We are trying to figure out how to setup Dependency Injection for situations where service classes can have different dependencies based on how they are used. In our specific case, we have a web app where 95% of the time the connection string is the same for the entire Request (this is a web application), but sometimes it can change.
For example, we might have 2 classes with the following dependencies (simplified version - service actually has 4 dependencies):
public LoginService (IUserRepository userRep)
{
}
public UserRepository (IContext dbContext)
{
}
In our IoC container, most of our dependencies are auto-wired except the Context for which I have something like this (not actual code, it's from memory ... this is StructureMap):
x.ForRequestedType().Use()
.WithCtorArg("connectionString").EqualTo(Session["ConnString"]);
For 95% of our web application, this works perfectly. However, we have some admin-type functions that must operate across thousands of databases (one per client). Basically, we'd want to do this:
public CreateUserList(IList<string> connStrings)
{
foreach (connString in connStrings)
{
//first create dependency graph using new connection string
????
//then call service method on new database
_loginService.GetReportDataForAllUsers();
}
}
My question is: How do we create that new dependency graph for each time through the loop, while maintaining something that can easily be tested?
To defer the creation of an object until runtime, you can use a factory:
public interface ILoginServiceFactory
{
ILoginService CreateLoginService(string connectionString);
}
Usage:
public void CreateUserList(IList<string> connStrings)
{
foreach(connString in connStrings)
{
var loginService = _loginServiceFactory.CreateLoginService(connString);
loginService.GetReportDataForAllUsers();
}
}
Within the loop, do:
container.With("connectionString").EqualTo(connString).GetInstance<ILoginService>()
where "connectionString" is the name of a string constructor parameter on the concrete implementation of ILoginService.
So most UserRepository methods use a single connection string obtained from session, but several methods need to operate against a list of connection strings?
You can solve this problem by promoting the connection string dependency from IContext to the repository and adding two additional dependencies - a context factory and a list of all the possible connections strings the repository might need to do its work:
public UserRepository(IContextFactory contextFactory,
string defaultConnectionString,
List<string> allConnectionStrings)
Then each of its methods can build as many IContext instances as they need:
// In UserRepository
public CreateUserList() {
foreach (string connString in allConnectionStrings) {
IContext context = contextFactory.CreateInstance(connString);
// Build the rest of the dependency graph, etc.
_loginService.GetReportDataForAllUsers();
}
}
public LoginUser() {
IContext context = contextFactory.CreateInstance(defaultConnectionString);
// Build the rest of the dependency graph, etc.
}
We ended up just creating a concrete context and injecting that, then changing creating a wrapper class that changed the context's connection string. Seemed to work fine.
Related
I am setting up an asp.Net Mvc 4 app and looking to configure it using the Onion Architecture Pattern.
In the past I have used the Unit of Work Pattern like this
public class UnitOfWork : IUnitOfWork, IDisposable
{
private IRepository<CallModel> _callRepo;
private IRepository<UserModel> _userRepo;
public IRepository<CallModel> CallRepo
{
get
{
if (_callRepo == null)
{
_callRepo = new Repository<CallModel>();
}
return _callRepo;
}
}
public IRepository<UserModel> UserRepo
{
get
{
if (_userRepo == null)
{
_userRepo = new Repository<UserModel>();
}
return _userRepo;
}
}
}
I would then pass the instance of the UnitOfWork Class to the Controller to do simple CRUD stuff like this.
public class QuestionsController : Controller
{
private IUnitOfWork _unitOfWork;
[Inject]
public QuestionsController(IUnitOfWork unitOfWork)
{
_unitOfWork = unitOfWork;
}
I have seperated the app into three projects.
Core
Infrastructure
Web
I have my Interfaces all in the Core project and the implementation of the IRepository interface in the Infrastructure project.
If I put the UnitOfWork Class in the Core Project then since it calls for a new Repository in the Infrastructure project I am creating a dependency from the Core to the Infrastructure.
If I include it in the Infrastructure then the Web project (which has the controllers) will have a dependency on the Infrastructure and the whole Solution ends up looking less like an Onion and more like spaghetti.
I have my Interfaces all in the Core project and the implementation of the IRepository interface in the Infrastructure project. If I put the UnitOfWork Class in the Core Project then since it calls for a new Repository in the Infrastructure project I am creating a dependency from the Core to the Infrastructure.
Hmm, not really. Your unit of work class should have a dependency on IRepository, not the Repository implementation itself. If you are using Dependency Injection, this should not pose a problem, as it should find the right type and provide it at runtime. I'm not sure whether the Onion architecture is even possible without using DI.
See david.s's answer as well, as this is exactly how I set things up--have a project for the sole purpose of wiring up dependencies.
What I do is have another project named DependencyResolution which has references to Core and Infrastructure an where I configure my IoC container. Then I can refence only DependencyResolution from the Web project.
I would do like david.s create project named DependencyResolution but let it referance Web, Core and Infrastructure.
In that project you could do:
[assembly: PreApplicationStartMethod(typeof(Start), "Register")]
namespace DependencyResolution
{
public static class Start
{
public static void Register()
{
UnityConfig.Register();
}
}
}
and to register DI.
namespace DependencyResolution
{
public static class UnityConfig
{
public static void Register()
{
DependencyResolver.SetResolver(new UnityDependencyResolver());
}
}
}
So no referance between Web and infrastructure is needed.
Best regards
For what it's still worth, I have implemented my own library that applies the UnitOfWork-pattern a little differently than I've seen in any code sample before, but I have found it to work very well in practice. In short: I kinda copied the way .NET Transactions work by creating a scope and then enlisting resources in the ambient unitofwork(-manager) where necessary. What basically happens is that when a new message/request is being handled, this code is executed:
public void Handle<TMessage>(TMessage message)
{
using (var scope = CreateMessageProcessorContextScope())
{
HandleMessage(message);
scope.Complete();
}
}
Now just as with transactions, as soon as the Thread is still inside the scope, an ambient UnitOfWork-controller is present in which all resources that are used and changed during the request can enlist dynamically. They do this by implementing the IUnitOfWork-interface that has two methods:
public interface IUnitOfWork
{
bool RequiresFlush();
void Flush();
}
Instances that implement this interface can then enlist themselves as follows:
MessageProcessorContext.Current.Enlist(this);
Typically, a Repository-class will implement this interface, and when it detects it's managed aggregates are changed/added/removed, it can enlist itself (double enlistments are ignored).
In my case, the framework assumes that you are using an IOC-framework that will resolve all message-handlers and repositories for you, so I made enlistment to the ambient unit of work controller easier by letting it inject an instance of the current IUnitOfWorkManager into the constructor where required. This way the dependencies of the unit of work manager and the actual pieces that require to be flushed (repositories, services, etc) are reversed:
internal sealed class OrderRepository : IOrderRepository, IUnitOfWork
{
private readonly IUnitOfWorkManager _manager;
private readonly Dictionary<Guid, Order> _orders;
public OrderRepository(IUnitOfWorkManager manager)
{
if (manager == null)
{
throw new ArgumentNullException("manager");
}
_manager = manager;
}
bool IUnitOfWork.RequiresFlush()
{
return _orders.Values.Any(order => order.HasChanges());
}
void IUnitOfWork.Flush()
{
// Flush here...
}
public void Add(Order order)
{
_orders.Add(order.Id, order);
_manager.Enlist(this);
}
}
As soon as a request has been handled succesfully (no exceptions thrown), scope.Complete() will be called which triggers the controller to check with all enlisted items whether they (still) need to be flushed (by calling RequiresFlush()), and if so, flushes them (by calling Flush()).
All in all, this allows for a very maintainable solution (in my perspective) in which new repositories and other dependencies can be added on the fly without changing any master unitofwork class, just like the TransactionManager doesn't need to know upfront which items may take part in any given Transaction.
I am writing a library that will provide a collection of public types to its consumers.
I want to make types from this library dependency injection friendly. This means that every class needs to have a constructor through which it is possible to specify every single dependency of the object being initialized. I also want the library to adhere to the convention over configuration principle. This means that if a consumer wants the default behavior, he may use a parameterless constructor and the object will somehow construct the dependencies for itself.
In example (C#):
public class Samurai {
private readonly IWeapon _weapon;
// consumers will use this constructor most of the time
public Samurai() {
_weapon = ??? // get an instance of the default weapon somehow
}
// consumers will use this constructor if they want to explicitly
// configure dependencies for this instance
public Samurai(IWeapon weapon) {
_weapon = weapon;
}
}
My first solution would be to use the service locator pattern.
The code would look like this:
...
public Samurai() {
_weapon = ServiceLocator.Instance.Get<IWeapon>();
}
...
I have a problem with this, though. Service locator has been flagged as an anti-pattern (link) and I completely agree with these arguments. On the other hand, Martin Fowler advocates use of the service locator pattern exactly in this situation (library projects) (link). I want to be careful and eliminate the possible necessity to rewrite the library after it shows up that service locator really was a bad idea.
So in conclusion - do you think that service locator is fine in this scenario? Should I solve my problem in a completely different way? Any thought is welcome...
If you want to make life easier for users who are not using a DI container, you can provide default instances via a dedicated Defaults class which has methods like this:
public virtual Samurai CreateDefaultSamurai()
{
return new Samurai(CreateDefaultWeapon());
}
public virtual IWeapon CreateDefaultWeapon()
{
return new Shuriken();
}
This way you don't need to pollute the classes themselves with default constructors, and your users aren't at risk of using those default constructors unintentionally.
There is an alternative, that is injecting a specific provider, let's say a WeaponProvider in your case into your class so it can do the lookup for you:
public interface IWeaponProvider
{
IWeapon GetWeapon();
}
public class Samurai
{
private readonly IWeapon _weapon;
public Samurai(IWeaponProvider provider)
{
_weapon = provider.GetWeapon();
}
}
Now you can provide a local default provider for a weapon:
public class DefaultWeaponProvider : IWeaponProvider
{
public IWeapon GetWeapon()
{
return new Sword();
}
}
And since this is a local default (as opposed to one from a different assembly, so it's not a "bastard injection"), you can use it as part of your Samurai class as well:
public class Samurai
{
private readonly IWeapon _weapon;
public Samurai() : this(new DefaultWeaponProvider())
{
}
public Samurai(IWeaponProvider provider)
{
_weapon = provider.GetWeapon();
}
}
I have used the following approach in my C# project. The goal was to achieve dependency injection (for unit / mock testing) whilst not crippling the implementation of the code for a "normal use case" (i.e. having a large amount of new()'s that are cascaded through the execution flow).
public sealed class QueueProcessor : IQueueProcessor
{
private IVbfInventory vbfInventory;
private IVbfRetryList vbfRetryList;
public QueueProcessor(IVbfInventory vbfInventory = null, IVbfRetryList vbfRetryList = null)
{
this.vbfInventory = vbfInventory ?? new VbfInventory();
this.vbfRetryList = vbfRetryList ?? new VbfRetryList();
}
}
This allows DI but also means any consumer doesn't have to worry about what the "default instance flow" should be.
Let's say we have class:
public class WithDependencies
{
public WithDependencies(IAmDependencyOne first, IAmDependencyTwo second)
// ...
}
Now the question. How do you create objects of WithDependencies class in an application?
I know there are many ways.
new WithDependencies(new DependencyOne(), new DependencyTwo());
new WithDependencies(IoC.Resolve(IDependencyOne), IoC.Resolve(IDependencyTwo());
// register IDependencyOne, IDependencyTwo implementations at app start
IoC.Resolve(WithDependencies);
// register IDependencyOne, IDependencyTwo implementations at app start
// isolate ourselves from concrete IoC Container
MyCustomWithDependenciesFactory.Create();
and so on...
What do you think is the way to do it?
Edit:
Because I don't get answers or I don't understand them I'll try to ask again. Let's say that on some event (button, timer, whatever) I need new object WithDependencies(). How do I create it? Assume IoC container is already configured.
It depends on the context, so it's impossible to provide a single answer. Conceptually you'd be doing something like this from the Composition Root:
var wd = new WithDependencies(new DependencyOne(), new DependencyTwo());
However, even in the absence of a DI Container, the above code isn't always unambiguously the correct answer. In some cases, you might want to share the same dependency among several consumers, like this:
var dep1 = new DependencyOne();
var wd = new WithDependencies(dep1, new DependencyTwo());
var another = AnotherWithDependencies(dep1, new DependencyThree());
In other cases, you might not want to share dependencies, in which case the first option is more correct.
This is just a small glimpse of an entire dimension of DI concerned with Lifetime Management. Many DI Containers can take care of that for you, which is one excellent argument to prefer a DI Container over Poor Man's DI.
Once you start using a DI Container, you should follow the Register Resolve Release pattern when resolving types, letting Auto-wiring take care of the actual composition:
var wd = container.Resolve<WithDependencies>();
The above example assumes that the container is already correctly configured.
If you need to create a dependency which has its own dependencies, you can either A) do it yourself, or B) ask something else to do it for you. Option A negates the benefits of dependency injection (decoupling, etc.), so I would say option B is a better starting point. Now, we have chosen to use the factory pattern, no matter whether it takes the form of a service locator (i.e. IoC.Resolve), a static factory, or an instance factory. The point is that we have delegated that responsibility to an external authority.
There are a number of trade-offs required for static accessors. (I went over them in another answer, so I won't repeat them here.) In order to avoid introducing a dependency on the infrastructure or the container, a solid option is to accept a factory for creating WithDependencies when we need an instance somewhere else:
public class NeedsWithDependencies
{
private readonly IWithDependenciesFactory _withDependenciesFactory;
public NeedsWithDependencies(IWithDependenciesFactory withDependenciesFactory)
{
_withDependenciesFactory = withDependenciesFactory;
}
public void Foo()
{
var withDependencies = _withDependenciesFactory.Create();
...Use the instance...
}
}
Next, we can create a container-specific implementation of the factory:
public class WithDependenciesFactory : IWithDependenciesFactory
{
private readonly IContainer _container;
public WithDependenciesFactory(IContainer container)
{
_container = container
}
public WithDependencies Create()
{
return _container.Resolve<WithDependencies>();
}
}
Now NeedsWithDependencies is completely isolated from any knowledge of how WithDependencies gets created; it also exposes all its dependencies in its constructor, instead of hiding dependencies on static accessors, making it easy to reuse and test.
Defining all those factories can get a little cumbersome, though. I like Autofac's factory relationship type, which will detect parameters of the form Func<TDependency> and automatically inject a function which serves the same purpose as the hand-coded factory above:
public class NeedsWithDependencies
{
private readonly Func<WithDependencies> _withDependenciesFactory;
public NeedsWithDependencies(Func<WithDependencies> withDependenciesFactory)
{
_withDependenciesFactory = withDependenciesFactory;
}
public void Foo()
{
var withDependencies = _withDependenciesFactory();
...Use the instance...
}
}
It also works great with runtime parameters:
public class NeedsWithDependencies
{
private readonly Func<int, WithDependencies> _withDependenciesFactory;
public NeedsWithDependencies(Func<int, WithDependencies> withDependenciesFactory)
{
_withDependenciesFactory = withDependenciesFactory;
}
public void Foo(int x)
{
var withDependencies = _withDependenciesFactory(x);
...Use the instance...
}
}
Sometimes I try to get rid of factories or at least not depend directly on them, so Dependency Injection (without factories) is useful of course.
Therefore I use Google Juice, cause its a small little framework using Java Annotations and you can quickly change your injections / dependencies. Just take a look at it:
http://code.google.com/p/google-guice/
I have been doing my first Test Driven Development project recently and have been learning Ninject and MOQ. This is my first attempt at all this. I've found the TDD approach has been thought provoking, and Ninject and MOQ have been great. The project I am working on has not particularly been the best fit for Ninject as it is a highly configurable C# program that is designed to test the use of a web service interface.
I have broken it up into modules and have interfaces all over the shop, but I am still finding that I am having to use lots of constructor arguments when getting an implementation of a service from the Ninject kernel. For example;
In my Ninject module;
Bind<IDirEnum>().To<DirEnum>()
My DirEnum class;
public class DirEnum : IDirEnum
{
public DirEnum(string filePath, string fileFilter,
bool includeSubDirs)
{
....
In my Configurator class (this is the main entry point) that hooks all the services together;
class Configurator
{
public ConfigureServices(string[] args)
{
ArgParser argParser = new ArgParser(args);
IDirEnum dirEnum = kernel.Get<IDirEnum>(
new ConstructorArgument("filePath", argParser.filePath),
new ConstructorArgument("fileFilter", argParser.fileFilter),
new ConstructorArgument("includeSubDirs", argParser.subDirs)
);
filePath, fileFilter and includeSubDirs are command line options to the program. So far so good. However, being a conscientious kind of guy, I have a test covering this bit of code. I'd like to use a MOQ object. I have created a Ninject module for my tests;
public class TestNinjectModule : NinjectModule
{
internal IDirEnum mockDirEnum {set;get};
Bind<IDirEnum>().ToConstant(mockDirEnum);
}
And in my test I use it like this;
[TestMethod]
public void Test()
{
// Arrange
TestNinjectModule testmodule = new TestNinjectModule();
Mock<IDirEnum> mockDirEnum = new Mock<IDirEnum>();
testModule.mockDirEnum = mockDirEnum;
// Act
Configurator configurator = new Configurator();
configurator.ConfigureServices();
// Assert
here lies my problem! How do I test what values were passed to the
constructor arguments???
So the above shows my problem. How can I test what arguments were passed to the ConstructorArguments of the mock object? My guess is that Ninject is dispensing of the ConstuctorArguments in this case as the Bind does not require them? Can I test this with a MOQ object or do I need to hand code a mock object that implements DirEnum and accepts and 'records' the constructor arguments?
n.b. this code is 'example' code, i.e. I have not reproduced my code verbatim, but I think I have expressed enough to hopefully convey the issues? If you need more context, please ask!
Thanks for looking. Be gentle, this is my first time ;-)
Jim
There are a few problems with the way you designed your application. First of all, you are calling the Ninject kernel directly from within your code. This is called the Service Locator pattern and it is considered an anti-pattern. It makes testing your application much harder and you are already experiencing this. You are trying to mock the Ninject container in your unit test, which complicates things tremendously.
Next, you are injecting primitive types (string, bool) in the constructor of your DirEnum type. I like how MNrydengren states it in the comments:
take "compile-time" dependencies
through constructor parameters and
"run-time" dependencies through method
parameters
It's hard for me to guess what that class should do, but since you are injecting these variables that change at run-time into the DirEnum constructor, you end up with a hard to test application.
There are multiple ways to fix this. Two that come in mind are the use of method injection and the use of a factory. Which one is feasible is up to you.
Using method injection, your Configurator class will look like this:
class Configurator
{
private readonly IDirEnum dirEnum;
// Injecting IDirEnum through the constructor
public Configurator(IDirEnum dirEnum)
{
this.dirEnum = dirEnum;
}
public ConfigureServices(string[] args)
{
var parser = new ArgParser(args);
// Inject the arguments into a method
this.dirEnum.SomeOperation(
argParser.filePath
argParser.fileFilter
argParser.subDirs);
}
}
Using a factory, you would need to define a factory that knows how to create new IDirEnum types:
interface IDirEnumFactory
{
IDirEnum CreateDirEnum(string filePath, string fileFilter,
bool includeSubDirs);
}
Your Configuration class can now depend on the IDirEnumFactory interface:
class Configurator
{
private readonly IDirEnumFactory dirFactory;
// Injecting the factory through the constructor
public Configurator(IDirEnumFactory dirFactory)
{
this.dirFactory = dirFactory;
}
public ConfigureServices(string[] args)
{
var parser = new ArgParser(args);
// Creating a new IDirEnum using the factory
var dirEnum = this.dirFactory.CreateDirEnum(
parser.filePath
parser.fileFilter
parser.subDirs);
}
}
See how in both examples the dependencies get injected into the Configurator class. This is called the Dependency Injection pattern, opposed to the Service Locator pattern, where the Configurator asks for its dependencies by calling into the Ninject kernel.
Now, since your Configurator is completely free from any IoC container what so ever, you can now easily test this class, by injecting a mocked version of the dependency it expects.
What is left is to configure the Ninject container in the top of your application (in DI terminology: the composition root). With the method injection example, your container configuration would stay the same, with the factory example, you will need to replace the Bind<IDirEnum>().To<DirEnum>() line with something as follows:
public static void Bootstrap()
{
kernel.Bind<IDirEnumFactory>().To<DirEnumFactory>();
}
Of course, you will need to create the DirEnumFactory:
class DirEnumFactory : IDirEnumFactory
{
IDirEnum CreateDirEnum(string filePath, string fileFilter,
bool includeSubDirs)
{
return new DirEnum(filePath, fileFilter, includeSubDirs);
}
}
WARNING: Do note that factory abstractions are in most cases not the best design, as explained here.
The last thing you need to do is to create a new Configurator instance. You can simply do this as follows:
public static Configurator CreateConfigurator()
{
return kernel.Get<Configurator>();
}
public static void Main(string[] args)
{
Bootstrap():
var configurator = CreateConfigurator();
configurator.ConfigureServices(args);
}
Here we call the kernel. Although calling the container directly should be prevented, there will always at least be one place in your application where you call the container, simply because it must wire everything up. However, we try to minimize the number of times the container is called directly, because it improves -among other things- the testability of our code.
See how I didn't really answer your question, but showed a way to work around the problem very effectively.
You might still want to test your DI configuration. That's very valid IMO. I do this in my applications. But for this, you often don't need the DI container, or even if your do, this doesn't mean that all your tests should have a dependency on the container. This relationship should only exist for the tests that test the DI configuration itself. Here is a test:
[TestMethod]
public void DependencyConfiguration_IsConfiguredCorrectly()
{
// Arrange
Program.Bootstrap();
// Act
var configurator = Program.CreateConfigurator();
// Assert
Assert.IsNotNull(configurator);
}
This test indirectly depends on Ninject and it will fail when Ninject is not able to construct a new Configurator instance. When you keep your constructors clean from any logic and only use it for storing the taken dependencies in private fields, you can run this, without the risk of calling out to a database, web service or what so ever.
I hope this helps.
High Level
With StructureMap, Can I define a assembly scan rule that for an interface IRequestService<T> will return the object named TRequestService
Examples:
FooRequestService is injected when IRequestService<FooRequest> is requested
BarRequestService is injected when IRequestService<BarRequest> is requested
Details
I have a generic interface defined
public interface IRequestService<T> where T : Request
{
Response TransformRequest(T request, User current);
}
and then I have multiple Request objects that implement this interface
public class FooRequestService : IRequestService<Foo>
{
public Response TransformRequest(Foo request, User current) { ... }
}
public class BarRequestService : IRequestService<Bar>
{
public Response TransformRequest(Bar request, User current) { ... }
}
Now I am at the point where I need to register these classes so that StructureMap knows how to create them because in my controller I want have the following ctor (which I want StructureMap to inject a FooRequestService into)
public MyController(IRequestService<Foo> fooRequestService) { ... }
Right now to get around my issue I have implemented an empty interface and instead of having the FooRequestService implement the generic interface I have it implement this empty interface
public interface IFooRequestService : IRequestService<Foo> { }
Then my controllers ctor looks like so, which works with StructureMaps' Default Convention Scanner
public MyController(IFooRequestService fooRequestService) { ... }
How could I create a rule with StructureMap's assembly scanner to register all objects named TRequestService with IRequestService<T> (where T = "Foo", "Bar", etc) so that I don't have to create these empty Interface definitions?
To throw something else into the mix, where I am handling StructureMap's assembly scanning does not have any reference to the assembly that defines IRequestService<T> so this has to use some sort of reflection when doing this. I scanned the answer to "StructureMap Auto registration for generic types using Scan" but it seems as though that answer requires a reference to the assembly that contains the interface definition.
I am on the path of trying to write a custom StructureMap.Graph.ITypeScanner but I am kind of stuck on what to do there (mainly because I have little experience with reflection).
You are on the right path with the scanner. Thankfully there is one built into StructureMap. Unfortunately it is not yet, as of this writing, released. Get the latest from trunk and you will see a few new things available within the scanner configuration. An example for your needs is below.
public class MyRegistry : Registry
{
public MyRegistry()
{
Scan(x =>
{
x.TheCallingAssembly();
//x.AssembliesFromApplicationBaseDirectory();
x.WithDefaultConventions();
x.ConnectImplementationsToTypesClosing(typeof (IRequestService<>));
});
}
}
First you need to tell the scanner configuration which assemblies to include in the scan. The commented AssembliesFromApplicationBaseDirectory() method also might help if you are not doing a registry per assembly.
To get your generic types into the container use ConnectImplementationsToTypesClosing.
For an example on how to setup use registries when setting up the container see:
http://structuremap.sourceforge.net/ConfiguringStructureMap.htm
If you like you can skip using registries in general and just do a scan within ObjectFactory.Initialize.
Hope this helps.