CloudFlare Rate-Limiting like service proof-of-concept implemented using RLN and InterRep

Prerequisites

Docker and docker-compose, for the MongoDb database.

Install instructions

yarn

Startup instructions

Start the DB and the DB explorer containers: docker-compose up -d,

Then run the apps:

yarn interrepMock
yarn server
yarn appExample
yarn singleClient # single client simulation
yarn multiClient # multi client simulation

extra (simulate multiple separate clients in real time):

yarn user1 # normal user
yarn user2 # user that spams
yarn user3 # normal user from other group

Description

This repository contains a PoC application for a CloudFlare-like rate limiting service implemented using the RLN construct and the InterRep linking service. The idea is that users already registered to InterRep (their identity commitment being added to some of the semaphore groups) can access applications protected by the Rate limiting service. If the users are caught spamming, then they will be "banned" from all of the applications protected by the service, by the properties of the RLN protocol. The users will not be be removed from the semaphore groups at the InterRep service, but only at the RLN service.

This is a simple PoC, and much more advanced rate limiting logic can be applied (i.e request limiting and proof generation based on more advanced heuristics), but the purpose of the PoC is how RLN can be applied and used for this kind of applications.

There are 4 different components:

InterRep (web2 to web3 account linking service with privacy properties), serves as access permit
Rate limiting service - rate limiting middleware between the users and the appps
Apps - applications protected by the Rate limiting service (need to register to the rate limiting service for protection first)
Users

The rate limiting service synchronizes the membership tree from the InterRep service. Currently a mock app for the InterRep service is used that stores two semaphore groups. The synchronization with the InterRep membership tree happens periodically (on a preset time interval), by reading the newly added identity commitments, for which the Rate Limiting service is unaware of. This can be improved by adding a smart contract event listener which listens for NewRootHash events from the InterRepGroups contract. The rate limiting service keeps it's own merkle tree, which contains all of the identity commitments from the InterRep's merkle tree. Once a user is banned, the merkle tree in the rate limiting service is modified (only the merkle tree for the group in which the banned user is part of), the leaf for the banned user is zeroed out. If the banned user tries to send new request, they will be blocked by the rate limiting service, because their proof will be invalid. The tree in the InterRep service is not modified in any way.

The connected users to the rate limiting service will get notifications via websockets (SocketIO) when a new user is registered to their group or a user is slashed. The clients can use these notifications to obtain a new witness from the RL service, in order to be able to generate valid proofs.

For more details about the idea, please refer to: https://ethresear.ch/t/decentralised-cloudflare-using-rln-and-rich-user-identities/10774

For the details about InterRep <-> RLN integration and tree syncing, please read the following doc: https://hackmd.io/@aeAuSD7mSCKofwwx445eAQ/SJpo9rwrt.

For a server implementation that synchronises the membership tree from an InterRep subgraph, please check the interrep_integration branch.

RLN Construct variables and usage

spam_threshold - the spam threshold is set to 3. The circuits (NRln) are built with limit=3, polynomial of degree 2 is used.
epoch - string concatenation of the website url and a random timestamp. This allows the spam filtering to be more granular - for url, per time interval. The users will be slashed if they send more than spam_threshold requests per time interval, at a given url.
signal - random string, used as an request id. Must be different for every request, otherwise the requests would be rejected
rln_identifier - random identifier, different for each application.

Extra: More about RLN: https://docs.google.com/presentation/d/1qkBetwCtBwPeK8VHVgit14cq1EWhJ04vM6ldTl6hpD0/edit?usp=sharing