[already possible] Respect Retry-After HTTP header

The Retry-After response header is used to indicate how long a client should wait before sending the next request. It would be great if RetryPolicy took this into account.

Pseudocode:

var serverWaitDuration = getServerWaitDuration(response.Headers.Get("Retry-After"));
var waitDuration = Math.Max(clientWaitDuration.TotalMilliseconds, serverWaitDuration.TotalMilliseconds);

[reisenberger]

Hi @nmurf! You can already do this with current Polly. You can use a wait-and-retry overload taking the returned response as input to the sleepDurationProvider delegate. Something like:

IAsyncPolicy<HttpResponseMessage> retryHonouringRetryAfter =
    Policy.Handle<HttpRequestException>
        .OrResult<HttpResponseMessage>(r => /* clauses for other status codes you want to handle */
            || r.StatusCode == (HttpStatusCode)429) // RetryAfter
        .WaitAndRetryAsync(
            retryCount: 3, 
            sleepDurationProvider: (retryCount, response, context) => {
                // taking the pseudocode from the question
                var serverWaitDuration = getServerWaitDuration(response.Result?.Headers.Get("Retry-After")); // explaining response.Result? : The response is a Polly type DelegateResult<HttpResponseMessage>. Use null defence, as response.Result will be null if the cause for retry was an exception (in which case there was on result).
                var waitDuration = Math.Max(clientWaitDuration.TotalMilliseconds, serverWaitDuration.TotalMilliseconds);
                return TimeSpan.FromMilliseconds(waitDuration); 
            },
            onRetryAsync: async (response, timespan, retryCount, context) => { 
                /* perhaps some logging, eg the retry count, the timespan delaying for */ 
            }
        );

This is pseudo-code (for speed of response) : you may have to vary some details. For example, Azure Cosmos DB client SDK will wrap this response up in an Exception with a RetryAfter property after making its own retries first. We’d love to see your final version!

This is documented in the Polly wiki here but I’ll aim to reference that also from the main Polly readme, to make it more discoverable. More info also in this similar question.

Let us know how you get on!

[nmurf]

Thanks for the quick response, @reisenberger! Great library – I did not notice the provider overload until you pointed it out.

For posterity, here is an implementation of the method to obtain the header value as a TimeSpan (assumes server is in UTC):

private TimeSpan getServerWaitDuration(DelegateResult<HttpResponseMessage> response)
{
	var retryAfter = response?.Result?.Headers?.RetryAfter;
	if (retryAfter == null)
		return TimeSpan.Zero;
	
	return retryAfter.Date.HasValue
		? retryAfter.Date.Value - DateTime.UtcNow
		: retryAfter.Delta.GetValueOrDefault(TimeSpan.Zero);
}

Here’s the rest:

var clientWaitDurations = DecorrelatedJitter(numRetries, firstWaitDuration.Value, maxWaitDuration.Value)
	.ToArray();

_policy = Policy
	.Handle<HttpRequestException>()
	.OrResult(filter)
	.WaitAndRetryAsync(
		retryCount: numRetries,
		sleepDurationProvider: (retryCount, response, context) => 
			TimeSpan.FromMilliseconds(
				Math.Max(
					clientWaitDurations[retryCount - 1].TotalMilliseconds, 
					getServerWaitDuration(response).TotalMilliseconds));