Ratelimit Architecture

[renuka-fernando]

Instead of going with the envoy/ratelimt gRPC service, we can use a library or our own implementation. This eliminates the extra network hop and extra container, and allows rate limiting to be handled like any other policy.

[renuka-fernando]

Without special casing RataLimit, I thought to include the Envoy ratelimiter in the policy engine as a Policy Implementation (A user can include a gRPC or REST service as a policy implementation as well). So this fits well with our Policy story. In this case, envoy/ratelimit works as a library for Policy Engine. During request processing, the Policy Engine would invoke it via an in-process function call instead of making an external gRPC request.

If we are not using the Envoy rate limit filter (Envoy is not directly call the envoy/ratelimit service), there is no value in using envoy/ratelimit service. Anyway, this is going to be a library for Policy Engine if we are going as described above.

If we go with our own implementation, we can use any rate limit algorithm as well. envoy/ratelimit is using the Fixed Windows algorithm.

In my opinion going with a rate limit library or our own implementation is better.