Mir is a framework for implementing, debugging, and analyzing distributed protocols. It has the form of a library that provides abstractions representing different components of a distributed system and an engine orchestrating their interaction.
Mir aims to be general enough to enable implementing distributed protocols in a way that is agnostic to network transport, storage, and cryptographic primitive implementation. All these (and other) usual components of a distributed protocol implementation are encapsulated in abstractions with well-defined interfaces. While Mir provides some implementations of those abstractions to be used directly "out of the box", the consumer is free to provide their own implementations.
The first intended use of Mir is as a scalable and efficient consensus layer in Filecoin subnets and, potentially, as a Byzantine fault-tolerant ordering service in Hyperledger Fabric. However, Mir hopes to be a building block of a next generation of distributed systems, being used by many applications.
Currently Mir includes an implementation of the Trantor modular state machine replication system. It has also been used to implement and evaluate the Alea-BFT protocol.
Mir is a framework for implementing distributed protocols (also referred to as distributed algorithms) meant to run on a distributed system. The basic unit of a distributed system is a node. Each node locally executes (its portion of) a protocol, sending and receiving messages to and from other nodes over a communication network.
Mir models a node of such a distributed system and presents the consumer (the programmer using Mir)
with a Node
abstraction.
The main task of a Node
is to process events.
A node contains one or multiple Module
s that implement the logic for event processing.
Each module independently consumes, processes, and outputs events.
This approach bears resemblance to the actor model,
where events exchanged between modules correspond to messages exchanged between actors.
The Node implements an event loop where all events created by modules are stored in a buffer and
distributed to their corresponding target modules for processing.
For example, when the networking module receives a protocol message over the network,
it generates a MessageReceived
event (containing the received message)
that the node implementation routes to the protocol module, which processes the message,
potentially outputting SendMessage
events that the Node implementation routes back to the networking module.
The architecture described above enables a powerful debugging approach. All Events in the event loop can, in debug mode, be recorded, inspected, or even modified and replayed to the Node using a debugging interface.
In practice, when instantiating a Node, the consumer of Mir provides implementations of these modules to Mir. For example, instantiating a node of a state machine replication system might look as follows:
// Example Node instantiation
node, err := mir.NewNode(
/* some more technical arguments ... */
&modules.Modules{
// ...
"app": NewChatApp(),
"protocol": TOBProtocol,
"net": grpcNetworking,
"crypto": ecdsaCrypto,
},
eventInterceptor,
writeAheadLog,
)
Here the consumer provides modules for networking
(implements sending and receiving messages over the network),
the protocol logic (using some total-order broadcast protocol),
the application (implementing the logic of the replicated app),
and a cryptographic module (able to produce and verify digital signatures using ECDSA).
the eventInterceptor
implements recording of the events passed between the modules
for analysis and debugging purposes.
The writeAheadLog
is a special module that enables the node to recover from crashes.
For more details, see the Documentation.
The programmer working with Mir is free to provide own implementations of these modules, but Mir already comes bundled with several module implementations out of the box.
Mir-BFT is a scalable atomic broadcast protocol. The Mir framework initially started as an implementation of that protocol - thus the related name - but has since been made completely independent of Mir-BFT. Even the implementation of the Mir-BFT protocol itself has been abandoned and replaced by its successor, ISS, which is intended to be the first protocol implemented within Mir. However, since Mir is designed to be modular and versatile, ISS is just one (the first) of the protocols implemented in Mir.
This library is in development. This document describes what the library should become rather than what it currently is. This document itself is more than likely to still change. You are more than welcome to contribute to accelerating the development of the Mir library as an open-source contributor. Have a look at the Contributions section if you want to help out!
Assuming Go version 1.18 or higher is installed, the tests can be run by executing go test ./...
in the root folder of the repository.
The dependencies should be downloaded and installed automatically. Some of the dependencies may require gcc installed.
On Ubuntu Linux, it can be done by invoking sudo apt install gcc
.
If the sources have been updated, it is possible that some of the generated source files need to be updated as well.
More specifically, the Mir library relies on Protocol Buffers and gomock.
The protoc
compiler and the corresponding Go plugin need to be installed as well as mockgen
.
On Ubuntu Linux, those can be installed using
sudo snap install --classic protobuf
go install google.golang.org/protobuf/cmd/protoc-gen-go@latest
go install github.com/golang/mock/[email protected]
Make sure (by configuring your shell interpreter) that the directory with Go binaries (usually ~/go/bin
by default) is included in the PATH
environment variable.
On a default Ubuntu Linux system, for example, this can be achieved by running
echo 'PATH=$PATH:~/go/bin' >> ~/.profile
and restarting the terminal.
Once the dependencies are ready, the generated sources can be updated by executing go generate ./...
in the root directory.
For a description of the design and inner workings of the library, see Mir Library Overview. We also keep a log of Architecture Decision Records (ADRs).
For a small demo application, see /samples/chat-demo
For an automated deployment of Mir on a set of remote machines, see the remote deployment instructions.
To get started using (and contributing to) Mir, in addition to this README, we recommend the following:
- Watch the first introductory video
- Read the Mir Library Overview
- Watch the second introductory video. (Very low-level coding, this is not how Mir coding works in real life - it is for developers to understand how Mir internally works. Realistic coding videos will follow soon.)
- Check out the chat-demo sample application to learn how to use Mir for state machine replication.
- To see an example of using a DSL module (allowing to write pseudocode-like code for the protocol logic), look at the implementation of Byzantine Consistent Broadcast (BCB) being used in the corresponding sample application. Original pseudocode can also be found in these lecture notes (Algorithm 4 (Echo broadcast [Rei94])).
- A more complex example of DSL code is the implementation of the SMR availability layer
(concretely the
multisigcollector
).
To learn about the first complex system being built on Mir, have a look at Trantor, a complex SMR system being implemented using Mir.
Contributions are more than welcome!
If you want to contribute, have a look at the open issues. If you have any questions (specific or general), do not hesitate to drop an email to the active maintainer(s) or write a message to the developers in the public Slack channel #mir-dev of the Filecoin project.
The Mir library source code is made available under the Apache License, version 2.0 (Apache-2.0), located in the LICENSE file.
This project is a continuation of the development started under the name
mirbft
as a Hyperledger Lab.