it seems that it's advisible to keep the Controller 'thin', but that's where I currently have this code as it needs to be executed when the page loads.
Where should I be putting DB access code etc.? I've included an excerpt from my controller below.
Thanks.
public ActionResult Index()
{
TranslationRequestModel trm = new TranslationRequestModel();
// Get the list of supported languages from the DB
var db = new TransDBDataContext();
IEnumerable<SelectListItem> languages = db.trans_SupportedLanguages
.Select(c => new SelectListItem
{
Value = Convert.ToString(c.ID),
Text = c.Name.ToString()
});
ViewData["SourceLanguages"] = languages;
ViewData["TargetLanguages"] = languages;
return View();
Your database access code should be in a repository. Example:
public interface ITranslationRepository
{
Translation GetTransaltion();
}
and the controller would use this repository:
public class TransaltionController : Controller
{
private readonly ITranslationRepository _repository;
public TransaltionController(ITranslationRepository repository)
{
_repository = repository;
}
public ActionResult Index()
{
// query the repository to fetch a model
Translation translation = _repository.GetTransaltion();
// use AutoMapper to map between the model and the view model
TranslationViewModel viewModel = Mapper.Map<Translation, TranslationViewModel>(model);
// pass the view model to the view
return View(viewModel);
}
}
So the basic idea is the following:
The controller queries a repository to fetch a model
The controller maps this model to a view model (AutoMapper is great for this job)
The controller passes the view model to the view
The view is strongly typed to the view model and uses it to edit/display
As far as the implementation of this repository is concerned feel free to use any data access technology you like (EF, NHibernate, Linq to XML, WCF calls to remote resources over the internet, ...)
There are the following advantages:
The controller logic is completely decoupled from the data access logic
Your controllers can be unit tested in isolation
Your models are not littered with properties that should belong to the UI layer (such as SelectListItem) and thus are reusable across other types of application than ASP.NET MVC.
The view model is a class which is specifically tailored to the needs of the view meaning that it will contain specific formatted properties and the view code will be extremely readable.
Your views are strongly typed => no more ViewData and ugly magic strings
Suggest that your data-access code should be contained in its own project/assembly. That is referenced by the UI tier (ASP.NET MVC application). That'll help achieve the goal of keeping your Controllers thin, and keep all the data access code out of your MVC UI project.
That typically leads to another question/discussion about domain entities: when mapping to the data store. Some architects like to have the entities in their own separate assembly. This encourages reuse in other applications. Some like to keep the entity model and data access code in the same project/assembly. This is totally up to you and your environment.
For an example, let's say it's a billing application; holding customers, invoices, etc.
Your implementation will be different, depending on your data access strategy (an ORM like LINQ To SQL, EF, nHibernate, SubSonic, or plain old ADO.NET, or reading from a flat file).
// Assembly: InvoicingDL
public class CustomerRepo
{
public IQueryable<Customer> ListCustomers()
{
return MyDatabase.Customers(); //however you'd get all your customers
}
//etc
}
// Assembly: InvoicingDL
public class InvoicingRepo
{
public IQueryable<Invoice> GetCustomerInvoices(int custID)
{
return MyDatabase.Invoices.Where(i=>i.CustomerID==custID);
}
//etc
}
Check out the Repository pattern
https://web.archive.org/web/20110503184234/http://blogs.hibernatingrhinos.com/nhibernate/archive/2008/10/08/the-repository-pattern.aspx
http://www.mindscapehq.com/blog/index.php/2008/05/12/using-the-unit-of-work-per-request-pattern-in-aspnet-mvc/
The idea is you abstract your data access in something called a repository that returns domain objects. Your controller can then use this repository to get the appropriate objects from the database and assign them to the model.
I've been getting several errors:
cannot add an entity with a key that is already in use
An attempt has been made to attach or add an entity that is not new, perhaps having been loaded from another datacontext
In case 1, this stems from trying to set the key for an entity versus the entity. In case 2, I'm not attaching an entity but I am doing this:
MyParent.Child = EntityFromOtherDataContext;
I've been using using the pattern of wrap everything with a using datacontext. In my case, I am using this in a web forms scenario, and obviously moving the datacontext object to a class wide member variables solves this.
My questions are thus 2 fold:
How can I get rid of these errors and not have to structure my program in an odd way or pass the datacontext around while keeping the local-wrap pattern? I assume I could make another hit to the database but that seems very inefficient.
Would most people recommend that moving the datacontext to the class wide scope is desirable for web pages?
Linq to SQL is not adapted to disconnected scenarios. You can copy your entity to a DTO having a similar structure as the entity and then pass it around. Then copy the properties back to an entity when it's time to attach it to a new data context. You can also deserialize/reserialize the entity before attaching to a new data context to have a clean state. The first workaround clearly violates the DRY principle whereas the second is just ugly. If you don't want to use any of these solution the only option left is to retrieve the entity you're about to modify by its PK by hitting the DB. That means an extra query before every update. Or use another ORM if that's an option for you. Entity Framework 4 (included with .NET 4) with self-tracking entities is what I'm using currently on a web forms project and everything is great so far.
DataContext is not thread-safe and should only be used with using at the method level, as you already do. You can consider adding a lock to a static data context but that means no concurrent access to the database. Plus you'll get entities accumulated in memory inside the context that will turn into potential problems.
For those that came after me, I'll provide my own take:
The error "an attempt has been made to add or attach an entity that is not new" stems from this operation:
Child.Parent = ParentEntityFromOtherDataContext
We can reload the object using the current datacontext to avoid the problem in this way:
Child.Parent = dc.Entries.Select(t => t).Where(t => t.ID == parentEntry.ID).SingleOrDefault();
Or one could do this
MySubroutine(DataContext previousDataContext)
{
work...
}
Or in a web forms scenario, I am leaning to making the DataContext a class member such as this:
DataContext _dc = new DataContext();
Yes, the datacontext is suppose to represent a unit of work. But, it is a light-weight object and in a web forms scenario where a page is fairly transient, the pattern can be changed from the (using dc = new dc()) to simply using the member variable _dc. I am leaning to this last solution because it will hit the database less and require less code.
But, are there gotchas to even this solution? I'm thinking along the lines of some stale data being cached.
What I usually do is this
public abstract class BaseRepository : IDisposable
{
public BaseRepository():
this(new MyDataContext( ConfigurationManager.ConnectionStrings["myConnection"].ConnectionString))
{
}
public BaseRepository(MyDataContext dataContext)
{
this.DataContext = dataContext;
}
public MyDataContext DataContext {get; set;}
public void Dispose()
{
this.DataContext.Dispose();
}
}
Then imagine I have the following repository
public class EmployeeRepository : BaseRepository
{
public EmployeeRepository():base()
{
}
public EmployeeRepository(MyDataContext dataContext):base(dataContext)
{
}
public Employee SelectById(Guid id)
{
return this.DataContext.Employees.FirstOrDefault(e=>e.Id==id);
}
public void Update(Employee employee)
{
Employee original = this.Select(employee.Id);
if(original!=null)
{
original.Name = employee.Name;
//others
this.DataContext.SubmitChanges();
}
}
}
And in my controllers (I am using asp.net mvc)
public ActionResult Update(Employee employee)
{
using(EmployeeRepository employeeRepository = new EmployeeRepository())
{
if(ModelState.IsValid)
{
employeeRepository.Update(employee);
}
}
//other treatment
}
So the datacontext is properly disposed and I can use it across the same instance of my employee repository
Now imagine that for a specific action I want the employee's company to be loaded (in order to be displyed in my view later), I can do this:
public ActionResult Select(Guid id)
{
using(EmployeeRepository employeeRepository = new EmployeeRepository())
{
//Specifying special load options for this specific action:
DataLoadOptions options = new DataLaodOptions();
options.LoadWith<Employee>(e=>e.Company);
employeeRepository.DataContext.LoadOptions = options;
return View(employeeRepository.SelectById(id));
}
}