AlephBFT is an asynchronous and Byzantine fault-tolerant consensus protocol aimed at ordering arbitrary messages (transactions). It has been designed to operate continuously under conditions where there is no bound on message-delivery delay and under the assumption that there is a significant probability of malicious behavior, making it an excellent fit for blockchain-related applications. For more information, check the white paper.
This repository contains a Rust implementation of AlephBFT that offers a convenient API enabling seamless application to various problems. The prime application of the repository is the consensus engine (sometimes called the "finality gadget") of the Aleph Zero blockchain.
The code is split into several Rust packages, each having its own directory -
see the Cargo.toml file, which defines the layout of the whole workspace.
The main package, aleph-bft, is located in the consensus directory.
Additionally, every other package has a short README describing its role
in the AlephBFT toolset.
Every package is documented on docs.rs. Comprehensive documentation is available as a mdBook.
The book can be built locally (assuming you have installed rustup):
cargo install mdbook
cd docs
mdbook serve --open
Highlights:
- The protocol is asynchronous, so it's driven by consensus events as opposed to some clock ticks.
- The performance is still optimal in a partially synchronous environment.
- BFT - secure if less than one third of the committee is malicious.
- Secure against fork bombs.
- Lowered network overhead of sending DAG parent information.
- Thorough testing, including malicious scenarios, and high code coverage.
More details are available in the book.
- Import AlephBFT in your crate
[dependencies] aleph-bft = "^0.37"
- The main entry point is the
run_sessionfunction, which returns a Future that runs the consensus algorithm. To call this function, you need to pass a configuration (defaults are available in the package), and implement certain traits, which will provide all the necessary functionalities, such as networking and message signing. A comprehensive guide is available in the documentation.
We provide two basic examples of running AlephBFT, both of which are not cryptographically secure, and assume honest, but possibly malfunctioning, participants.
The first one, ordering, implements a simple node that produces data items, and then waits for them to be finalized. It can also perform a simulated crash after creating a specified number of items.
For example, you may run the following command:
cd ./examples/ordering
./run.sh
that will launch 2 properly working nodes, and 2 nodes that will crash 3 times each.
The delay before relaunching a crashed node will be set to 1 second.
A faulty node will create 25 items before every crash, and another 25 in the end.
Every node will therefore create 100 items in total, and then wait for other nodes before finishing its run.
See:
./run.sh -h
for further details.
Note that if the number of properly working nodes is less or equal than two times the number of faulty nodes, they will be unable to advance the protocol on their own.
The script will try to start nodes at predefined IP addresses, 127.0.0.1:100XX, where XX denotes the node id. If the port is unavailable, the node will log an error and keep trying to aquire it, waiting 10 seconds between consecutive attempts.
Running this script will result in generating log files node0.log, node1.log, ... corresponding to subsequent nodes.
A directory called aleph-bft-examples-ordering-backup will be created to store data required by the crash recovery mechanism, and the logs from subsequent runs will be appended to existing log files.
The cache and logs will be automatically cleared when launching the script again.
The second example, blockchain, is meant for benchmarking AlephBFT in the blockchain setting.
It implements a simple round-robin blockchain assuming honest participation.
The easiest way to run it is to use the provided script as follows (assuming we start in the root directory):
cd ./examples/blockchain
./run.sh 5
where, again, 5 denotes the number of nodes that will be started.
Here we only assume that the address 127.0.0.1:43000 is available, as the network implementation contains a simple node discovery mechanism.
The achieved transactions per second will be among the final log messages in these files.
For further details, see
cargo run -- --help
The repository is mainly self-contained. It is implemented using Rust's async features and depends only on the
futures crate from the standard library. Moreover, it has some usual dependencies like
log and rand and one bigger for encoding, namely parity-scale-codec. In future work, we plan to get
rid of this dependency.
This release was built and tested against the nightly-2022-10-30 Rust toolchain.
If you want to use another version, edit the rust-toolchain file, or use an override with higher priority.
There are many unit tests and several integration tests that may be run by standard command
cargo test --lib or cargo test --lib --skip medium if you want to run just small tests.
Alternatively, you may run the run_local_pipeline.sh script.
We provide fuzzing tests that try to crash the whole application by creating arbitrary data for the network layer
and feeding it into the member implementation. To run those tests you need to install afl and cargo-fuzz.
cargo-fuzz requires you to use a nightly Rust toolchain. afl differs from cargo-fuzz in that it requires
so called corpus data to operate, i.e. some non-empty data set that do not crash the application.
Both tools are using LLVM's instrumentation capabilities in order to guide the fuzzing process basing on code-coverage statistics.
cargo install cargo-fuzz
cargo install aflcargo fuzz run --features="libfuzz" fuzz_targetYou will need to generate some seed data first in order to run it.
# create some random input containing network data from a locally executed test
mkdir afl_in
cargo build --bin gen_fuzz
./target/debug/gen_fuzz >./afl_in/seedYou might need to reconfigure your operating system in order to proceed - in such a case follow the instructions printed by the afl tool in your terminal.
cargo afl build --features="afl-fuzz" --bin fuzz_target_afl
cargo afl fuzz -i afl_in -o afl_out target/debug/fuzz_target_aflThe gen_fuzz binary is also able to verify data for the afl tool.
cargo build --bin gen_fuzz
./target/debug/gen_fuzz | ./target/debug/gen_fuzz --check-fuzzYou may generate the code coverage summary using the gen_cov_data.sh script and then a detailed
raport for every file with cov_report.sh. Make sure to first install all the required
tools with install_cov_tools.sh.
- Papers: current version, old version
- docs: crate documentation, reference
- Asynchronous liveness is an important theoretical property and there is a lot of technical sophistication that comes in the design of AlephBFT in order to achieve it, however on the practical side there is still little evidence that performing such attacks against liveness in real-world scenarios is possible. Still, no matter how unlikely such attacks might be, we take them very seriously and plan to add randomness to AlephBFT in one of the future releases. We decided to go for a version without randomness first, as it gives an incredibly simple and at the same time secure and robust BFT consensus protocol. Adding randomness introduces some complexity into the protocol, so it makes sense to add it on top of a well-tested, working product. The API of the protocol will not change, and we will make the use of randomness configurable.
- We see a big value in keeping a critical piece of code such as a consensus protocol as
self-contained as possible, so we would like to get rid of the only major dependency -
parity-scale-codec
AlephBFT is licensed under the terms of the Apache License 2.0.
The implementation in this repository is funded by Aleph Zero Foundation.