Coerce-rs

Coerce-rs is an asynchronous (async/await) Actor runtime and distributed system framework for Rust. It allows for extremely simple yet powerful actor-based distributed system development. With minimal code, you can build a highly scalable, fault-tolerant modern actor-driven application.

Crates

Crate	Purpose	Latest Version
coerce	The main Coerce runtime and framework
coerce-redis	Actor persistence provider for Redis. Enables event source and snapshots to be read and written from Redis.
coerce-macros	Useful macros allowing for quick implementations of snapshots, JSON-serialisable remote messages and more.
coerce-k8s	Kubernetes discovery provider, automatically discover cluster peers hosted in Kubernetes, based on a configurable pod-selection label

Using Coerce in your own project

First step to using Coerce in your project is to add the coerce crate dependency, this can be done by adding the following to your Cargo.toml:

[dependencies]
coerce = { version = "0.8", features = ["full"] }

Optional: enabling tracing/valuable

Coerce provides support for tracing/valuable, which can be used for enriching logs with information on the actor context. This is currently an unstable feature, which can be enabled by adding the coerce/tracing-unstable feature and the following section to your .cargo/config.toml file:

[build]
rustflags = ["--cfg", "tracing_unstable"]

Note: if your project already depends on tracing crate, you'll need to enable the valuable feature too!

Features

Actors

• Type-safe actors
• Supervision / child spawning
• Location-transparent ActorRef<A> types (ActorRef may comprise of a LocalActorRef<A> or a RemoteActorRef<A>)
• Metrics available out of the box

Remoting

• Communicate with an actor from anywhere in the cluster
• Actors can be deployed locally or to other remote nodes
• Protobuf network protocol
• Actor-driven networking layer

Distributed Sharding

• Actor IDs can resolve to specific shards, which can be spread across a cluster of Coerce nodes
• Automatic load balancing, shards will be fairly allocated across the cluster
• Self-recovering when nodes are lost, actors can be automatically restarted on other healthy nodes

Persistence

• Journaling / event sourcing
• Snapshotting
• Pluggable storage providers (in-memory and redis readily available, MySQL is planned)

Distributed PubSub

• Actors can subscribe to programmable topics from anywhere in the cluster
• System-level topic provided to receive updated system state (e.g new nodes joining, nodes lost etc.)

HTTP API

• Easily accessible metrics and information useful for diagnosis

Building and testing the Coerce libraries

Building Coerce is easy. All you need is the latest Rust stable or nightly installed, along with Cargo.

# Clone the repository
git clone https://github.com/leonhartley/coerce-rs && cd coerce-rs

## run Cargo build to build the entire workspace, including the examples and the tests
cargo build

## Alternatively, if you'd like to build the library, dependencies and run the tests
cargo test --all-features

How to run the examples

Sharded Chat example

ActorSystem

Every actor belongs to an ActorSystem.

async/await Actors

An actor is just another word for a unit of computation. It can have mutable state, it can receive messages and perform actions. One caveat though.. It can only do one thing at a time. This can be useful because it can alleviate the need for thread synchronisation, usually achieved by locking (using Mutex, RwLock etc).

How is this achieved in Coerce?

Coerce uses Tokio's MPSC channels (tokio::sync::mpsc::channel), every actor created spawns a task listening to messages from a Receiver, handling and awaiting the result of the message. Every reference (ActorRef<A: Actor>) holds a Sender<M> where A: Handler<M>, which can be cloned.

Actors can be stopped and actor references can be retrieved by ID from anywhere in your application. IDs are String but if an ID isn't provided upon creation, a new Uuid will be generated. Anonymous actors are automatically dropped ( and Stopped) when all references are dropped. Tracked actors (using global fn new_actor) must be stopped.