The preamble:
I have designed a strongly interfaced and fully mockable data layer class that expects the business layer to create a TransactionScope
when multiple calls should be included in a single transaction.
The problem: I would like to unit test that my business layer makes use of a TransactionScope
object when I expect it to.
Unfortunately, the standard pattern for using TransactionScope
is a follows:
using(var scope = new TransactionScope())
{
// transactional methods
datalayer.InsertFoo();
datalayer.InsertBar();
scope.Complete();
}
While this is a really great pattern in terms of usability for the programmer, testing that it's done seems... unpossible to me. I cannot detect that a transient object has been instantiated, let alone mock it to determine that a method was called on it. Yet my goal for coverage implies that I must.
The Question: How can I go about building unit tests that ensure TransactionScope
is used appropriately according to the standard pattern?
Final Thoughts: I've considered a solution that would certainly provide the coverage I need, but have rejected it as overly complex and not conforming to the standard TransactionScope
pattern. It involves adding a CreateTransactionScope
method on my data layer object that returns an instance of TransactionScope
. But because TransactionScope contains constructor logic and non-virtual methods and is therefore difficult if not impossible to mock, CreateTransactionScope
would return an instance of DataLayerTransactionScope
which would be a mockable facade into TransactionScope
.
While this might do the job it's complex and I would prefer to use the standard pattern. Is there a better way?
I'm just now sitting with the same problem and to me there seems to be two solutions:
Edit: I've created a CodePlex-project for this now: http://legendtransactions.codeplex.com/
I'm leaning towards creating a set of interfaces for working with transactions and a default implementation that delegates to the System.Transaction-implementations, something like:
public interface ITransactionManager { ITransaction CurrentTransaction { get; } ITransactionScope CreateScope(TransactionScopeOption options); } public interface ITransactionScope : IDisposable { void Complete(); } public interface ITransaction { void EnlistVolatile(IEnlistmentNotification enlistmentNotification); } public interface IEnlistment { void Done(); } public interface IPreparingEnlistment { void Prepared(); } public interface IEnlistable // The same as IEnlistmentNotification but it has // to be redefined since the Enlistment-class // has no public constructor so it's not mockable. { void Commit(IEnlistment enlistment); void Rollback(IEnlistment enlistment); void Prepare(IPreparingEnlistment enlistment); void InDoubt(IEnlistment enlistment); }
This seems like a lot of work but on the other hand it's reusable and it makes it all very easily testable.
Note that this is not the complete definition of the interfaces just enough to give you the big picture.
Edit: I just did some quick and dirty implementation as a proof of concept, I think this is the direction I will take, here's what I've come up with so far. I'm thinking that maybe I should create a CodePlex project for this so the problem can be solved once and for all. This is not the first time I've run into this.
public interface ITransactionManager { ITransaction CurrentTransaction { get; } ITransactionScope CreateScope(TransactionScopeOption options); } public class TransactionManager : ITransactionManager { public ITransaction CurrentTransaction { get { return new DefaultTransaction(Transaction.Current); } } public ITransactionScope CreateScope(TransactionScopeOption options) { return new DefaultTransactionScope(new TransactionScope()); } } public interface ITransactionScope : IDisposable { void Complete(); } public class DefaultTransactionScope : ITransactionScope { private TransactionScope scope; public DefaultTransactionScope(TransactionScope scope) { this.scope = scope; } public void Complete() { this.scope.Complete(); } public void Dispose() { this.scope.Dispose(); } } public interface ITransaction { void EnlistVolatile(Enlistable enlistmentNotification, EnlistmentOptions enlistmentOptions); } public class DefaultTransaction : ITransaction { private Transaction transaction; public DefaultTransaction(Transaction transaction) { this.transaction = transaction; } public void EnlistVolatile(Enlistable enlistmentNotification, EnlistmentOptions enlistmentOptions) { this.transaction.EnlistVolatile(enlistmentNotification, enlistmentOptions); } } public interface IEnlistment { void Done(); } public interface IPreparingEnlistment { void Prepared(); } public abstract class Enlistable : IEnlistmentNotification { public abstract void Commit(IEnlistment enlistment); public abstract void Rollback(IEnlistment enlistment); public abstract void Prepare(IPreparingEnlistment enlistment); public abstract void InDoubt(IEnlistment enlistment); void IEnlistmentNotification.Commit(Enlistment enlistment) { this.Commit(new DefaultEnlistment(enlistment)); } void IEnlistmentNotification.InDoubt(Enlistment enlistment) { this.InDoubt(new DefaultEnlistment(enlistment)); } void IEnlistmentNotification.Prepare(PreparingEnlistment preparingEnlistment) { this.Prepare(new DefaultPreparingEnlistment(preparingEnlistment)); } void IEnlistmentNotification.Rollback(Enlistment enlistment) { this.Rollback(new DefaultEnlistment(enlistment)); } private class DefaultEnlistment : IEnlistment { private Enlistment enlistment; public DefaultEnlistment(Enlistment enlistment) { this.enlistment = enlistment; } public void Done() { this.enlistment.Done(); } } private class DefaultPreparingEnlistment : DefaultEnlistment, IPreparingEnlistment { private PreparingEnlistment enlistment; public DefaultPreparingEnlistment(PreparingEnlistment enlistment) : base(enlistment) { this.enlistment = enlistment; } public void Prepared() { this.enlistment.Prepared(); } } }
Here's an example of a class that depends on the ITransactionManager to handle it's transactional work:
public class Foo { private ITransactionManager transactionManager; public Foo(ITransactionManager transactionManager) { this.transactionManager = transactionManager; } public void DoSomethingTransactional() { var command = new TransactionalCommand(); using (var scope = this.transactionManager.CreateScope(TransactionScopeOption.Required)) { this.transactionManager.CurrentTransaction.EnlistVolatile(command, EnlistmentOptions.None); command.Execute(); scope.Complete(); } } private class TransactionalCommand : Enlistable { public void Execute() { // Do some work here... } public override void Commit(IEnlistment enlistment) { enlistment.Done(); } public override void Rollback(IEnlistment enlistment) { // Do rollback work... enlistment.Done(); } public override void Prepare(IPreparingEnlistment enlistment) { enlistment.Prepared(); } public override void InDoubt(IEnlistment enlistment) { enlistment.Done(); } } }
Ignoring whether this test is a good thing or not....
Very dirty hack is to check that Transaction.Current is not null.
This is not a 100% test since someone could be using something other than TransactionScope to achieve this but it should guard against the obvious 'didn't bother to have a transaction' parts.
Another option is to deliberately try to create a new TransactionScope with incompatible isolation level to whatever would/should be in use and TransactionScopeOption.Required
. If this succeeds rather than throwing an ArgumentException there wasn't a transaction. This requires you to know that a particular IsolationLevel is unused (something like Chaos is a potential choice)
Neither of these two options is particularly pleasant, the latter is very fragile and subject to the semantics of TransactionScope remaining constant. I would test the former rather than the latter since it is somewhat more robust (and clear to read/debug).
I found a great way to test this using Moq and FluentAssertions. Suppose your unit under test looks like this:
public class Foo
{
private readonly IDataLayer dataLayer;
public Foo(IDataLayer dataLayer)
{
this.dataLayer = dataLayer;
}
public void MethodToTest()
{
using (var transaction = new TransactionScope())
{
this.dataLayer.Foo();
this.dataLayer.Bar();
transaction.Complete();
}
}
}
Your test would look like this (assuming MS Test):
[TestClass]
public class WhenMethodToTestIsCalled()
{
[TestMethod]
public void ThenEverythingIsExecutedInATransaction()
{
var transactionCommitted = false;
var fooTransaction = (Transaction)null;
var barTransaction = (Transaction)null;
var dataLayerMock = new Mock<IDataLayer>();
dataLayerMock.Setup(dataLayer => dataLayer.Foo())
.Callback(() =>
{
fooTransaction = Transaction.Current;
fooTransaction.TransactionCompleted +=
(sender, args) =>
transactionCommitted = args.Transaction.TransactionInformation.Status == TransactionStatus.Committed;
});
dataLayerMock.Setup(dataLayer => dataLayer.Bar())
.Callback(() => barTransaction = Transaction.Current);
var unitUnderTest = new Foo(dataLayerMock.Object);
unitUnderTest.MethodToTest();
// A transaction was used for Foo()
fooTransaction.Should().NotBeNull();
// The same transaction was used for Bar()
barTransaction.Should().BeSameAs(fooTransaction);
// The transaction was committed
transactionCommitted.Should().BeTrue();
}
}
This works great for my purposes.
I'm a Java developer, so I'm uncertain about the C# details, but it seems to me that you need two unit tests here.
The first one should be a "blue sky" test that succeeds. Your unit test should ensure that all records that are ACID appear in the database after the transaction is committed.
The second one should be "wonky" version that does the InsertFoo operation and then throws an exception before attempting the InsertBar. A successful test will show that the exception has been thrown and that neither the Foo nor Bar objects have been committed to the database.
If both of these pass, I'd say that your TransactionScope is working as it should.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With