MSTEST - async Testinitialize guarantee test fail

Probably the cleanest way to do this is to have TestInitialize start the asynchronous operation, as such:

[TestClass]
public class UnitTestAsync
{
    private Task<int> val = null;

    [TestInitialize]
    public void TestInitializeMethod()
    {
        val = TestInitializeMethodAsync();
    }

    private async Task<int> TestInitializeMethodAsync()
    {
        return await LongRunningMethod();
    }

    private async Task<int> LongRunningMethod()
    {
        await Task.Delay(20);
        return 10;
    }

    [TestMethod]
    public async Task TestMehod2()
    {
        Assert.AreEqual(10, await val);
    }
}

What you want to do is to use .Result or .Wait() to synchronously block the TestInitialize decorated method. You can do the following:

private int val = 0;

[TestInitialize]
public void TestMehod1()
{
    Task<object> result = await LongRunningMethod();
    result.Wait();

    val = 10;
}

[TestMethod]
public void  TestMehod2()
{
    Assert.AreEqual(10, val);
}

Your code is correct!

To clarify this answer is 5 years, 2 months after the initial question. Back then having async [TestInitialize] might have been a compile error, but these days it isn't.

It's possible to have async [TestInitialize], async [ClassInitialize] and async [TestMethod] just simply use await.

Using async and await properly is probably the cleanest way to do it. I have something like the following in my code where I need to get our category structure in order to be able to test if the my classes work well with the category structure that we have.

private Category rootCategory;

[TestInitialize]
public async Task TestInitialize()
{
    var loader = new CategoryLoader();
    rootCategory = await loader.GetAllCategoriesAsync();
}

[TestInitialize] runs before every [TestMethod], so depending on what I'm trying to test here, it might be better to load only once and then make all the assertions, to not pay for the loading time multiple times. But you need to be careful so that the tests don't affect each other to get consistent and correct results.

Just a note that this is not a unit test anymore since I'm testing integration with external service.