RateLimiting - Incorrect limiting

Question

I have a RabbitMQ Queue, filled with thousands of messages. I need my consumer to consume 1 message per second, so I have implemented a RateLimit policy using Polly. My configuration is as follows:

public static IAsyncPolicy GetPolicy(int mps)
{
    if (mps <= 0)
    {
        throw new ArgumentOutOfRangeException(nameof(mps));
    }
    
    return Policy
        .HandleResult<HttpResponseMessage>(result => {
            return result.StatusCode == System.Net.HttpStatusCode.TooManyRequests;
        })
        .Or<Polly.RateLimit.RateLimitRejectedException>()
        .WaitAndRetryForeverAsync((retryNum, context) => {
            Console.WriteLine($"Retrying. Num: {retryNum}");
            return TimeSpan.FromSeconds(1);
        }).WrapAsync(
            Policy.RateLimitAsync(mps, TimeSpan.FromSeconds(1)));
}

where mps is 1

Now what I've noticed is the following:

• In the beginning, 50 messages are consumed from my Queue, in a span of 1 second. RateLimiter looks like not working
• Then, one message per second is consumed, with WaitAndRetryForeverAsync executing multiple (tens) of times

If I set the mps to 50, the following happens:

• In the beginning 50 messages are immediately consumed
• Then 20 messages per second are consumed (and not 50 as expected)

Is there a bug with the Policy.RateLimitAsync call?
Am I doing something wrong?

Answer 1

I have to emphasize here that at the time of writing the rate limiter policy is considered quite new. It is available since 7.2.3, which is the current stable version. So, it is not as mature as other policies.

---

Based on its documentation I think it's unclear how does it really work.

Let me show you what I mean through a simple example

var localQueue = new Queue<int>();
for (int i = 0; i < 1000; i++)
{
    localQueue.Enqueue(i);
}

RateLimitPolicy rateLimiter = Policy
    .RateLimit(20, TimeSpan.FromSeconds(1));

while (localQueue.TryPeek(out _))
{
    rateLimiter.Execute(() =>
    {
        Console.WriteLine(localQueue.Dequeue());
        Thread.Sleep(10);
    });
}

If you run this program it will print 0 and 1 then it crashes with RateLimitRejectedException.

• Why?
• Why does it not print the first 20 and then crashes?

The answer is that the policy is defined in a way that it can prevent abuse. We wait only 10 milliseconds between two operations. It is considered as an abuse from a policy perspective.

So, without this abuse prevention we would consume the allowed bandwidth under 200 milliseconds and we could not perform any further action in the remaining 800 milliseconds.

Change sleep duration to 49

The result would be more or less the same

• It might be able to reach other number than 1 but it could not reach 20 for sure

Change sleep duration to 50

It could happily consume the whole queue without ever gets halted

Why? Because 1000 milliseconds / 20 allowed execution = 50 milliseconds token

This means that your allowed executions are distributed evenly over time.

Allow burst

Let's play a little bit with the burst mode.

Let's set the maxBurst to 2 and the sleep to 49

RateLimitPolicy rateLimiter = Policy
    .RateLimit(20, TimeSpan.FromSeconds(1), 2);

while (localQueue.TryPeek(out _))
{
    rateLimiter.Execute(() =>
    {
        Console.WriteLine(localQueue.Dequeue());
        Thread.Sleep(49);
    });
}

In this mode the exception will be thrown between 10-20. (Run it multiple times)

Let's change the maxBurst to 4. It will crash between 20-60...

---

Conclusion

So, the rate limiter does not work in the way as you might expect. It allows evenly distributed traffic.