How to decorate objects created by a custom factory using .NET Core DI? - .net-core

Given I have a factory class responsible for constructing instances of a certain service that has constructor parameters that can only be resolved at runtime, is there a way to leverage container-driven decoration?
Consider the following class which relies on a parameter that is only defined at runtime:
interface IFooService
{
void DoServicyStuff();
}
class MyFooService : IFooService
{
public MyFooService(string somePeskyRuntimeArgument)
{
this.peskyValue = somePeskyRuntimeArgument;
}
public void DoServicyStuff()
{
// do some stuff here with the peskyValue...
}
}
Since the value can only be provided at runtime, we need to move away from the constructor injection and into a method-level parameter passing. This is commonly achieved using a factory implementation like this:
interface IFooServiceFactory
{
IFooService CreateService(string heyItsNowAMethodLevelPeskyParameter);
}
class FooServiceFactory : IFooServiceFactory
{
public IFooService CreateService(string heyItsNowAMethodLevelPeskyParameter)
{
return new MyFooService(heyItsNowAMethodLevelPeskyParameter);
}
}
While this works fine if the intent is to just abstract away the construction of the service, it poses a challenge to decorate the IFooService instance.
For scenarios where no runtime parameter is involved, this can be easily achieved by tapping into the container to provide our service for us. The example below uses the Scrutor library to decorate the interface with a logging decorator implementation:
class FooServiceFactory : IFooServiceFactory
{
private readonly IServiceProvider serviceProvider;
public FooServiceFactory(IServiceProvider serviceProvider)
{
this.serviceProvider = serviceProvider
}
public IFooService CreateService(string heyItsNowAMethodLevelPeskyParameter)
{
return this.serviceProvider.GetRequiredInstance<IFooService>();
}
}
...
services
.AddTransient<IFooService, MyFooService>()
.AddTransient<IFooServiceFactory, FooServiceFactory>()
.Decorate<IFooService, LoggingFooService>();
But since MyFooService takes a primitive value as an argument, we cannot rely on GetRequiredService<T> to obtain the instance, as it will fail to find "a registration for string" when building the concrete class.
Similarly, changing the factory to rely on ActivatorUtilities's .CreateInstance or .CreateFactory methods will end up creating the objects while completely ignoring the container registrations, thus leaving us without any decorator.
I know I have at least 2 options to decorate the objects manually, namely:
Using the factory itself to manually create the decorator:
public IFooService CreateService(string heyItsNowAMethodLevelPeskyParameter)
{
return new LoggingService(
new MyFooService(heyItsNowAMethodLevelPeskyParameter));
}
Using a factory decorator to inject a decorator after the instance is created:
abstract class FooServiceFactoryDecorator : IFooServiceFactory
{
private readonly IFooServiceFactory fooServiceFactory;
protected FooServiceFactory(IFooServiceFactory fooServiceFactory)
{
this.fooServiceFactory = fooServiceFactory;
}
public virtual IFooService CreateService(string heyItsNowAMethodLevelPeskyParameter)
{
return this.fooServiceFactory.CreateService(heyItsNowAMethodLevelPeskyParameter);
}
}
class LoggingFooServiceFactory : FooServiceFactoryDecorator
{
private readonly IFooServiceFactory fooServiceFactory;
public FooServiceFactory(IFooServiceFactory fooServiceFactory)
{
this.fooServiceFactory = fooServiceFactory;
}
public override IFooService CreateService(string heyItsNowAMethodLevelPeskyParameter)
{
return new LoggingFooService(
this.fooServiceFactory.CreateService(heyItsNowAMethodLevelPeskyParameter));
}
}
...
services
.AddTransient<IFooServiceFactory, FooServiceFactory>()
.Decorate<IFooServiceFactory, LoggingFooServiceFactory>()
Neither of these allows me to directly use .Decorate on top of the service interface. The first option works but is heavily coupled (meaning I'd have to keep changing it if I want to add other decorators into the mix), while the second version is less coupled, but still forces me to writing one factory decorator per service decorator and thus leads into a much more complex solution.
Another pain point is dependencies on the decorators themselves (for example, ILogger<T> on the LoggingFooService), which I could potentially solve by leveraging ActivatorUtilities to create the decorators instead of newing them up manually.
I could also potentially generalize the "factory decorator" so that the decoration function is parameterized and thus the class can be reused, but it is still very convoluted and hard to maintain, while also not providing as good a syntax for consumers to add new decorators.
class DecoratedFooServiceFactory<TDecorator> : FooServiceFactoryDecorator
where TDecorator : IFooService
{
private readonly IFooServiceFactory fooServiceFactory;
private readonly IServiceProvider serviceProvider;
public FooServiceFactory(
IFooServiceFactory fooServiceFactory,
IServiceProvider serviceProvider)
{
this.fooServiceFactory = fooServiceFactory;
this.serviceProvider = serviceProvider;
}
public override IFooService CreateService(string heyItsNowAMethodLevelPeskyParameter)
{
return ActivatorUtilities.CreateInstance<TDecorator>(
this.serviceProvider,
this.fooServiceFactory.CreateService(heyItsNowAMethodLevelPeskyParameter));
}
}
...
services
.AddTransient<IFooServiceFactory, FooServiceFactory>()
.Decorate<IFooServiceFactory, DecoratedFooServiceFactory<LoggingFooService>>()
And finally, if I ever want to move away from using a factory and want to change to using the service directly, this will cause a significant setup change where I'd then have to configure all the decorators again in the container directly instead of just removing the factory registration as one normally would do.
How can I use a factory like this, while still keeping the capability of configuring decorators at the container level using the simple Scrutor syntax?

Ok, a couple of disclaimers first:
I agree with Steven here in that this looks like an anti-pattern and you will probably be better off redesigning your code to not require run-time values on service construction.
I additionally want to caution against using scrutor-like Decorate. While much less confident in this than in the first point, I believe hiding logging in decorators is much less convenient in the long run than it seems at first. Or at least that's what I saw after about a year of trying them out.
That said, let's see what can be done.
First, let's put some constraints on where the value is coming from. Specifically, let's say we can have a service providing that value, that looks like this:
public interface IValueProvider
{
string Get();
}
This actually allows us to have quite a bit of range. Implementation of that interface can:
Get value from external API - once or periodically in the background. It can even call it every time Get is called, but this is a very bad idea, as it will make construction asynchronous.
Get value that is stored in memory and allow some other service to update it. Say, expose a 'configuration' endpoint where a user can set a new value every once in a while.
Calculate the value based on some algorithm of your choice.
Once you have this service, you can register it like this:
public void ConfigureServices(IServiceCollection services)
{
services.AddSingleton<IValueProvider, AwesomeValueProvider>();
services.AddSingleton<IFooServiceFactory, FooServiceFactory>();
services.AddTransient<IFooService>(sp =>
{
var factory = sp.GetRequiredService<IFooServiceFactory>();
var valueProvider = sp.GetRequiredService<IValueProvider>();
return factory.Create(valueProvider.Get());
});
}
Hope this helps

Related

Blazor. Task from another service

I'm studying blazor server.
Deployed a solution from a standard vs template.
Created two server-side services, TestService1 and TestService2.
In TestService1 i have task
GetMyData()
How can i call with task from TestService2?
If i trying
var serv1 = new TestService1()
i have to fill in all the variables of the constructor that is in TestService1.
What is easiest way?
In line with the comment on your question, the best way to go about this in Blazor is to utilize the built-in dependency injection mechanism.
I assume that your services look like the following:
public class TestService1
{
public object GetMyData()
{
}
}
public class TestService2
{
private readonly TestService1 _testService1 { get; set; }
public class TestService2(TestService1 ts1)
{
_testService1 = ts1;
}
public void DoesSomething()
{
var data = _testService1.GetMyData();
//...
}
}
First, you'd need to register these with Blazor at startup, so in your Startup.cs in the ConfigureServices method, add the following, assuming you have an empty constructor available for TestService1:
services.AddSingleton<TestService1>();
Because you'll need to instantiate an instance of TestService1 into TestService2 to call methods on it, you'll have to handle registration of TestService2 differently since you'll need to procure an instance of TestService1 from the DI service to instantiate it:
services.AddSingleton<TestService2>(s => {
var testService1 = s.GetService<TestService1>();
return new TestService2(testService1);
});
It's possible you may need to scope the services differently (e.g. used scoped instead of singletons). You can read about the differences here.
Now something is presumably calling TestService2 to kick all this off, so let's pretend it's running in a component in your Blazor app. You'd inject TestService2 into the component with the following:
#inject TestService2 _testService2
<h1>Hello!</h1>
#code {
protected override void OnInitialized()
{
_testService2.DoesSomething();
}
}
As part of the initialization then of this component, it'll automatically inject a TestService2 instance (based on the scoping you specified at DI initialization) to your component and will call the DoesSomething method on it. When injected, it looks to DI to instantiate the TestService1 service to the constructor as you've also specified, leaving it free to call that method and the call commences as intended.
Please let me know if you'd like any clarification somewhere!

"Add" appears to be working in WebFlux, but seems like there must be a "more accepted way"

New to WebFlux, reactive, and handlers.
I am able to get a Mono<> from a ServerRequest and process the contained POJO to add a new tuple to a database. But, it seems like there should be a "better" or "more accepted" way to write this code.
Any help/input with the code in AccountRequestHandler would be appreciated, especially with explanations of the rationale behind the recommend change(s).
Router implementation (stripped down to only "POST")...
#Configuration
public class AccountRequestRouter {
#Bean
public RouterFunction<ServerResponse> route(AccountRequestHandler requestHandler) {
return nest(path("/v2"),
nest(accept(APPLICATION_JSON),
.andRoute(RequestPredicates.POST("/accounts"), requestHandler::addAccount)
));
}
}
Handler implementation...
The code where I'm actually doing the add, and then separately creating a ServerResponse, is what I'm focused on. It seems "clunky", especially since AccountService.addAccount() returns a Mono on completion.
#Component
public class AccountRequestHandler {
#Autowired
private mil.navy.ccop.service.accounts.account.AccountService accountService;
public Mono<ServerResponse> addAccount(ServerRequest request) {
return request.bodyToMono(Account.class).flatMap(account -> {
accountService.addAccount(account);
return ServerResponse.ok().build();
})
.switchIfEmpty(ServerResponse.badRequest()
.contentType(APPLICATION_JSON)
.build(Mono.empty()));
}
}
AccountService implementation (again, stripped down)...
#Service
class AccountService {
#Autowired
private AccountRepository accounts;
public AccountService() {
}
public Mono<Void> addAccount(Account account) {
Account proxy;
// make sure that accountId is set to support auto-generation of synthetic key value
proxy = new Account(-1, account.getShortName(), account.getLongName(), account.getDescription());
accounts.save(proxy);
return Mono.empty();
}
}
Appreciating all the help in ramping up on this style of programming....
well first of all, you have 2 addAccount, that can be a bit confusing.
Second of all, what kind of "repository" are you writing too? if its an sql repo you need to properly wrap it in a Mono.fromCallable() otherwise it will block the Reactive thread pool and you can have really bad performance.
Yes there are other ways of doing things. A lot of people tend to do things in flatmap or map and sure it is completely possible to do things here, but for the semantics i'd say it is less good.
map and flatmap are usually used to perform some sort of computation on the inner value of the mono and then return the same or a new value and or type inside the mono.
i would rewrite this like such.
return void here:
public void addAccount(Account account) {
Account proxy;
// make sure that accountId is set to support auto-generation of synthetic key value
proxy = new Account(-1, account.getShortName(), account.getLongName(), account.getDescription());
accounts.save(proxy);
}
And here:
public Mono<ServerResponse> addAccount(ServerRequest request) {
return request.bodyToMono(Account.class)
.doOnSuccess(account -> {
accountService.addAccount(account);
}).then(ServerResponse.ok().build())
.switchIfEmpty(ServerResponse.badRequest()
.contentType(APPLICATION_JSON)
.build());
}
there are a number of different doOn methods that are ment to be used to consume and do "side effects" on things. Like doOnSuccess, doOnError, doOnCancel etc. etc.
you also have then and thenReturn which will just return whatever you put in them. Then returns whatever Mono you put in it. thenReturn wraps whatever value you put into it into a Mono and returns it.

Handling specimen creation inconsistencies between AutoFixture and Moq

I am using AutoMoqCustomization in my test conventions.
Consider the code below. Everything works great until I add a constructor to one of the concrete classes. When I do, I get "could not find a parameterless constructor". We know AutoFixture doesn't have an issue with the constructor because it delivered me the test object one which proved to be assignable from IThings... no failure there. So it must be moq.
This makes some sense because I assume builder was generated by moq and passed into the GetCommands method. So I think I can see that control has been passed from AutoFixture to moq at that point.
That takes care of the why, but what should I do about it? Is there a way to instruct moq on how to deal with the ThingOne or is there a way to instruct AutoFixture to ignore moq for IThingBuilders and instead do something Fixtury?
public class TestClass
{
public interface IThingBuilders
{
T1 Build<T1>() where T1 : IThings;
}
public interface IThings
{
}
public class ThingOne : IThings
{
public ThingOne(string someparam)
{
}
}
public class ThingTwo : IThings
{
}
public class SomeClass
{
public List<IThings> GetCommands(IThingBuilders builder)
{
var newlist = new List<IThings>();
newlist.Add(builder.Build<ThingOne>());
newlist.Add(builder.Build<ThingTwo>());
return newlist;
}
}
[Theory, BasicConventions]
public void WhyCannotInstantiateProxyOfClass(ThingOne one, ThingTwo two, IThingBuilders builder, SomeClass sut)
{
Assert.IsAssignableFrom<IThings>(one);
Assert.IsAssignableFrom<IThings>(two);
var actual = sut.GetCommands(builder);
Assert.Equal(1, actual.OfType<ThingOne>().Count());
Assert.Equal(1, actual.OfType<ThingTwo>().Count());
}
}
As there's no extensibility point in Moq that enables AutoFixture to hook in and supply a value of ThingOne, there's not a whole lot you can do.
However, you can use the SetReturnsDefault<T> method of Moq. Modifying the above test would then be like this:
[Theory, BasicConventions]
public void WhyCannotInstantiateProxyOfClass(
ThingOne one, ThingTwo two, IThingBuilders builder, SomeClass sut)
{
Assert.IsAssignableFrom<IThings>(one);
Assert.IsAssignableFrom<IThings>(two);
Mock.Get(builder).SetReturnsDefault(one); // Add this to make the test pass
var actual = sut.GetCommands(builder);
Assert.Equal(1, actual.OfType<ThingOne>().Count());
Assert.Equal(1, actual.OfType<ThingTwo>().Count());
}
This is a bit easier than having to write a specific Setup/Returns pair, but not much. You could move that code to an AutoFixture Customization, but again, since this is a generic method on a a Mock instance, you'll explicitly need to call this for e.g. ThingOne in order to set the default for that return type. Not particularly flexible.

Verifying indirectly called methods with Moq on a mocked object

My app has a ProviderFactory static class that has static utility methods passing back static instances of things like a logger. The rest of my app then can just grab a/the reference to the logger from anywhere without having to pass in the logger (common design practice).
So, another part of my app, the DbCacheProvider, has methods that make calls to the logger so internally it gets a reference to the logger from the factory and then issues calls to it.
My question is that using Moq, I want to verify methods on the logger are being called by the methods within the DbCacheProvider. I can do this using dependency injection when I pass a mock logger into the DbCacheProvider as a parameter, but I'm not passing the logger in (not do I want to). So, how would I verify the DbCacheProvider is making calls to the logger?
If you don't want to pass the logger in through the constructor you'd need to change your ProviderFactory while running unit tests to return your mocked logger.
Anyway there are a couple of reasons it's often suggested to set up dependency injection:
Your tests are more straightforward and don't involve finagling with custom factories
IoC frameworks like Unity, Ninject and Autofac make it easy to create objects when their dependencies are set up this way. If you set up all of your objects this way, the framework will do all the heavy lifting of creating the right objects and passing them in for you. The dependency injection is done automatically and won't be a burden for you.
Old question without an answer, I had a similar problem and solved it like this:
I have the following sample code and need to verify that not only was a method called but was called with a specific value.
public interface ILog
{
void Info(string message);
}
public interface ILogFactory
{
ILog GetLogger();
}
This is the class being tested, where the interface items are being injected:
public class NewAction
{
readonly ILogFactory _logger;
public NewAction(ILogFactory logger)
{
_logger = logger;
}
public void Step1()
{
_logger.GetLogger().Info("Step 1");
}
public void Step2()
{
_logger.GetLogger().Info("Step 2");
}
}
This is obviously a very simplistic view of my actual code, but I needed to verify that Step1 and Step2 are behaving as expected and passed the correct values to the Log, this would mean I also needed to ensure they occurred in the right order. My test:
[TestClass]
public class UnitTest1
{
[TestMethod]
public void TestMethod1()
{
// Arrange
var log = new Mock<ILog>();
var factory = new Mock<ILogFactory>();
factory.Setup(l => l.GetLogger()).Returns(log.Object);
// Act
var action = new NewAction(factory.Object);
action.Step1();
action.Step2();
// Assert
factory.Verify(l => l.GetLogger());
log.Verify(l => l.Info(It.Is<string>(s => s == "Step 1")));
log.Verify(l => l.Info(It.Is<string>(s => s == "Step 2")));
}
}
Hope this helps.

How to pass unit of work container into constructor of repository using dependency injection

I'm trying to work out how to complete my implementation of the Repository pattern in an ASP.NET web application.
At the moment, I have a repository interface per domain class defining methods for e.g. loading and saving instances of that class.
Each repository interface is implemented by a class which does the NHibernate stuff. Castle Windsor sorts out the DI of the class into the interface according to web.config. An example of an implemented class is provided below:
public class StoredWillRepository : IStoredWillRepository
{
public StoredWill Load(int id)
{
StoredWill storedWill;
using (ISession session = NHibernateSessionFactory.OpenSession())
{
storedWill = session.Load<StoredWill>(id);
NHibernateUtil.Initialize(storedWill);
}
return storedWill;
}
public void Save(StoredWill storedWill)
{
using (ISession session = NHibernateSessionFactory.OpenSession())
{
using (ITransaction transaction = session.BeginTransaction())
{
session.SaveOrUpdate(storedWill);
transaction.Commit();
}
}
}
}
As pointed out in a previous thread, the repository class needs to accept an unit of work container (i.e. ISession) rather than instantiating it in every method.
I anticipate that the unit of work container will be created by each aspx page when needed (for example, in a property).
How do I then specify that this unit of work container instance is to be passed into the constructor of StoredWillRepository when Windsor is creating it for me?
Or is this pattern completely wrong?
Thanks again for your advice.
David
I have a persistence framework built on top of NHibernate that is used in a few Web apps. It hides the NH implementation behind an IRepository and IRepository<T> interface, with the concrete instances provided by Unity (thus I could in theory swap out NHibernate for, say, Entity Framework fairly easily).
Since Unity doesn't (or at least the version I'm using doesn't) support the passing in of constructor parameters other than those that are dependency injections themselves, passing in an extant NH ISession isn't possible; but I do want all objects in the UOW to share the same ISession.
I solve this by having a controlling repository class that manages access to the ISession on a per-thread basis:
public static ISession Session
{
get
{
lock (_lockObject)
{
// if a cached session exists, we'll use it
if (PersistenceFrameworkContext.Current.Items.ContainsKey(SESSION_KEY))
{
return (ISession)PersistenceFrameworkContext.Current.Items[NHibernateRepository.SESSION_KEY];
}
else
{
// must create a new session - note we're not caching the new session here... that's the job of
// BeginUnitOfWork().
return _factory.OpenSession(new NHibernateInterceptor());
}
}
}
}
In this example, PersistenceFrameworkContext.Current.Items accesses an IList<object> that is stored either ThreadStatic if not in a Web context, or within HttpContext.Current.Items if it is in a Web context (to avoid thread-pool problems). The first call to the property instantiates the ISession from the stored factory instance, subsequent calls just retrieve it from storage. The locking will slow things down slightly but not as much as just locking an appdomain-scoped static ISession instance.
I then have BeginUnitOfWork and EndUnitOfWork methods to take care of the UOW - I have specifically disallowed nested UOWs because frankly they were a pain to manage.
public void BeginUnitOfWork()
{
lock (_lockObject)
{
if (PersistenceFrameworkContext.Current.Items.ContainsKey(SESSION_KEY))
EndUnitOfWork();
ISession session = Session;
PersistenceFrameworkContext.Current.Items.Add(SESSION_KEY, session);
}
}
public void EndUnitOfWork()
{
lock (_lockObject)
{
if (PersistenceFrameworkContext.Current.Items.ContainsKey(SESSION_KEY))
{
ISession session = (ISession)PersistenceFrameworkContext.Current.Items[SESSION_KEY];
PersistenceFrameworkContext.Current.Items.Remove(SESSION_KEY);
session.Flush();
session.Dispose();
}
}
}
Finally, a pair of methods provide access to the domain-type-specific repositories:
public IRepository<T> For<T>()
where T : PersistentObject<T>
{
return Container.Resolve<IRepository<T>>();
}
public TRepository For<T, TRepository>()
where T : PersistentObject<T>
where TRepository : IRepository<T>
{
return Container.Resolve<TRepository>();
}
(Here, PersistentObject<T> is a base class providing ID and Equals support.)
Access to a given repository is thus in the pattern
NHibernateRepository.For<MyDomainType>().Save();
This is then facaded over such that you can use
MyDomainType.Repository.Save();
Where a given type has a specialised repository (ie needs more than it can get from IRepository<T>) then I create an interface deriving from IRepository<T>, an extending implementation inheriting from my IRepository<T> implementation, and in the domain type itself I override the static Repository property using new
new public static IUserRepository Repository
{
get
{
return MyApplication.Repository.For<User, IUserRepository>();
}
}
(MyApplication [which is called something less noddy in the real product] is a facade class which takes care of supplying the Repository instance via Unity so you have no dependency on the specific NHibernate repository implementation within your domain classes.)
This gives me full pluggability via Unity for the repository implementation, easy access to the repository in code without jumping through hoops, and transparent, per-thread ISession management.
There's lots more code than just what's above (and I've simplified the example code a great deal), but you get the general idea.
MyApplication.Repository.BeginUnitOfWork();
User user = User.Repository.FindByEmail("wibble#wobble.com");
user.FirstName = "Joe"; // change something
user.LastName = "Bloggs";
// you *can* call User.Repository.Save(user), but you don't need to, because...
MyApplication.Repository.EndUnitOfWork();
// ...causes session flush which saves the changes automatically
In my Web app, I have session-per-request, so BeginUnitOfWork and EndUnitOfWork get called in BeginRequest and EndRequest respectively.
I have a pretty similar structure to yours, and here's how I solve your question:
1) To specify my container on each method, I have a separate class ("SessionManager") which I then invoke via a static property. By doing so, here's an example using my Save implementation:
private static ISession NHibernateSession
{
get { return SessionManager.Instance.GetSession(); }
}
public T Save(T entity)
{
using (var transaction = NHibernateSession.BeginTransaction())
{
ValidateEntityValues(entity);
NHibernateSession.Save(entity);
transaction.Commit();
}
return entity;
}
2) My container is not created on each ASPX page. I instantiate all of my NHibernate goodness on the global.asax page.
** A few more things spring up **
3) You don't need to have a helper to instantiate the Load. You might as well use Get instead of Load. More information # Difference between Load and Get.
4) Using your current code, you would have to repeat pretty much the same code for each domain object you need (StoredWillRepository, PersonRepository, CategoryRepository, etc..?), which seems like a drag. You could very well use a generic class to operate over NHibernate, like:
public class Dao<T> : IDao<T>
{
public T SaveOrUpdate(T entity)
{
using (var transaction = NHibernateSession.BeginTransaction())
{
NHibernateSession.SaveOrUpdate(entity);
transaction.Commit();
}
return entity;
}
}
In my implementation, I could then use something like:
Service<StoredWill>.Instance.SaveOrUpdate(will);
Technically, the answer to my question is to use the overload of container.Resolve which allows you to specify the constructor argument as an anonymous type:
IUnitOfWork unitOfWork = [Code to get unit of work];
_storedWillRepository = container.Resolve<IStoredWillRepository>(new { unitOfWork = unitOfWork });
But let's face it, the answers provided by everyone else have been much more informative.

Resources