How to avoid AddHttpClient shared polly policies

[CodingOctocat]

I have a WebService that is being used in multiple VMs via dependency injection, the WebService is configured with some Polly-provided policies, but I've found that these policies share state, which means that if vm1 is using the WebService for exactly ~2 requests per second (which just don't trigger a rate-limit), then the The other vm's can't make requests because if they do, then a rate limit exception is thrown.

How can this be resolved? The expected behavior should be that the WebServices that different vm's depend on are isolated.

Translated with DeepL.com (free version)

// Inject
services.AddHttpClient<IWebService, WebService>(c => c.BaseAddress = new Uri(context.Configuration["WebApi:BaseUrl"]!))
    .ConfigurePrimaryHttpMessageHandler(() => _ignoreSslErrorHandler)
    .AddPolicyHandler(GetRetryPolicy(3))
    .AddPolicyHandler(GetRateLimitPolicy(2));

// WebService ctor
public WebService(HttpClient httpClient, IResponseParser<QueryResponse<Model>> parser, IConfiguration configuration)
{
    _httpClient = httpClient;
    _responseParser = parser;
    _configuration = configuration;
}

[martincostello]

I don't fully understand your question, but it sounds like you want per-client rate limits.

If that's what you want, then you need to configure your rate limits to work that way using some client-specific information to shard the rate limits so they aren't global to all calls to your web service.

If you want to do that, you should use the Polly.Core rate limiting functionality coupled with client-rate limits using the built-in .NET rate limiting functionality.

The code snippets above use our "v7" Polly API that uses the older ASP.NET Core integration with HttpClientFactory.

[CodingOctocat]

Thanks! I looked at the IHttpClientFactory docs. AddHttpClient injects the service as transient, the problem I'm having is that services.AddResiliencePipeline(‘DefaultStrategies’, ...) , pipelineProvider.GetPipeline(‘DefaultStrategies’);, the WebService is getting the same pipeline, now I give up the pipeline dependency injection and create the pipeline instance directly in the WebService. The problem is solved.

[CodingOctocat]

hi, Why is RateLimiterRejectedException.RetryAfter always null?

[martincostello]

That property is populated only when the underlying rate limiting API gives a value to us:

Polly/src/Polly/RateLimit/RateLimitEngine.cs

Line 13 in 4e6f93f

(bool permit, TimeSpan retryAfter) = rateLimiter.PermitExecution();

Why it isn't would depend on exactly how the rate limiter is configured and what it interacts with.

[CodingOctocat]

The underlying rate limiting API is Polly.RateLimiting or System.Threading.RateLimiting? What confuses me is that I followed the polly documentation for the code, but retryAfter is actually null.

[martincostello]

The former builds on top of the latter - .NET itself provides the rate limiting implementation.