Handle CancellationToken for Retry

[peter-csala]

Pull Request

The issue or feature being addressed

#2375

Details on the issue fix or feature implementation

• Retry: handle cancelled CancellationToken as OperationCanceledException
• Assessment
  • Hedging: handles cancelled CancellationToken as OperationCanceledException
  • Timeout: does not handle cancelled CancellationToken as OperationCanceledException
    • We might want to tackle that as a separate PR

Confirm the following

• I started this PR by branching from the head of the default branch
• I have targeted the PR to merge into the default branch
• I have included unit tests for the issue/feature
• I have successfully run a local build

[peter-csala]

It seems like the timeout strategy also does not handle the linked CancellationToken's cancellation as OperationCanceledException

The following code prints 1 as a result

var cts = new CancellationTokenSource();
var p = new ResiliencePipelineBuilder<int>().AddTimeout(TimeSpan.FromSeconds(1)).Build();
var r = await p.ExecuteAsync(async token =>
{
	await Task.Delay(500, token);
	cts.Cancel();
	return 1;
}, cts.Token);

Console.WriteLine(r); // 1

[peter-csala]

For hedging it seems like we are getting OperationCanceledException at the next attempt.

var options = new HedgingStrategyOptions<int>
{
	ShouldHandle = _ => PredicateResult.True(),
	MaxHedgedAttempts = 3,
	Delay = TimeSpan.FromMilliseconds(100),
	OnHedging = static args =>
	{
		Console.WriteLine("Hedging...");
		return default;
	}
};

var cts = new CancellationTokenSource();
var p = new ResiliencePipelineBuilder<int>().AddHedging(options).Build();
var r = await p.ExecuteAsync(async token =>
{
	await Task.Delay(200);
	cts.Cancel();
	return 3;
}, cts.Token);

Console.WriteLine(r);

Output:

Hedging...
Unhandled exception. System.OperationCanceledException: The operation was canceled.
   at Polly.Utils.ExceptionUtilities.TrySetStackTrace[T](T exception) in /_/src/Polly.Core/Utils/ExceptionUtilities.cs:line 23
   ...

[martintmk]

How about just checking for context.CancellationToken.IsCancellationRequested() and then constructing the exception manually?

Alternatively, you can do the same check inside ExecuteCallbackSafeAsync and it will apply everywhere where it is used.

[peter-csala]

I tried to follow the same pattern what we already have for the _timeProvider.DelayAsync:

Polly/src/Polly.Core/Retry/RetryResilienceStrategy.cs

Lines 106 to 113 in 4a44d0e

	try
	{
	await _timeProvider.DelayAsync(delay, context).ConfigureAwait(context.ContinueOnCapturedContext);
	}
	catch (OperationCanceledException e)
	{
	return Outcome.FromException<T>(e);
	}

But yes for sure, I can rewrite the code to avoid try catch. Do you want?