ZIO-Entity

Event sourcing refers to a collection of patterns based on persisting the full history of a domain as a sequence of immutable “events”, rather than persisting just the current state.

CQRS (Command-Query-Responsible-Segregation) is a simple pattern that strictly segregates the responsibility of handling command input into an autonomous system from the responsibility of handling side-effect-free query/read access on the same system.

ZIO-Entities are distributed business domain objects, with a unique identifier, whose state is changed using event-sourcing and automatically persisted into a DB in a functional way.

They apply the CQRS pattern through read-side views. An Entity with a specific id is a Singleton in the cluster, and this means that you will never have to deal with concurrency within nodes as ZIO-Entities implement the "Single Writer Principle" .

Inspiration

This project is inspired by Lagom and Aecor.

Rationale

Historically all event sourcing frameworks have failed. The reason, in my opinion is the high barrier to entry. The concept of event sourcing is simple enough but, applying the concepts requires a deep understanding of the underlying infrastructure.

Other frameworks have issues with testing, they use future, tests are non-deterministic, and they could be flaky. Stubbing the underlying stores is difficult and sub-optimal.

These frameworks are pretty opinionated and very soon you can hit limitations. Other libraries are instead too difficult to use or not maintained anymore.

ZIO-Entity wants to be a simple-to-use library that brings distributed event sourcing in the ZIO world.

Features

ZIO
Easy and versatile API
RPC style Entities
ZIO Stream CQRS
Distributed by nature with pluggable runtimes
Pluggable stores
Schema evolution
Testable in milliseconds

ZIO

ZIO-Entity is integrated in the ZIO ecosystem with all the advantages that the effect library can provide. As a result, tests can run in ms, they are deterministic, fast and easy to reason.

Easy and versatile API

Calling an Entity is as easy as writing

counters("John").increase(10)

RPC style Entities

DDD Entities, use some magic (aka macro) in order to allow RPC style invocation. Amount of boilerplate code is drastically reduced and an entity can be invoked as a normal class. Testing a Zio-Entity is a lot easier since it can be tested like normal code.

The library will distribute the request in the cluster and serialize commands using either Scodec or Protobuf.

The optional annotation @Id can be used to maintain schema compatibility if method is renamed. The id used will be the unique number set in the annotation.

ZIO Stream CQRS

Process ReadSide (CQRS) using ZIO Stream.

Distributed by nature with pluggable runtimes

All the calls against a ZIO-entity are distributed. The distribution mechanism is called Runtime. At the moment Akka-Cluster, Local, LocalWithProtocol and Test are ready. New runtimes using Zookeeper and Native implementations are in the works and new ones can be easily added.

Pluggable stores

The log and snapshot stores can be configured with implementations available like Memory and Postgres with Cassandra in the works.

Schema evolution ready

Plug in protobuf, avro, json in order to manage database and communication evolution.

Testable in milliseconds

Being part of ZIO ecosystem and using ZIO Stream, tests can be easy, quick and deterministic, no more eventually, no more flakyness. Test tools are available in order to test async interaction in a Reactive way.

Installation

Include ZIO-Entity core in your project by adding the following to your build.sbt file:

libraryDependencies += "io.github.thehonesttech" %% "zio-entity-core" % "<latest version>"

You can add the runtime you prefer (currently only akka-runtime is complete)

libraryDependencies += "io.github.thehonesttech" %% "zio-entity-akkaruntime" % "<latest version>"

and the storage you prefer (Postgres in this case)

libraryDependencies += "io.github.thehonesttech" %% "zio-entity-postgres" % "<latest version>"

Example

Interacting with entities is very simple, and they behave like normal ZIO effects:

for {
  counter <- entity[String, Counter, Int, CountEvent, String]
  res <- counter("key").increase(3)
  state <- counter("key").getValue
} yield state

Or subscribing to projections (here a simple projection that counts events):

for {
  counter <- entity[String, Counter, Int, CountEvent, String]
  state <- Ref.make(0)
  killSwitch <- counter
    .readSideSubscription(
      ReadSideParams("read", 
        ConsumerId("1"),
        CounterEntity.tagging, 2, ReadSide.countIncreaseEvents(state, _, _)), _.getMessage)
} yield killSwitch

Below an example of the counter entity:

sealed trait CountEvent

case class CountIncremented(number: Int) extends CountEvent

case class CountDecremented(number: Int) extends CountEvent

trait Counter {
  @Id(1)
  def increase(number: Int): IO[String, Int]

  @Id(2)
  def decrease(number: Int): IO[String, Int]

  @Id(3)
  def getValue: IO[String, Int]
}

class CounterCommandHandler(combinators: Combinators[Int, CountEvent, String]) extends Counter {
  import combinators._
  def increase(number: Int): IO[String, Int] =
    read flatMap { res =>
      append(CountIncremented(number)).as(res + number)
    }

  def decrease(number: Int): IO[String, Int] =
    read flatMap { res =>
      append(CountDecremented(number)).as(res - number)
    }

  def getValue: IO[String, Int] = read
}

Define your event foldable logic:

val eventHandlerLogic: Fold[Int, CountEvent] = Fold(
  initial = 0,
  reduce = {
    case (state, CountIncremented(number)) => UIO.succeed(state + number)
    case (state, CountDecremented(number)) => UIO.succeed(state - number)
    case _ => impossible
  }
)

Define the rpc protocol with the Command handler, the state, the event and the error types:

  implicit val counterProtocol: EntityProtocol[Counter, String] =
  RpcMacro.derive[Counter, String]

Choose the Runtime and build layer

trait Combinators[+State, -Event, Reject] {
  def read: IO[Reject, State]
  def append(es: Event, other: Event*): IO[Reject, Unit]
  def ignore: UIO[Unit] = IO.unit
  def reject[A](r: Reject): IO[Reject, A]
}

trait Entity[Key, Algebra, State, Event, Reject]

private val stores: ZLayer[Any, Nothing, Has[Stores[String, CountEvent, Int]]] =
    Clock.live to MemoryStores.make[String, CountEvent, Int](100.millis, 2)
  
  private val counter: ZLayer[ZEnv, Throwable, Has[Entity[String, Counter, Int, CountEvent, String]]] =
    (Clock.live and stores and Runtime.actorSettings("Test")) to Runtime
      .entityLive("Counter", CounterEntity.tagging,
        EventSourcedBehaviour[Counter, Int, CountEvent, String](
          new CounterCommandHandler(_), CounterEntity.eventHandlerLogic, _.getMessage))
      .toLayer

tobia80/zio-entity