Why is Task.Delay(1) necessary to advance clock when unit testing with .NET TimeProvider?

Question

I have configured a retry based pipeline with Polly.

// Using Polly for retry logic
private readonly ResiliencePipeline _retryPipeline = new ResiliencePipelineBuilder { TimeProvider = timeProvider }
    .AddRetry(new RetryStrategyOptions
    {
        ShouldHandle = new PredicateBuilder().Handle<ConditionalCheckFailedException>(),
        Delay = TimeSpan.FromMilliseconds(_backoffFactor),
        MaxRetryAttempts = _maxAttempts - 1,

        // Linear backoff increases the delay each time by the backoff factor
        BackoffType = DelayBackoffType.Linear,
        OnRetry = onRetryArguments =>
        {
            logger.LogWarning(
                "Failed to acquire lock. Retrying. {@LogContext}",
                new { onRetryArguments });
            return ValueTask.CompletedTask;
        }
    })
    .Build();

Which I execute using

// Attempt to store the lock with backoff retry 
LockResult result = await _retryPipeline.ExecuteAsync(
   async _ => await AttemptLockStorageAsync(lockId, expiryMilliseconds, attempts++),
   cancellationTokenSource.Token);

When unit testing, I find that I have to add a Task.Delay(1) in order for Polly to perform the retries

// Act
Func<Task<DistributedLock>> func = async () =>
{
    Task<DistributedLock> result = _distributedLockService.AcquireLockAsync(lockId);
    for (int i = 1; i <= 4; i++)
    {
        _timeProvider.Advance(TimeSpan.FromMilliseconds(1000 * i + 1));
        await Task.Delay(1);
    }

    return await result;
};

// Assert
// We expect that we should be able to attempt 5 full times, rather than getting a TaskCancelledException.
(await func.Should().ThrowAsync<TimeoutException>()).WithMessage(
    $"Could not acquire lock {lockId}. Attempted 5 times.");

Why is the Task.Delay necessary?

Edit

TimeProvider provided to the SUT via the primary constructor.

public class DistributedLockService(
    IDistributedLockRepository distributedLockRepository,
    ILogger<DistributedLockService> logger,
    TimeProvider timeProvider)
    : IDisposable, IDistributedLockService

FakeTimer is provided in the unit test constructor

    private readonly FakeTimeProvider _timeProvider = new();

    public DistributedLockServiceTests()
    {
        _timeProvider.SetUtcNow(DateTimeOffset.Parse("2024-01-23", CultureInfo.InvariantCulture));
        _distributedLockService = new DistributedLockService(
            _distributedLockRepository.Object,
            _logger.Object,
            _timeProvider);
    }

Filed a bug report based on minimal reproduction https://github.com/App-vNext/Polly/issues/1932

Answer 1

As documented here

https://github.com/dotnet/extensions/pull/5169/files

The Advance method is used to simulate the passage of time. This can be useful in tests where you need to control the timing of asynchronous operations. When awaiting a task in a test that uses FakeTimeProvider, it's important to use ConfigureAwait(true).

Here's an example:

await provider.Delay(TimeSpan.FromSeconds(delay)).ConfigureAwait(true);

This ensures that the continuation of the awaited task (i.e., the code that comes after the await statement) runs in the original context.

For a more realistic example, consider the following test using Polly:

using Polly;
using Polly.Retry;

public class SomeService(TimeProvider timeProvider)
{
    // Don't do this in real life, not thread safe
    public int Tries { get; private set; }

    private readonly ResiliencePipeline _retryPipeline = new ResiliencePipelineBuilder { TimeProvider = timeProvider }
        .AddRetry(
            new RetryStrategyOptions
            {
                ShouldHandle = new PredicateBuilder().Handle<InvalidOperationException>(),
                Delay = TimeSpan.FromSeconds(1),
                MaxRetryAttempts = 2,
                BackoffType = DelayBackoffType.Linear,
            })
        .Build();

    public async Task<int> PollyRetry(double taskDelay, double cancellationSeconds)
    {
        CancellationTokenSource cts = new(TimeSpan.FromSeconds(cancellationSeconds), timeProvider);
        Tries = 0;

        // get a context from the pool and return it when done
        var context = ResilienceContextPool.Shared.Get(
            // ensure execution continues on captured context 
            continueOnCapturedContext: true, 
            cancellationToken: cts.Token);

        var result = await _retryPipeline.ExecuteAsync(
            async _ =>
            {
                Tries++;

                // Simulate a task that takes some time to complete
                await Task.Delay(TimeSpan.FromSeconds(taskDelay), timeProvider).ConfigureAwait(true);

                if (Tries <= 2)
                {
                    throw new InvalidOperationException();
                }

                return Tries;
            },
            context);

        ResilienceContextPool.Shared.Return(context);

        return result;
    }
}

using Microsoft.Extensions.Time.Testing;

public class SomeServiceTests
{
    [Fact]
    public void PollyRetry_ShouldHave2Tries()
    {
        var timeProvider = new FakeTimeProvider();
        var someService = new SomeService(timeProvider);

        // Act
        var result = someService.PollyRetry(taskDelay: 1, cancellationSeconds: 6);

        // Advancing the time more than one second should resolves the first execution delay.
        timeProvider.Advance(TimeSpan.FromMilliseconds(1001));

        // Advancing the time more than the retry delay time of 1s,
        // and less then the task execution delay should start the second try
        timeProvider.Advance(TimeSpan.FromMilliseconds(1050));

        // Assert
        result.IsCompleted.Should().BeFalse();
        someService.Tries.Should().Be(2);
    }
}