Ways of unit testing data access layer

29,758

Solution 1

It's unfortunate that you can't find a tool that puts your database into a known state and lets you run your CustomerRepository against the database to test the CustomerRepository. However, the answer is not to start using mocks to mock out all of the ADO calls. By doing that, you end up creating a unit test that doesn't really test any logic: it's just testing that the code is written the way you think it should be written.

Let's say that I end up writing a SQL INSERT as my command to create the customer in the SQL database. Now let's say that we're making a change so that the customer table has different fields (that breaks our INSERT command) and that now we should be using a stored procedure to create the customer. The test with mocks would still pass, even though the implementation it's testing is now broken. Additionally, if you fixed the implementation to use stored procedures, your unit tests would now fail. What is the point of a unit test if it continues to pass when it should fail but then would fail when you fixed the system?

See this question for some possible alternatives. It looks like the marked answer is to actually just end up using DBUnit in C# using IKVM.

So, there might be alternative avenues to continue to explore, but mocking the ADO calls is just going to lead to brittle tests that don't really test anything important.

Solution 2

This layer's job is to connect the code to the database. It has to encapsulate the knowledge about db connection and syntax. In usually maps the domain language to database language. I look at this part of the unit tests as integration test, and as such I test that the database schema is equivalent against real or test database. More on the subject here.

Solution 3

To test DataAccess Layer you will need a more complex structure.

DataAccess Layer will call references from Repository objects. Repo object will call references from Entity Framework DbSets via UnitOfWork design pattern.

DataAccess Layer (TOP)
|
UnitOfWork
|
Repository Pattern Classes
|
EF Context
|
Actual Database

After you set structure, then you will mock repository classes. For example, items will be inserted to DB instead will go to mock object. Later you will assert against your mock object to see item inserted or not.

Please take a look at Implementing the Repository and Unit of Work Patterns

Share:
29,758

Related videos on Youtube

wenic
Author by

wenic

Updated on July 09, 2022

Comments

  • wenic
    wenic almost 2 years

    I have be trying to look for an effective way in unit testing my data access layer in C#. I'm primary a Java developer and have only used C# for about 6 months, in the past i've used a library called DBUnit to test against a known state database. I haven't been able to find a similar active library that can be used, closest seems to be nDBUnit but it hasn't been active for awhile now.

    There seems to be a lot of conflicting methods on how and why in C#. Ideally I want to test the data access layer using mocking without the need to connect to a database and then unit test the store procedure in a separate set of tests.

    In the system I'm working on, the data access layer is to use ADO.net (without the use of the Entity Framework) to call store procedures on a SQL Server.

    Below is a sample code of what I have to work with; to go down the mocking path, I would have to be able to mock the SqlCommand (using IDbCommand) and/or mock the SqlConnection.

    So my question is what seems to be the best way (if there is such a thing) to do this? So far the only way would be to make Proxy object that is passed into the constructor so it can return the mocked Sql* objects for testing.

    I haven't had a chance to look at all the available C# mock libraries available yet.

    public class CustomerRepository : ICustomerRepository
    {
       private string connectionString;
    
       public CustomerRepository (string connectionString)
       {
         this.connectionString = connectionString;
       }
    
       public int Create(Customer customer)
       {
    
         SqlParameter paramOutId = new SqlParameter("@out_id", SqlDbType.Int);
         paramOutId.Direction = ParameterDirection.Output;
         List<SqlParameter> sqlParams = new List<SqlParameter>()
         {
           paramOutId,
           new SqlParameter("@name", customer.Name)
         }
    
         SqlConnection connection = GetConnection();
         try
         {
           SqlCommand command = new SqlCommand("store_proc_name", connection);
    
           command.CommandType = CommandType.StoredProcedure;
    
           command.Parameters.AddRange(sqlParams.ToArray());
    
           int results = command.ExecuteNonQuery();
    
           return (int) paramOutId.Value;
         }
         finally
         {
           CloseConnection(connection);
         }
    
       }
    
    }
    
  • wenic
    wenic over 11 years
    Thanks, I gave this a better look over the weekend, I agree that the test should access a real database, with our Java projects it has worked just like that especially as tables and column names have changed as the projects grew. I took a look at the IKM method, I'd rather not introduce anything too complex for the other developers to manage and understand.
  • wenic
    wenic over 11 years
    Thanks, I think I looked too munch into this, I originally wanted to do that but with the lack of such libraries in c# I started to look for an alternative.
  • Gaui
    Gaui almost 10 years
    Exactly, this is why you should create a service that uses a repository/DAL. You mock the repository, inject it into the service and test that service. Then you're testing the logic in the service in isolation. If the repository breaks (returns incorrect data), the logic will fail and the test will fail. It's always funny to see people mocking a repository and then asserting returns from that repository. You're not testing anything with that.