As a follow-up to this question, I wanted to understand how my invoking of a Service manually can be improved. This became longer than I wanted, but I feel the background info is needed.
When doing a pub/sub (broadcast), the normal sequence and flow in the Messaging API isn't used, and I instead get a callback when a pub/sub message is received, using IRedisClient, IRedisSubscription:
_subscription.OnMessage = (channel, msg) =>
{
onMessageReceived(ParseJsonMsgToPoco(msg));
};
The Action onMessageReceived will then, in turn, invoke a normal .NET/C# Event, like so:
protected override void OnMessageReceived(MyRequest request)
{
OnMyEvent?.Invoke(this, new RequestEventArgs(request));
}
This works, I get my request and all that, however, I would like it to be streamlined into the other flow, the flow in the Messaging API, meaning, the request finds its way into a Service class implementation, and that all normal boilerplate and dependency injection takes place as it would have using Messaging API.
So, in my Event handler, I manually invoke the Service:
private void Instance_OnMyEvent(object sender, RequestEventArgs e)
{
using (var myRequestService = HostContext.ResolveService<MyRequestService>(new BasicRequest()))
{
myRequestService.Any(e.Request);
}
}
and the MyRequestService is indeed found and Any called, and dependency injection works for the Service.
Question 1:
Methods such as OnBeforeExecute, OnAfterExecute etc, are not called, unless I manually call them, like: myRequestService.OnBeforeExecute(e) etc. What parts of the pipeline is lost? Can it be reinstated in some easy way, so I don't have to call each of them, in order, manually?
Question 2:
I think I am messing up the DI system when I do this:
using (var myRequestService = HostContext.ResolveService<MyRequestService>(new BasicRequest()))
{
myRequestService.OnBeforeExecute(e.Request);
myRequestService.Any(e.Request);
myRequestService.OnAfterExecute(e.Request);
}
The effect I see is that the injected dependencies that I have registered with container.AddScoped, isn't scoped, but seems static. I see this because I have a Guid inside the injected class, and that Guid is always the same in this case, when it should be different for each request.
container.AddScoped<IRedisCache, RedisCache>();
and the OnBeforeExecute (in a descendant to Service) is like:
public override void OnBeforeExecute(object requestDto)
{
base.OnBeforeExecute(requestDto);
IRedisCache cache = TryResolve<IRedisCache>();
cache?.SetGuid(Guid.NewGuid());
}
So, the IRedisCache Guid should be different each time, but it isn't. This however works fine when I use the Messaging API "from start to finish". It seems that if I call the TryResolve in the AppHostBase descendant, the AddScoped is ignored, and an instance is placed in the container, and then never removed.
What parts of the pipeline is lost?
None of the request pipeline is executed:
myRequestService.Any(e.Request);
Is physically only invoking the Any C# method of your MyRequestService class, it doesn't (nor cannot) do anything else.
The recommended way for invoking other Services during a Service Request is to use the Service Gateway.
But if you want to invoke a Service outside of a HTTP Request you can use the RPC Gateway for executing non-trusted services as it invokes the full Request Pipeline & converts HTTP Error responses into Typed Error Responses:
HostContext.AppHost.RpcGateway.ExecuteAsync()
For executing internal/trusted Services outside of a Service Request you can use HostContext.AppHost.ExecuteMessage as used by ServiceStack MQ which applies Message Request Request/Response Filters, Service Action Filters & Events.
I have registered with container.AddScoped
Do not use Request Scoped dependencies outside of a HTTP Request, use Singleton if the dependencies are ThreadSafe, otherwise register them as Transient. If you need to pass per-request storage pass them in IRequest.Items.
Related
I am in the process of migrating NServiceBus up to v6 and am at a roadblock in the process of removing reference to IBus.
We build upon a common library for many of our applications (Website, Micro Services etc) and this library has the concept of IEventPublisher which is essentially a Send and Publish interface. This library has no knowledge of NSB.
We can then supply the implementation of this IEventPublisher using DI from the application, this allows the library's message passing to be replaced with another technology very easily.
So what we end up with is an implementation similar to
public class NsbEventPublisher : IEventPublisher
{
IEndpointInstance _instance;
public NsbEventPublisher(IEndpointInstance endpoint)
{
instance = endpoint;
}
public void Send(object message)
{
instance.Send(message, sendOptions);
}
public void Publish(object message)
{
instance.Publish(message, sendOptions);
}
}
This is a simplification of what actually happens but illustrates my problem.
Now when the DI container is asked for an IEventPublisher it knows to return a NsbEventPublisher and it knows to resolve the IEndpointInstance as we bind this in the bootstrapper for the website to the container as a singleton.
All is fine and my site runs perfect.
I am now migrating the micro-services (running in NSB.Host) and the DI container is refusing to resolve IEndpointInstance when resolving the dependencies within a message handler. Reading the docs this is intentional and I should be using IMessageHandlerContext when in a message handler.
https://docs.particular.net/nservicebus/upgrades/5to6/moving-away-from-ibus
The docs even elude to the issue I have in the bottom example around the class MyContextAccessingDependency. The suggestion is to pass the message context through the method which puts a hard dependency on the code running in the context of a message handler.
What I would like to do is have access to a sender/publisher and the DI container can give me the correct implementation. The code does not need any concept of the caller and if it was called from a message handler or from a self hosted application that just wants to publish.
I see that there is two interfaces for communicating with the "Bus" IPipelineContext and IMessageSession which IMessageHandlerContext and IEndpointInstance interfaces extend respectively.
What I am wondering is there some unification of the two interfaces that gets bound by NSB into the container so I can accept an interface that sends/publishes messages. In a handler it is an IMessageHandlerContext and on my self hosted application the IEndPointInstance.
For now I am looking to change my implementation of IEventPublisher depending on application hosting. I was just hoping there might be some discussion about how this approach is modeled without a reliable interface to send/publish irrespective of what initiated the execution of the code path.
A few things to note before I get to the code:
The abstraction over abstraction promise, never works. I have never seen the argument of "I'm going to abstract ESB/Messaging/Database/ORM so that I can swap it in future" work. ever.
When you abstract message sending functionality like that, you'll lose some of the features the library provides. In this case, you can't perform 'Conversations' or use 'Sagas' which would hinder your overall experience, e.g. when using monitoring tools and watching diagrams in ServiceInsight, you won't see the whole picture but only nugets of messages passing through the system.
Now in order to make that work, you need to register IEndpointInstance in your container when your endpoint starts up. Then that interface can be used in your dependency injection e.g. in NsbEventPublisher to send the messages.
Something like this (depending which IoC container you're using, here I assume Autofac):
static async Task AsyncMain()
{
IEndpointInstance endpoint = null;
var builder = new ContainerBuilder();
builder.Register(x => endpoint)
.As<IEndpointInstance>()
.SingleInstance();
//Endpoint configuration goes here...
endpoint = await Endpoint.Start(busConfiguration)
.ConfigureAwait(false);
}
The issues with using IEndpointInstance / IMessageSession are mentioned here.
Just getting my head around message queues and Redis MQ, excellent framework.
I understand that you have to use .RegisterHandler(...) to determine which handler will process the type of message/event that is in the message queue.
So if I have EventA, EventB etc should I have one Service which handles each of those Events, like :
public class DomainService : Service {
public object Any(EventA eventA) {...}
public object Any(EventB eventA) {...}
}
So these should be only queue/redis list created?
Also, what If I want a chain of events to happen, so for example a message of type EventA also has a handler that sends an Email providing handlers earlier on the chain are successful?
ServiceStack has no distinction between services created for MQ's, REST, HTML or SOAP services, they're the same thing. i.e. they each accept a Request DTO and optionally return a Response DTO and the same service can handle calls from any endpoint or format, e.g HTML, REST, SOAP or MQ.
Refer to ServiceStack's Architecture diagram to see how MQ fits in.
Limitations
The only things you need to keep in mind are:
Like SOAP, MQ's only support 1 Verb so your methods need to be named Post or Any
Only Action Filters are executed (i.e. not Global or Attribute filters)
You get MqRequest and MqResponse stubs in place of IHttpRequest, IHttpResponse. You can still use .Items to pass data through the request pipeline but any HTTP actions like setting cookies or HTTP Headers are benign
Configuring a Redis MQ Host
The MQ Host itself is completely decoupled from the rest of the ServiceStack framework, who doesn't know the MQ exists until you pass the message into ServiceStack yourself, which is commonly done inside your registered handler, e.g:
var redisFactory = new PooledRedisClientManager("localhost:6379");
var mqHost = new RedisMqServer(redisFactory, retryCount:2);
mqHost.RegisterHandler<Hello>(m => {
return this.ServiceController.ExecuteMessage(m);
});
//shorter version:
//mqHost.RegisterHandler<Hello>(ServiceController.ExecuteMessage);
mqHost.Start(); //Starts listening for messages
In your RegisterHandler<T> you specify the type of Request you want it to listen for.
By default you can only Register a single handler for each message and in ServiceStack a Request is tied to a known Service implementation, in the case of MQ's it's looking for a method signature first matching: Post(Hello) and if that doesn't exist it looks for the fallback Any(Hello).
Can add multiple handlers per message yourself
If you want to invoke multiple handlers then you would just maintain your own List<Handler> and just go through and execute them all when a request comes in.
Calling different services
If you want to call a different service, just translate it to a different Request DTO and pass that to the ServiceController instead.
When a MQ Request is sent by anyone, e.g:
mqClient.Publish(new Hello { Name = "Client" });
Your handler is invoked with an instance of type IMessage where the Request DTO is contained in the Body property. At that point you can choose to discard the message, validate it or alter it.
MQ Requests are the same as any other Service requests
In most cases you would typically just forward the message on to the ServiceController to process, the implementation of which is:
public object ExecuteMessage<T>(IMessage<T> mqMsg)
{
return Execute(mqMsg.Body, new MqRequestContext(this.Resolver, mqMsg));
}
The implementation just extracts the Request DTO from the mqMsg.Body and processes that message as a normal service being passed a C# Request DTO from that point on, with a MqRequestContext that contains the MQ IHttpRequest, IHttpResponse stubs.
I need to extract several header values at the start of each request and place them into a ClientContext object that can be injected into my application code by MEF. I am using Preview 5 of the WCF Web API and don't see a way to do this.
In 'standard' WCF, I would create a class that implements IExtension<OperationContext> and have the following property to wire it all together:
[Export(typeof(IClientContext)]
[PartCreationPolicy(CreationPolicy.NonShared)]
public static ClientContextExtension Current
{
get
{
var operationContext = OperationContext.Current;
if (operationContext == null)
return null;
var extension = operationContext.Extensions.Find<ClientContextExtension>();
if (extension == null)
{
extension = new ClientContextExtension();
operationContext.Extensions.Add(extension);
}
return extension;
}
}
A custom DelegatingHandler calls ClientContextExtension.Current and sets the properties from the header values. Unfortunately, with WCF Web API, OperationContext.Current is always null!
I cannot figure out a way to make this work with the Web API. Any help is appreciated!!!
I've come up with a working solution but remain open to other options. First, some rationale behind the original approach...
Because WCF uses thread pooling, anything based on a per-thread model may (and will) have a lifetime that extends beyond an individual request. I needed a way to store client context information pulled from the HTTP headers for each request as the information will be different each time. This means I can't persist the context information per-thread because the thread will be re-used.
Or can I?
The flaw in my logic was that thread re-use was the problem. In reality, each thread is only every servicing a single request at one time thereby making any information in that thread isolated to that request. Therefore, all I need to do is make sure that the information is relavent to that request and my problem is solved.
My solution was to refactor the Current property to reference a private static field marked with the [ThreadStatic()] attribute, ensuring that each instance was specific to the thread. Then, in my DelegatingHandler, which executes for each request, I reset the properties of the object for that request. Subsequent calls to Current during that request return the request-specific information and the next request handled by the thread gets updated in the DelegatingHandler so as far as my other code is concerned, the context is per-request.
Not perfect, but it at least gets me up and running for the moment. As I said, I am open to other solutions.
UPDATE
Upon closer inspection, this solution is not working as there is no thread affinity between the DelegatingHandler and the service code that is making use of the context object. As a result, sometimes my call to retrieve the ThreadStatic object works as expected but on other occasions I get a new instance because the code is operating on a different thread than the handler.
So, disregard this solution. Back to the drawing board.
UPDATE TO MY UPDATE
After discussing my problem with Glenn Block, it turns out that it is just a matter of making sure the context is set on the same thread the request handler (the service) is executing. The solution is to use an HttpOperationHandler instead of a MessageHandler.
According to Glenn, message handlers operate asynchronously which means they could execute on a different thread from the request handler (service) so we should never do anything in a message handler that requires thread affinity. On the other hand, operation handlers run synchronously on the same thread as the request handler, therefore we can rely on thread affinity.
So, I simply moved my code from a MessageHandler to an HttpOperationHandler and have the desired results.
You can read a full explanation here: http://sonofpirate.blogspot.com/2011/11/modeling-client-context-in-wcf-web-api.html
You can try to use a
HttpOperationHandler<HttpRequestMessage, HttpRequestMessage>
There you should be able to access the headers.
I'm making a WCF service with netTcpBinding which has a main lobby with multiple chatrooms which the clients can enter. The Lobby class implements ILobby as the service contract.
When a client wishes to enter a room I want to callback the client exposing a new Channel containing the InstanceContext for the room he just entered but after much searching I am doubting that this is possible.
For example on the Service side I might have
class Lobby : ILobby
{
Dictionary<string, Chatroom> rooms;
public void JoinRoom(string roomname)
{
if (rooms[roomname].TryEnter()) {}
}
}
class ChatRoom : IChatRoom
{
public bool TryEnter(string username)
{
ILobbyCallback callback =
OperationContext.Current.GetCallbackChannel<ILobbyCallback>();
// How do I do this next bit?
callback.JoinedRoom(pass some instance context here);
return true;
}
}
On the client side callback method I want
public void JoinedRoom(InstanceContext for the room on the service side)
{
// Create a new WCF proxy using above InstanceContext
// Create a WPF UI for the new room passing the proxy so it can communicate
// with the room class directly without going via the root service
}
Is this possible? What's the best practice for spawning new classes with their own contracts on the service side? Or do I just have to bundle everything into one massive MyService class and handle everything myself?
You cannot pass instance context as parameter to any operation contract. It doesn't make sense because that context has local scope. It is called "instance context" = it is context of current service instance. In duplex scenario both client and server has its own service:
Clients calls server's service through its proxy
Server calls client' service through received callback channel
Server's service instance context has meaning only on the server. It is not clear what you are trying to achieve (except very complex architecture).
If you want to share context on client you can try to pass around the instance context used for the very first proxy you created - I'm not sure if it will work but you can try it
If you want to share service instance context between multiple proxies you must develop your own IInstanceContextProvider and perhaps also your own IInstanceProvider (depending on what you want to achieve), wrap them in behavior and add them to the service. That will put whole complexity of session handling and correct instance releasing under your control (it obviously has its pros and cons).
But is it really needed? When I look at your code I see that one service and one proxy is enough. Also your JoinRoom operation doesn't need to use callback at all, it can be just request response method.
I have a Windows Service Application
in which i create WCF services in it.
One of the services is data
services: add, delete,
read , updatte data via
WCF.
WCF use NHibernate for data manipulation
So my guestions are:
Any advice (best practice) for session management for Hibernate using with WCF?
Anybody knows anything about
WcfOperationSessionContext (hibernate 3.0) class?
how to use it with WCF?
Well to make it concrete :
Suppose that i have WCF Service called DataServices
class WCFDataService .....
{
void SaveMyEntity(MyEntity entity)
{
.....................?? // How to do? Best Way
// Should i take one session and use it all times
// Should i take session and dipsose when operation finished then get
//new session for new operations?
// If many clients call my WCF service function at the same time?
// what may go wrong?
// etc....
}
}
And I need a NHibernateServiceProvider class
class NHibernateServiceProvider ....
{
// How to get Session ?? Best way
ISession GetCurrentSession(){.... }
DisposeSession(){ ....}
}
Best Wishes
PS: I have read similiar entries here and other web pages. But can not see "concrete" answers.
The WcfOperationSessionContext, similar to ThreadStaticSessionContext and WebRequestSessionContext is an implementation for a session context. The session context is used to bind (associate) a ISession instance to a particular context.
The session in the current context can be retrieved by calling ISessionFactory.GetCurrentSession().
You can find more information about session context here.
The WcfOperationSessionContext represents a context that spans for the entire duration of a WCF operation. You still need to handle the binding of the session in the begining of the operation and the unbinding/commiting/disposal of the session at the end of the operation.
To get access to the begin/end actions in the wcf pipeline you need to implement a IDispatchMessageInspector. You can see a sample here.
Also regarding WCF integration: if you use ThreadStatic session context it will appear to work on development, but you will hit the wall in production when various components (ex: authorization, authentication ) from the wcf pipeline are executed on different threads.
As for best practices you almost nailed it: Use WcfOperationSessionContext to store the current session and the IDispatchMessageInspector to begin/complete your unit of work.
EDIT - to address the details you added:
If you configured WcfOperationSessionContext and do the binding/unbinding as i explained above, all you have to do to is inject the ISessionFactory into your service and just use factory.GetCurrentSession(). I'll post a sample prj if time permits.
Here is the sample project
The model we use for managing NHibernate sessions with WCF is as follows:
1) We have our own ServiceHost class that inherits from System.ServiceModel.ServiceHost which also implements ICallContextInitializer. We add the service host instance to each of the operations in our service as follows:
protected override void InitializeRuntime()
{
base.InitializeRuntime();
foreach (ChannelDispatcher cd in this.ChannelDispatchers)
{
foreach (EndpointDispatcher ed in cd.Endpoints)
{
foreach (DispatchOperation op in ed.DispatchRuntime.Operations)
{
op.CallContextInitializers.Add(this);
}
}
}
}
public void AfterInvoke(object correlationState)
{
// We don't do anything after the invoke
}
public object BeforeInvoke(InstanceContext instanceContext, IClientChannel channel, Message message)
{
OperationContext.Current.Extensions.Add(new SessionOperationContext());
return null;
}
The BeforeInvoke simply makes sure that the OperationContext for each WCF call has it's own session. We have found problems with IDispatchMessageInspector where the session is not available during response serialisation - a problem if you use lazy loading.
2) Our SessionOperationContext will then be called to attach itself and we use the OperationCompleted event to remove ourselves. This way we can be sure the session will be available for response serialisation.
public class SessionOperationContext : IExtension<OperationContext>
{
public ISession Session { get; private set; }
public static SessionOperationContext Current
{
get
{
OperationContext oc = OperationContext.Current;
if (oc == null) throw new InvalidOperationException("Must be in an operation context.");
return oc.Extensions.Find<SessionOperationContext>();
}
}
public void Attach(OperationContext owner)
{
// Create the session and do anything else you required
this.Session = ... // Whatever instantiation method you use
// Hook into the OperationCompleted event which will be raised
// after the operation has completed and the response serialised.
owner.OperationCompleted += new EventHandler(OperationCompleted);
}
void OperationCompleted(object sender, EventArgs e)
{
// Tell WCF this extension is done
((OperationContext)sender).Extensions.Remove(this);
}
public void Detach(OperationContext owner)
{
// Close our session, do any cleanup, even auto commit
// transactions if required.
this.Session.Dispose();
this.Session = null;
}
}
We've used the above pattern successfully in high-load applications and it seems to work well.
In summary this is similar to what the new WcfOperationSessionContext does (it wasn't around when we figured out the pattern above;-)) but also overcomes issues surrounding lazy loading.
Regarding the additional questions asked: If you use the model outlined above you would simply do the following:
void SaveMyEntity(MyEntity entity)
{
SessionOperationContext.Current.Session.Save(entity);
}
You are guaranteed that the session is always there and that it will be disposed once the WCF operation is completed. You can use transactions if required in the normal way.
Here is a post describing, in detail, all the steps for registering and using the WcfOperationSessionContext. It also includes instructions for using it with the agatha-rrsl project.
Ok, after few days of reading internet posts etc. all approaches shown in the internets seems to be wrong. When we are using UnitOfWork pattern with NH 3^ with nhibernate transaction this all aprochaes are producing exceptions. To test it and proof that we need to create test enviroment with MSMQ transaction queue, special interface with OneWay operation contract with transaction required set on it. This approach should works like this:
1. We put transactionally message in queue.
2. Service is getting transactionally messege from queue.
3. Everything works queue is empty.
In some cases not so obious with internet approaches this does not work properly. So here are expamples which we tested that are wrong and why:
Fabio Maulo approach: Use ICallContextInitializer - open NH session/transaction on BeforeCall, after that WCF is executing service method, on AfterCall in context initializer we call session.Flush + transaction.commit. Automaticly session will be saved when transaction scope will commit operation. In situation when on calling transaction.Complete exception will be thrown WCF service will shutdown! Question can be ok, so take transaction.Complete in try/catch clausule - great! - NO wrong! Then transaction scope will commit transaction and message will be taken from queue but data will not be saved !
Another approach is to use IDispatchMessageInspector - yesterday I thought this is best approach. Here we need to open session/transaction in method AfterReceiveRequest, after WCF invoke service operation on message dispatcher inspector BeforeSendReply is called. In this method we have info about [reply] which in OneWay operation is null, but filled with fault information if it occured on invoking service method. Great I thought - this is this ! but NOT! Problem is that at this point in WCF processing pipe we have no transaction ! So if transaction.Complete throw error or session.Flush will throw it we will have not data saved in database and message will not come back to queue what is wrong.
What is the solution?
IOperationInvoker and only this!
You need to implement this interface as a decorator pattern on default invoker. In method Invoke before call we are openning session/transaction open then we call invoke default invoker and after that call transaction.complete in finally clausule we call session.flush. What types of problem this solves:
1. We have transaction scope on this level so when complete throws exception message will go back to queue and WCF will not shutdown.
2. When invocation will throw exception transaction.complete will not be called what will not change database state
I hope this will clear everyones missinformation.
In some free time I will try to write some example.