FluiDex Backend

FluiDex team is building the first permissionless layer2 orderbook DEX on Ethereum, powered by PLONK zk-rollup.

This repo contains all the backend stuff, including exchange matching engine, rollup state manager, prover cluster (master and workers), and zk circuit codes. You can read through our design rationale here.

Currently it is only a demo/PoC version, and many features are still WIP.

Architecture

Components & Submodules

Submodules:

circuits: ZK-Rollup circuits written in circom. It lies in the rollup-state-manager submodule.
dingir-exchange: the matching engine server. It matches eddsa-signed L2 orders from users, and generates trades. It writes all the 'events' (e.g., orders/trades/balance updates) to the global Kafka message bus.
rollup-state-manager: maintaining the global rollup state Merkle tree. It fetches events/transactions from the Kafka message bus, and updates the Merkle tree accordingly, and generates L2 blocks.
prover-cluster: a master-workers cluster for proving PLONK ZK-SNARK circuits. It loads L2 blocks generated by rollup-state-manager, then proves them, and writes the proofs to databases.
regnbue-bridge: a L1-L2 bridge for fast withdrawal/deposit. Currently in the demo version, it acts like a faucet, sending some initial tokens to each new user of the FluiDex zk-rollup network.

Some external services:

Kafka: used as the global message bus.
PostgreSQL: the main database we use. It stores the match engine history/state, prover-cluster state, rollup (L2 blocks/L2 txs) state etc.
TimescaleDB: time-series databases, used for exchange market data (e.g., K-Line).

Some zero knowledge development tools developed by Fludiex team are used to process the circuits, including snarkit and plonkit

How to run it

Ubuntu 20.04 is the only supported environment currently. You could speed up the building following https://doc.rust-lang.org/cargo/guide/build-cache.html#shared-cache, and more documents can be found here.

# install some dependencies and tools
# including rust / docker / docker-compose / nodejs etc.
$ bash scripts/install_all.sh

# compile zk circuits and setup circuits locally
# start databases and message queue with docker compose
# and launch all the services
# and a mock orders/trades generator
$ bash run.sh

# stop all the processes and destroy docker compose clusters
$ bash stop.sh

Some useful commands have been added to Makefile:

# print the L2 blocks status, total block number, prover block number, etc.
$ make prover_status

# print the latest trades generated by matchengine
$ make new_trades

Now you can also attach a prove client cluster to the backend, see the document

Persist Data

NOTE: for the first time, DO NOT set DX_CLEAN before run.sh. set env DX_CLEAN with false (case insensitive) to skip data purging stage when execute stop.sh.

TODOs

Data availability. And change original inputs as private, then use their hash as the (single) public input.

Known Issues

In order signature verification, user nonce and order id should be, but haven't yet been, signed.
For test convenience, common reference string (CRS) is setup locally rather than by MPC.

fluidex / fluidex-backend