Please note: this article has been superceded by the documentation for the ChannelAdam WCF Library.
Background
In my previous article, How To Call WCF Services Properly, I researched and highlighted the typical problems with using WCF clients (such as memory and connection leaks), and concluded that the most correct solution for consuming WCF services would:
- Perform the Close/Abort pattern without a race condition;
- Handle the situation when the service operation throws exceptions;
- Handle the situations when both the
Close
andAbort
methods throw exceptions; and - Handle asynchronous exceptions such as the
ThreadAbortException
.
And below was my sample code for correctly using a WCF client, performing exception handling and correctly performing the Close/Abort pattern.
SampleServiceClient client = null;
try
{
client = new SampleServiceClient();
var response = client.SampleOperation(1234);
// Do some business logic
}
catch (FaultException<MyCustomException>)
{
// Do some business logic for this SOAP Fault Exception
}
catch (FaultException)
{
// Do some business logic for this SOAP Fault Exception
}
catch (CommunicationException)
{
// Catch this expected exception so it is not propagated further.
// Perhaps write this exception out to log file for gathering statistics...
}
catch (TimeoutException)
{
// Catch this expected exception so it is not propagated further.
// Perhaps write this exception out to log file for gathering statistics...
}
catch (Exception)
{
// An unexpected exception that we don't know how to handle.
// Perhaps write this exception out to log file for support purposes...
throw;
}
finally
{
// This will:
// - be executed if any exception was thrown above in the 'try' (including ThreadAbortException); and
// - ensure that CloseOrAbortServiceChannel() itself will not be interrupted by a ThreadAbortException
// (since it is executing from within a 'finally' block)
CloseOrAbortServiceChannel(client);
// Unreference the client
client = null;
}
private void CloseOrAbortServiceChannel(ICommunicationObject communicationObject)
{
bool isClosed = false;
if (communicationObject == null || communicationObject.State == CommunicationState.Closed)
{
return;
}
try
{
if (communicationObject.State != CommunicationState.Faulted)
{
communicationObject.Close();
isClosed = true;
}
}
catch (CommunicationException)
{
// Catch this expected exception so it is not propagated further.
// Perhaps write this exception out to log file for gathering statistics...
}
catch (TimeoutException)
{
// Catch this expected exception so it is not propagated further.
// Perhaps write this exception out to log file for gathering statistics...
}
catch (Exception)
{
// An unexpected exception that we don't know how to handle.
// Perhaps write this exception out to log file for support purposes...
throw;
}
finally
{
// If State was Faulted or any exception occurred while doing the Close(), then do an Abort()
if (!isClosed)
{
AbortServiceChannel(communicationObject);
}
}
}
private static void AbortServiceChannel(ICommunicationObject communicationObject)
{
try
{
communicationObject.Abort();
}
catch (Exception)
{
// An unexpected exception that we don't know how to handle.
// If we are in this situation:
// - we should NOT retry the Abort() because it has already failed and there is nothing to suggest it could be successful next time
// - the abort may have partially succeeded
// - the actual service call may have been successful
//
// The only thing we can do is hope that the channel's resources have been released.
// Do not rethrow this exception because the actual service operation call might have succeeded
// and an exception closing the channel should not stop the client doing whatever it does next.
//
// Perhaps write this exception out to log file for gathering statistics and support purposes...
}
}
Unfortunately, due to the way Microsoft designed WCF, if you want to use a WCF service client cleanly without connection or memory leaks, then above is the type of extensive code that must be written for every service operation call.
I then promised to share an even better way to achieve this - by using some programming tricks to provide a nice, clean API for working with WCF clients, that significantly reduces the amount of that code that developers have to write.
What Is The Easy, Clean and Robust Solution?
Well, here it is! Drum roll please...
After more than a month of developing late into the evening and early morning, allow me to introduce the ChannelAdam WCF Library - an open source (Apache License 2.0) library that allows you to forget about the complex subtleties of WCF clients, channels and communication states, and lets you just use your service client (very similar to how you probably are doing it now - but without the leaks!).
When you use the ChannelAdam WCF Library's ServiceConsumer
, you know that you can keep calling service operations through the ServiceConsumer
and it will 'just work' and not leak connections or memory - no matter what happens!
The ChannelAdam WCF Library's ServiceConsumer
hides the inner complexity of service channels. For example, if a service channel / communication object / service client moves into a Fault state, the ServiceConsumer
will correctly perform the Close/Abort pattern and dispose of the channel. The next time a request is made through the ServiceConsumer
, a new channel is created under the covers to seamlessly process the request.
Where Do I Get The Library?
You can install the ChannelAdam.Wcf NuGet package from within Visual Studio, or find it on the NuGet.org gallery at
https://www.nuget.org/packages/ChannelAdam.Wcf.
If you are keen to see some of the programming tricks and how it
works, or possibly contribute features to it, the source code for the
ChannelAdam WCF Library is on CodePlex at https://channeladamwcf.codeplex.com.
Show Me The Basics! How Do I Use It?
Let's assume for the duration of this article that there is a web service named FakeService
and in Visual Studio we have generated a service reference and interface to FakeService
, and there is now an IFakeService
interface and a FakeServiceClient
class to use.
That is all normal and basic Visual Studio functionality.
Let's also assume that the service has the following operation signature:
int AddIntegers(int first, int second);
Create a ServiceConsumer with the ServiceConsumerFactory
The first thing you need to understand is that you consume your service operations through the usage of the ChannelAdam.ServiceModel.ServiceConsumer
class.
The second thing you need to understand is that the best way to create the ServiceConsumer
is by using the Create
method on the ChannelAdam.ServiceModel.ServiceConsumerFactory
.
In order to use the ServiceConsumerFactory.Create
method, simply pass in as a parameter a factory method for creating the service client.
For example:
IServiceConsumer<IFakeService> service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient()));
The ServiceConsumerFactory.Create
method returns a IServiceConsumer<TServiceInterface>
type. In all further examples, I will use the var
keyword instead - such as:
var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient()));
Alternatively, you can pass in as a string parameter the endpoint configuration name from the service client section of your application's .config file. For example:
var service = ServiceConsumerFactory.Create<IFakeService>("BasicHttpBinding_IFakeService"));
Four Ways to Clean Up the ServiceConsumer
There are four ways in which the ServiceConsumer
can be cleaned up:
- Explicitly with the Close method;
- Explicitly with the Dispose method;
- Implicitly via the Using statement; or
- Automatically by the Garbage Collector.
The Close Method
The ServiceConsumer
and underlying channel can be closed immediately by calling the Close
method on the ServiceConsumer
instance.
service.Close();
If you want to, you still can immediately call another service operation via that same ServiceConsumer
instance, as under the covers, a new service channel will be created seamlessly and automatically.
The Dispose Method
The ServiceConsumer
and underlying channel can be disposed of immediately by calling the Dispose
method on the ServiceConsumer
instance.
service.Dispose();
As per standard IDisposable
concepts, you should not attempt to reuse the ServiceConsumer
after disposing it.
The Using Statement
Instead of calling the Close
or Dispose
methods directly, you can use the Using
statement on the ServiceConsumer
because it is a well behaved implementation of IDisposable
(unlike Microsoft's service clients!).
using (var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient())))
{
...
}
Garbage Collection
If you happen to accidently forget to wrap your ServiceConsumer
instance in a Using
statement or call the Close
method, rest assured that the ServiceConsumer
will still be cleaned up! When the ServiceConsumer
instance goes out of scope, the .NET Garbage Collector will [eventually] kick in and call its destructor/finalise method, thus correctly cleaning up the service channel and preventing any leaks.
Since there is an indeterminate amount of time before the Garbage Collector kicks in, I would recommend the Using
statement approach, as it minimises the number of open channels and other resources currently in use, over the lifetime of your application.
How To Call a Service Operation
There are two ways in which to call a service operation via the ServiceConsumer
:
- Via the
Operations
property; or - Via the
Consume
method.
The Operations Property
The Operations
property exposes the interface of your service. It can be used just like the normal service client that you already would be familiar with.
Below is an example of how to call the AddIntegers
operation via the Operations
property.
using (var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient())))
{
try
{
int actualValue = service.Operations.AddIntegers(1, 1);
Assert.AreEqual(2, actualValue);
}
...
}
And below is an example of how to do some exception handling.
using (var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient())))
{
try
{
int actualValue = service.Operations.AddIntegers(1, 1);
Assert.AreEqual(2, actualValue);
}
catch (FaultException<MyCustomException> fe)
{
Console.WriteLine("Service operation threw a custom fault: " + fe.ToString());
}
catch (FaultException fe)
{
Console.WriteLine("Service operation threw a fault: " + fe.ToString());
}
catch (Exception ex)
{
Console.WriteLine("An unexpected error occurred while calling the service operation: " + ex.ToString());
}
}
You may notice that there is now no need to catch a CommunicationException
or TimeOutException
and try to perform the Close/Abort pattern, as the underlying channel is dealt with automatically for you by the ServiceConsumer
.
The Consume Method
As an alternative to the Operations
property, you can call a service operation via the Consume
method on the ServiceConsumer
instance.
The Consume
method returns an instance of IOperationResult<TReturnTypeOfYourOperation>
which wraps the result of the operation. This interface has the following helpful properties:
-
TReturnTypeOfYourOperation Value
- which gives you access to the return value from the operation; -
bool HasException
- whether there was an exception thrown by the service operation; -
bool HasNoException
- whether there was not an exception thrown by the service operation; -
bool HasFaultException
- whether an exception of typeFaultException
was thrown by the service operation; -
bool HasFaultException<OfThisType>
- whether an exception ofFaultException<OfThisType>
was thrown by the service operation; and -
Exception Exception
- the exception that was thrown by the service operation.
Below is an example of how to call the AddIntegers
operation via the Consume
method.
using (var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient())))
{
IOperationResult<int> result = service.Consume(operation => operation.AddIntegers(1, 1));
if (result.HasNoException)
{
Assert.AreEqual(2, result.Value);
}
...
}
And below is an example of how to do some exception handling with IOperationResult
.
using (var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient())))
{
var result = service.Consume(operation => operation.AddIntegers(1, 1));
if (result.HasNoException)
{
int actualValue = result.Value;
Assert.AreEqual(2, actualValue);
}
else
{
if (result.HasFaultExceptionOfType<MyCustomException>())
{
Console.WriteLine("Service operation threw a custom fault: " + result.Exception.ToString());
}
else if (result.HasFaultException)
{
Console.WriteLine("Service operation threw a fault: " + result.Exception.ToString());
}
else
{
Console.WriteLine("An unexpected error occurred while calling the service operation: " + result.Exception.ToString());
}
}
}
More Advanced Usage - With IoC Containers
It is easy to use the ServiceConsumerFactory
from Inversion of Control (IoC) containers. Below are examples with three popular IoC containers.
Let's assume that there is the following class and we want to perform constructor injection with it.
public class ConstructorInjectedController
{
public IServiceConsumer<IFakeService> FakeService { get; set; }
/// <summary>
/// Initializes a new instance of the <see cref="ConstructorInjectedController"/> class.
/// </summary>
/// <param name="fakeService">The fake service.</param>
public ConstructorInjectedController(IServiceConsumer<IFakeService> fakeService)
{
this.FakeService = fakeService;
}
}
AutoFac
Here is how to use the ServiceConsumerFactory
with AutoFac.
var builder = new Autofac.ContainerBuilder();
builder.Register(c => ServiceConsumerFactory.Create<IFakeService>("BasicHttpBinding_IFakeService")).InstancePerDependency();
builder.RegisterType<ConstructorInjectedController>();
var autofacContainer = builder.Build();
var controller = autofacContainer.Resolve<ConstructorInjectedController>();
SimpleInjector
Here is how to use the ServiceConsumerFactory
with SimpleInjector.
var simpleInjectorContainer = new SimpleInjector.Container();
simpleInjectorContainer.Register(() => ServiceConsumerFactory.Create<IFakeService>("BasicHttpBinding_IFakeService"), SimpleInjector.Lifestyle.Transient);
simpleInjectorContainer.Verify();
var controller = simpleInjectorContainer.GetInstance<ConstructorInjectedController>();
Unity
And, here is how to use the ServiceConsumerFactory
with Unity.
var unityContainer = new UnityContainer();
unityContainer
.RegisterType<IServiceConsumer<IFakeService>>(
new TransientLifetimeManager(),
new InjectionFactory(c => ServiceConsumerFactory.Create<IFakeService>("BasicHttpBinding_IFakeService")))
//
// OR
//
//.RegisterType<IServiceConsumer<IFakeService>>(
// new InjectionFactory(c => ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient())))
//
// OR more 'correctly'
//
//.RegisterType<IFakeService, FakeServiceClient>(new TransientLifetimeManager())
//.RegisterType<IServiceConsumer<IFakeService>>(
// new InjectionFactory(c =>
// ServiceConsumerFactory.Create<IFakeService>(() => (ICommunicationObject)c.Resolve<IFakeService>())))
;
var controller = unityContainer.Resolve<ConstructorInjectedController>();
But Wait, There's More - Channel Close Trigger Strategies
When an exception occurs while calling the service operation, the ServiceConsumer
follows a specific strategy to determine whether the exception and/or current channel state is severe enough to necessitate and trigger the closing/aborting of the service channel.
These "Service Channel Close Trigger Strategies" are an implementation of the interface IServiceChannelCloseTriggerStrategy
. There is one method in this interface:
bool ShouldCloseChannel(ICommunicationObject channel, Exception exception);
If the method ShouldCloseChannel
returns true
, then the ServiceConsumer
will perform the Close/Abort pattern.
Out of the box, the DefaultServiceChannelCloseTriggerStrategy
is used. This strategy is very simple, effective and safe. It triggers a close to occur if the exception is anything but a FaultException
, regardless of the state of the service channel. It is highly probable that the channel is no longer usable in the case where there was any other type of exception.
If you find that this strategy does not suit your need, you can create your own class that implements that interface and specify your class as one of the overloads of the Create
method on the ServiceConsumerFactory
. Alternatively, you can directly set the property ChannelCloseTriggerStrategy
on the ServiceConsumer
instance itself.
But Wait, There's More - Exception Behaviour Strategies
As you know, the ServiceConsumer
automatically handles the clean up of service channels. If an exception occurs at any point throughout the usage of the channel, the ServiceConsumer
will catch it and close the channel if necessary. In some cases, for example, if an exception occurs while attempting to close the channel, the ServiceConsumer
does not want to propagate the exception back to the calling code, as it is not the caller's concern. In order to be a good citizen and prevent the ServiceConsumer
from swallowing exceptions, the concept of the "Exception Behaviour Strategy" was born.
Whenever an exception occurs throughout the lifetime of the ServiceConsumer
, a corresponding method on the configured Exception Behaviour Strategy class is invoked, thus allowing the exception to be logged and statistics recorded for an organisation's support purposes.
Each Exception Behaviour Strategy class implements the interface IServiceConsumerExceptionBehaviourStrategy
. This interface contains the following methods:
-
PerformFaultExceptionBehaviour
- the behaviour to perform when a fault exception occurs while the service operation is called; -
PerformCommunicationExceptionBehaviour
- the behaviour to perform when a communication exception occurs while the service operation is called; -
PerformTimeoutExceptionBehaviour
- the behaviour to perform when a timeout exception occurs while the service operation is called; -
PerformUnexpectedExceptionBehaviour
- the behaviour to perform when an unexpected exception occurs while the service operation is called; -
PerformCloseCommunicationExceptionBehaviour
- the behaviour to perform when a communication exception occurs during a close; -
PerformCloseTimeoutExceptionBehaviour
- the behaviour to perform when a timeout exception occurs during a close; -
PerformCloseUnexpectedExceptionBehaviour
- the behaviour to perform when an unexpected exception occurs during a close; -
PerformAbortExceptionBehaviour
- the behaviour to perform when an exception occurs during an abort; and -
PerformDestructorExceptionBehaviour
- the behaviour to perform when an exception occurs during a destructor/Finalize
method.
The ChannelAdam.Wcf Library comes with the following Exception Behaviour Strategies:
-
NullServiceConsumerExceptionBehaviourStrategy
- which does not write out any exception details anywhere; -
StandardOutServiceConsumerExceptionBehaviourStrategy
- which writes out all exceptions to the Standard Error stream; and -
StandardErrorServiceConsumerExceptionBehaviourStrategy
- which writes out all exceptions to the Standard Out stream.
By default, the ServiceConsumerFactory
configures each ServiceConsumer
with the StandardErrorServiceConsumerExceptionBehaviourStrategy
.
There are three ways to change the Exception Behaviour Strategy that is assigned when you create a ServiceConsumer
:
- Specify a different Exception Behaviour Strategy on one of the overloads of the
Create
method on theServiceConsumerFactory
; - Directly set the property
ChannelCloseTriggerStrategy
on theServiceConsumer
instance itself; or - Change the default - in some bootstrap code, set the static property
ServiceConsumerFactory.DefaultExceptionBehaviourStrategy
to your desired strategy. From then on, that strategy will be used by default.
As I expect will be the case, many organisations will want to use a logging library and write out the exceptions to a log file somewhere. Easy - just implement the interface and set it as the default in some bootstrap code.
But Wait, There's More - Retry Capability for Handling Transient Faults
When you call your web services, do you currently handle transient errors that may occur (such as the network briefly dropping out), and have some automatic retry logic?
Well, the ChannelAdam WCF Library has some built-in support for the Microsoft Practices Transient Fault Handling Library - but only when you use the Consume
method and not when you use the Operations
property.
Naive Retry Logic with the Operations Property
Below is an example naive implementation of retry logic (without the help of any library) while using the Operations
property.
int retryCount = 1;
Exception lastException = null;
using (var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient()))
{
while (retryCount > 0)
{
Console.WriteLine("#### Retry count: " + retryCount);
try
{
int actual = service.Operations.AddIntegers(1, 1);
Console.WriteLine("Actual: " + actual);
Assert.AreEqual(2, actual);
return;
}
catch (FaultException fe)
{
lastException = fe;
Console.WriteLine("Service operation threw a fault: " + fe.ToString());
}
catch (Exception ex)
{
lastException = ex;
Console.WriteLine("Technical error occurred while calling the service operation: " + ex.ToString());
}
retryCount--;
}
}
Assert.Fail("Service operation was not successfully called");
Unfortunately that code is lengthy and error prone.
Retry Logic with Operations Property and the Microsoft Library
The code above can be improved upon by using the Microsoft Practices Transient Fault Handling Library, as below.
You will notice in this example, I am using the ChannelAdam.ServiceModel.SoapFaultWebServiceTransientErrorDetectionStrategy
to determine if there was a transient error or not. If the given exception was a FaultException
, this strategy says that was not a transient error and therefore do not retry. If on the other hand any other type of exception occurs, then this strategy will tell the Microsoft library to retry according to the retry policy.
Exception lastException;
using (var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient()))
{
try
{
int actual = 0;
var retryStrategy = new Incremental(5, TimeSpan.FromSeconds(2), TimeSpan.FromSeconds(2));
var retryPolicy = new RetryPolicy<SoapFaultWebServiceTransientErrorDetectionStrategy>(retryStrategy);
retryPolicy.ExecuteAction(() =>
{
actual = service.Operations.AddIntegers(1, 1);
});
Console.WriteLine("Actual: " + actual);
Assert.AreEqual(2, actual);
return;
}
catch (FaultException fe)
{
lastException = fe;
Console.WriteLine("Service operation threw a fault: " + fe.ToString());
}
catch (Exception ex)
{
lastException = ex;
Console.WriteLine("Technical error occurred while calling the service operation: " + ex.ToString());
}
Assert.Fail("Service operation was not successfully called");
Retry Logic with the Consume Method
For me, both code examples above that use the Operations
property is clunky - which is why I prefer using the Consume
method!
When using the Consume
method, there are three ways to set the retry policy:
- Setting the static property
DefaultRetryPolicy
on theServiceConsumerFactory
- which will apply to all created instances; - As an overload to the
Create
method on theServiceConsumerFactory
- which will apply just to thatServiceConsumer
instance to be created; or - As one of the overloads on the
Consume
method itself.
For example, the overload on the Consume
method:
using (var service = ServiceConsumerFactory.Create<IFakeService>(() => new FakeServiceClient()))
{
var retryStrategy = new Incremental(5, TimeSpan.FromSeconds(2), TimeSpan.FromSeconds(2));
var retryPolicy = new RetryPolicy<SoapFaultWebServiceTransientErrorDetectionStrategy>(retryStrategy);
var result = service.Consume(operation => operation.AddIntegers(1, 1), retryPolicy);
if (result.HasNoException)
{
Console.WriteLine("Actual: " + result.Value);
Assert.AreEqual(2, result.Value);
}
else
{
if (result.HasFaultException)
{
Console.WriteLine("Service operation threw a fault: " + result.Exception.ToString());
}
else if (result.HasException)
{
Console.WriteLine("An unexpected error occurred while calling the service operation: " + result.Exception.ToString());
}
Assert.Fail("Service operation was not successfully called");
}
}
Perhaps you too may agree that using the retry strategy with the Consume
method is simpler and cleaner, in comparison with the Operations
property.
What Else Does the ServiceConsumer Do?
If you want to know at a deeper level how the ServiceConsumer
and its friends behave, perhaps you would like to read the Behaviour Specifications feature file.
Conclusion
The ChannelAdam WCF Library is designed to make your life as a developer easier when consuming WCF services, by abstracting away the complexities of service channel state transitions and the odd behaviour of the Microsoft WCF libraries that do not comply with even their own design guidelines.
In doing this, the ChannelAdam ServiceConsumer
once and for all prevents memory and connection leaks, gives developers the ability to plug in their own strategies for handling different situations (such as logging exceptions), and provides a built-in mechanism to retry service operation calls when transient faults occur.
I hope you like it, and I welcome your feedback!