6 releases

0.1.5 Mar 11, 2020
0.1.4 Feb 20, 2020
0.1.1 May 23, 2018
0.1.0 Apr 25, 2018

#1258 in Asynchronous

Download history 78/week @ 2024-06-16 153/week @ 2024-06-23 53/week @ 2024-06-30 124/week @ 2024-07-07 140/week @ 2024-07-14 231/week @ 2024-07-21 216/week @ 2024-07-28 140/week @ 2024-08-04 125/week @ 2024-08-11 226/week @ 2024-08-18 163/week @ 2024-08-25 162/week @ 2024-09-01 245/week @ 2024-09-08 236/week @ 2024-09-15 282/week @ 2024-09-22 15/week @ 2024-09-29

783 downloads per month
Used in 5 crates (2 directly)

Apache-2.0

23KB
272 lines

Event Sourcing

An eventsourcing library for Rust

One of the benefits of event sourcing is that in most cases, embracing this pattern does not require that much code. However, there's still a bit of boilerplate required as well as the discipline for ensuring the events, commands, and aggregates all perform their roles without sharing concerns.

The fundamental workflow to remember is that commands are applied to aggregates, which then emit one or more events. An aggregate's business logic is also responsible for returning a new state from a previous state combined with a new event. Put mathematically, this looks like:

f(state1, event) = state2

There are some event sourcing libraries that allow for, or even encourage, mutation of state in memory. I prefer a more functional approach, and the design of this library reflects that. It encourages you to write unit tests for your aggregate business logic that are predictable and can be executed in isolation without concern for how you receive events or how you persist them in a store.

To start, you create an undecorated enum for your Command type:

enum LocationCommand {
   UpdateLocation { lat: f32, long: f32, alt: f32 },
}

Next, you create an enum for your events and use a derive macro to write some boilerplate on your behalf. Note how the command variants are imperative statements while the event variants are verbs phrases in the past tense. While this is by convention and not enforced via code, this is a good practice to adopt.

#[derive(Serialize, Deserialize, Debug, Clone, Event)]
#[event_type_version("1.0")]
#[event_source("events://github.com/pholactery/eventsourcing/samples/location")]
enum LocationEvent {
   LocationUpdated { lat: f32, long: f32, alt: f32 },
}

We then define a type that represents the state to be used by an aggregate. With that in place, we write all of our business logic, the core of our event sourcing system, in the aggregate.

#[derive(Debug, Clone)]
struct LocationData {
    lat: f32,
    long: f32,
    alt: f32,
    generation: u64,
}

impl AggregateState for LocationData {
    fn generation(&self) -> u64 {
        self.generation
    }
}

struct Location;
impl Aggregate for Location {
   type Event = LocationEvent;
   type Command = LocationCommand;
   type State = LocationData;

   fn apply_event(state: &Self::State, evt: &Self::Event) -> Result<Self::State> {
       // TODO: validate event
       let ld = match evt {
           LocationEvent::LocationUpdated { lat, long, alt } => LocationData {
               lat: *lat,
               long: *long,
               alt: *alt,
               generation: state.generation + 1,
           },
       };
       Ok(ld)
   }

   fn handle_command(_state: &Self::State, cmd: &Self::Command) -> Result<Vec<Self::Event>> {
       // TODO: add code to validate state and command

       // if validation passes...
       Ok(vec![LocationEvent::LocationUpdated { lat: 10.0, long: 10.0, alt: 10.0 }])
   }
}

For more examples of usage, check out the examples directory in github.

Dependencies

~1.6–6.5MB
~128K SLoC