ThrottlingTroll.AzureFunctions 6.1.0

ThrottlingTroll.AzureFunctions

Rate limiting/throttling middleware for Azure Functions (.NET Isolated).

Install from NuGet:

dotnet add package ThrottlingTroll.AzureFunctions

How to configure

Quick example of an ingress config setting:

  "ThrottlingTrollIngress": {
    "Rules": [
      {
        "UriPattern": "/api/values",
        "RateLimit": {
          "Algorithm": "FixedWindow",
          "PermitLimit": 5,
          "IntervalInSeconds": 10
        }
      }
    ]
  }

ThrottlingTroll's configuration (both for ingress and egress) is represented by ThrottlingTrollConfig class. It contains a list of rate limiting Rules and some other settings and it can be configured:

• Statically, via host.json or any other config file.
• Programmatically, at service startup.
• Dynamically, by providing a callback, which ThrottlingTroll will periodically call to get updated values.

See more examples for all of these options below.

Configuring rules and limits

Each Rule defines a pattern that HTTP requests should match. A pattern can include the following properties (all are optional):

• UriPattern - a Regex pattern to match request URI against. Empty string or null means any URI. Note that this value is treated as a Regex, so symbols that have special meaning in Regex language must be escaped (e.g. to match a query string specify \\?abc=123 instead of ?abc=123).
• Method - comma-separated list of request's HTTP methods. E.g. GET,POST. Empty string or null means any method.
• HeaderName - request's HTTP header to check. If specified, the rule will only apply to requests with this header set to HeaderValue.
• HeaderValue - value for HTTP header identified by HeaderName. The rule will only apply to requests with that header set to this value. If HeaderName is specified and HeaderValue is not - that matches requests with any value in that header.
• IdentityId - request's custom Identity ID. If specified, the rule will only apply to requests with this Identity ID. Along with IdentityId you will also need to provide a custom identity extraction routine via IdentityIdExtractor setting.
• IdentityIdExtractor - a routine that takes a request object and should return the caller's identityId. Typically what serves as identityId is some api-key (taken from e.g. a header), or client's IP address, or the oid claim of a JWT token, or something similar. When IdentityIdExtractor is specified and IdentityId is not - different identityIds will automatically get their own rate counters, one counter per each unique identityId.

If any of the above properties is empty or not specified, this means matching any request.

Then each Rule must specify the rate limiting algorithm to be applied for matched requests and its parameters.

Supported rate limiting algorithms

• FixedWindow. No more than PermitLimit requests are allowed in IntervalInSeconds. Example:

  "ThrottlingTrollIngress": {
    "Rules": [
      {
        "RateLimit": {
          "Algorithm": "FixedWindow",
          "PermitLimit": 5,
          "IntervalInSeconds": 10
        }
      }
    ]
  }

• SlidingWindow. No more than PermitLimit requests are allowed in IntervalInSeconds, but that interval is split into NumOfBuckets. The main benefit of this algorithm over FixedWindow is that if a client constantly exceedes PermitLimit, it will never get any valid response and will always get 429 TooManyRequests. Example:

  "ThrottlingTrollIngress": {
    "Rules": [
      {
        "RateLimit": {
          "Algorithm": "SlidingWindow",
          "PermitLimit": 5,
          "IntervalInSeconds": 15,
          "NumOfBuckets": 3
        }
      }
    ]
  }

• Semaphore aka Concurrency Limiter. No more than PermitLimit requests are allowed to be executed concurrently. Example:

  "ThrottlingTrollIngress": {
    "Rules": [
      {
        "RateLimit": {
          "Algorithm": "Semaphore",
          "PermitLimit": 5,
          "TimeoutInSeconds": 60
        }
      }
    ]
  }

TimeoutInSeconds setting is optional, with default value set to 100. It defines the maximum time for the semaphore to be in "locked" state (when e.g. some request starts being processed and then the computing instance crashes).

Configuring whitelist

Requests that should be whitelisted (exempt from the above Rules) can be specified via WhiteList property:

  "ThrottlingTrollIngress": {

    "WhiteList": [
      { "UriPattern": "/api/healthcheck" },
      { "UriPattern": "api-key=my-unlimited-api-key" }
    ]
  }

Entries in the WhiteList array can have the same properties as in the Rules section.

Specifying UniqueName property when sharing a distributed cache instance

When using the same instance of a distributed cache for multiple different services you might also need to specify a value for UniqueName property:

  "ThrottlingTrollIngress": {

    "UniqueName": "MyThrottledService123"
  }

That value will be used as a prefix for cache keys, so that those multiple services do not corrupt each other's rate counters.

How to use for Ingress Throttling

To configure via host.json

1. Enable loading custom configuration sections from host.json:

builder.ConfigureAppConfiguration(configBuilder => {

    configBuilder.AddJsonFile("host.json", optional: false, reloadOnChange: true);
});

2. Add the following ThrottlingTrollIngress section to your host.json file:

  "ThrottlingTrollIngress": {

    "Rules": [
        ... here go rate limiting rules and limits...
    ],

    "WhiteList": [
        ... here go whitelisted URIs...
    ]
  }

3. Add the following call to your startup code:

builder.ConfigureFunctionsWorkerDefaults((hostBuilderContext, workerAppBuilder) => {

    workerAppBuilder.UseThrottlingTroll(hostBuilderContext);
});

To configure programmatically

Use the following configuration method at startup:

builder.ConfigureFunctionsWorkerDefaults((hostBuilderContext, workerAppBuilder) => {

    workerAppBuilder.UseThrottlingTroll(hostBuilderContext, options =>
    {
        options.Config = new ThrottlingTrollConfig
        {
            Rules = new[]
            {
                new ThrottlingTrollRule
                {
                    UriPattern = "/api/values",
                    LimitMethod = new SlidingWindowRateLimitMethod
                    {
                        PermitLimit = 5,
                        IntervalInSeconds = 10,
                        NumOfBuckets = 5
                    }
                },

                // add more rules here...
            }
        };
    });
});

To configure dynamically

Specify a callback that loads rate limits from some shared persistent storage and a time interval to periodically call it:

builder.ConfigureFunctionsWorkerDefaults((hostBuilderContext, workerAppBuilder) => {

    workerAppBuilder.UseThrottlingTroll(hostBuilderContext, options =>
    {
    
        options.GetConfigFunc = async () =>
        {
            var myThrottlingRules = await LoadThrottlingRulesFromDatabase();

            return new ThrottlingTrollConfig
            {
                Rules = myThrottlingRules
            };
        };

        options.IntervalToReloadConfigInSeconds = 10;
    });
});

NOTE: if your callback throws an exception, ThrottlingTroll will get suspended (will not apply any rules) until the callback succeeds again.

To limit clients based on their identityId

If you have a custom way to identify your clients and you want to limit their calls individually, specify a custom IdentityIdExtractor routine like this:

builder.ConfigureFunctionsWorkerDefaults((hostBuilderContext, workerAppBuilder) => {

    workerAppBuilder.UseThrottlingTroll(hostBuilderContext, options =>
    {
    
        options.Config = new ThrottlingTrollConfig
        {
            Rules = new[]
            {
                new ThrottlingTrollRule
                {
                    LimitMethod = new FixedWindowRateLimitMethod
                    {
                        PermitLimit = 3,
                        IntervalInSeconds = 15
                    },

                    IdentityIdExtractor = (request) =>
                    {
                        // Identifying clients e.g. by their api-key
                        return ((IIncomingHttpRequestProxy)request).Request.Query["api-key"]
                    }
                }
            }                    
        };
    });
});

Then ThrottlingTroll will count their requests on a per-identity basis.

To customize responses with a custom response fabric

Provide a response fabric implementation via ResponseFabric option:

builder.ConfigureFunctionsWorkerDefaults((hostBuilderContext, workerAppBuilder) => {

    workerAppBuilder.UseThrottlingTroll(hostBuilderContext, options =>
    {
        // Custom response fabric, returns 400 BadRequest + some custom content
        options.ResponseFabric = async (checkResults, requestProxy, responseProxy, requestAborted) => 
        {
            // Getting the rule that was exceeded and with the biggest RetryAfter value
            var limitExceededResult = checkResults.OrderByDescending(r => r.RetryAfterInSeconds).FirstOrDefault(r => r.RequestsRemaining < 0);
            if (limitExceededResult == null)
            {
                return;
            }

            responseProxy.StatusCode = (int)HttpStatusCode.BadRequest;

            responseProxy.SetHttpHeader("Retry-After", limitExceededResult.RetryAfterHeaderValue);

            await responseProxy.WriteAsync("Too many requests. Try again later.");
        };
    });
});

If you want ThrottlingTroll to proceed with the rest of your processing pipeline (instead of shortcutting to an error response), set ShouldContinueAsNormal to true in your response fabric:

builder.ConfigureFunctionsWorkerDefaults((hostBuilderContext, workerAppBuilder) => {

    workerAppBuilder.UseThrottlingTroll(hostBuilderContext, options =>
    {
        // Custom response fabric, returns 400 BadRequest + some custom content
        options.ResponseFabric = async (checkResults, requestProxy, responseProxy, requestAborted) => 
        {
            await Task.Delay(TimeSpan.FromSeconds(3));

            var ingressResponse = (IIngressHttpResponseProxy)responseProxy;
            ingressResponse.ShouldContinueAsNormal = true;
        };
    });
});

To let a client know their current balance status

When a limit is not exceeded yet, you might want to let your clients know how they're doing by sending their current counter values back to them (e.g. via a custom response header). The current list of check results is available to your code via FunctionContext.Items[ThrottlingTroll.ThrottlingTroll.LimitCheckResultsContextKey], and you can use it to produce your custom header values like this:

  [Function("my-function")]
  public HttpResponseData MyFunction([HttpTrigger(AuthorizationLevel.Anonymous, "get")] HttpRequestData req, FunctionContext context)
  {
      var response = req.CreateResponse(HttpStatusCode.OK);

      // Here is how to set a custom header with the number of remaining requests
      // Obtaining the current list of limit check results from HttpContext.Items
      var limitCheckResults = (List<LimitCheckResult>)context.Items[ThrottlingTroll.ThrottlingTroll.LimitCheckResultsContextKey]!;
      // Now finding the minimal RequestsRemaining number (since there can be multiple rules matched)
      var minRequestsRemaining = limitCheckResults.OrderByDescending(r => r.RequestsRemaining).FirstOrDefault();
      if (minRequestsRemaining != null)
      {
          // Now setting the custom header
          response.Headers.Add("X-Requests-Remaining", minRequestsRemaining.RequestsRemaining.ToString());
      }

      // Do the rest of request processing...

      response.WriteString("OK");
      return response;
  }

To delay responses instead of returning errors

To let ThrottlingTroll spin-wait until the counter drops below the limit set MaxDelayInSeconds to some positive value:

  "ThrottlingTrollIngress": {
    "Rules": [
      {
        "RateLimit": {
          "Algorithm": "Semaphore",
          "PermitLimit": 5
        },
        
        "MaxDelayInSeconds": 60
      }
    ]
  }

In combination with SemaphoreRateLimitMethod, RedisCounterStore and some custom IdentityIdExtractor (which identifies clients by e.g. some query string parameter) this allows to organize named distributed critical sections. Here is an example.

To assign custom costs to different requests

If some of your requests consume more resources than the others, you can assign custom (more than the default 1) costs to them like this:

builder.ConfigureFunctionsWorkerDefaults((hostBuilderContext, workerAppBuilder) => {

    workerAppBuilder.UseThrottlingTroll(hostBuilderContext, options =>
    {
      options.CostExtractor = request =>
      {
          // In this example cost comes as a 'cost' query string parameter
          string? cost = ((IIncomingHttpRequestProxy)request).Request.Query["cost"];
      
          return long.TryParse(cost, out long val) ? val : 1;
      };
    });
});

How to use for Egress Throttling

To configure via host.json

1. Enable loading custom configuration sections from host.json:

builder.ConfigureAppConfiguration(configBuilder => {

    configBuilder.AddJsonFile("host.json", optional: false, reloadOnChange: true);
});

2. Add the following ThrottlingTrollEgress section to your config file:

  "ThrottlingTrollEgress": {

    "Rules": [
        ... here go rate limiting rules and limits...
    ],

    "WhiteList": [
        ... here go whitelisted URIs...
    ]
    
  },

3. Use the following code to configure a named HttpClient at startup:

builder.ConfigureServices(services => { 

    services.AddHttpClient("my-throttled-httpclient").AddThrottlingTrollMessageHandler();
});

4. Get an instance of that HttpClient via IHttpClientFactory:

var throttledHttpClient = this._httpClientFactory.CreateClient("my-throttled-httpclient");

To configure programmatically

Create an HttpClient instance like this:

var myThrottledHttpClient = new HttpClient
(
    new ThrottlingTrollHandler
    (
        new ThrottlingTrollEgressConfig
        {
            Rules = new[]
            {
                new ThrottlingTrollRule
                {
                    UriPattern = "/some/external/url",
                    LimitMethod = new SlidingWindowRateLimitMethod
                    {
                        PermitLimit = 5,
                        IntervalInSeconds = 10,
                        NumOfBuckets = 5
                    }
                },
            }
        }
    )
);

NOTE: normally HttpClient instances should be created once and reused.

To configure dynamically

1. Use the following code to configure a named HttpClient at startup:

builder.ConfigureServices(services => { 

    services.AddHttpClient("my-throttled-httpclient").AddThrottlingTrollMessageHandler(options =>
    {
        options.GetConfigFunc = async () =>
        {
            var myThrottlingRules = await LoadThrottlingRulesFromDatabase();

            return new ThrottlingTrollConfig
            {
                Rules = myThrottlingRules
            };
        };

        options.IntervalToReloadConfigInSeconds = 10;
    });
});

2. Get an instance of that HttpClient via IHttpClientFactory:

var throttledHttpClient = this._httpClientFactory.CreateClient("my-throttled-httpclient");

To propagate from Egress to Ingress

If your service internally makes HTTP requests and you want to automatically propagate 429 TooManyRequests responses up to your service's clients, configure your ThrottlingTroll-enabled HttpClient with PropagateToIngress property set to true:

  "ThrottlingTrollEgress": {

    "PropagateToIngress": true
  }

This will make ThrottlingTrollHandler throw a dedicated ThrottlingTrollTooManyRequestsException, which then will be handled by ThrottlingTrollMiddleware. The Retry-After header value (if present) will also be propagated.

To make HttpClient do automatic retries when getting 429 TooManyRequests

Provide a response fabric implementation via ResponseFabric option and set ShouldRetry to true in it:

builder.ConfigureServices(services => {

    services.AddHttpClient("my-retrying-httpclient").AddThrottlingTrollMessageHandler(options =>
    {
        options.ResponseFabric = async (checkResults, requestProxy, responseProxy, cancelToken) =>
        {
            // Doing no more than 10 automatic retries
            var egressResponse = (IEgressHttpResponseProxy)responseProxy;
            egressResponse.ShouldRetry = egressResponse.RetryCount < 10;
        };
    });

});

HttpClient will then first wait for the amount of time suggested by Retry-After response header and then re-send the request. This will happen no matter whether 429 TooManyRequests status was returned by the external resource or it was the egress rate limit that got exceeded.

Supported Rate Counter Stores

By default ThrottlingTroll will store rate counters in memory, using MemoryCacheCounterStore (which internally uses System.Runtime.Caching.MemoryCache).

Other, distributed counter stores come as separate NuGet packages:

• ThrottlingTroll.CounterStores.Redis - most recommended one, uses Redis.
• ThrottlingTroll.CounterStores.AzureTable - uses Azure Tables (or Cosmos DB with Table API), easiest to configure (only takes a storage connection string), yet not recommended for production scenarios due to a potentially high contention.
• ThrottlingTroll.CounterStores.DistributedCache - uses ASP.NET Core's IDistributedCache and therefore not entirely consistent (because IDistributedCache lacks atomic operations).

You can also create your custom Counter Store by implementing the ICounterStore interface.

How to specify a Rate Counter Store to be used

Either put a desired ICounterStore implementation into DI container:

builder.ConfigureServices(services => {

    services.AddSingleton<ICounterStore>(
        provider => new DistributedCacheCounterStore(provider.GetRequiredService<IDistributedCache>())
    );
});

Or provide it via UseThrottlingTroll() method:

builder.ConfigureFunctionsWorkerDefaults((hostBuilderContext, workerAppBuilder) => {

    // Static programmatic configuration
    workerAppBuilder.UseThrottlingTroll(hostBuilderContext, options =>
    {
        options.CounterStore = new RedisCounterStore(ConnectionMultiplexer.Connect("my-redis-conn-string"));
    });
});

Samples

Here is a sample project, that demonstrates all the above concepts.