# Spawn Elixir SDK
<!-- MDOC !-->
Spawn Elixir SDK is the support library for the Spawn Actors System.
Spawn is a Stateful Serverless Platform for provide the multi-language Actor Model.
For a broader understanding of Spawn please consult its official [repository](https://github.com/eigr/spawn).
The advantage of the Elixir SDK over other SDKs is in Elixir's native ability to connect directly to an Erlang network. For this reason the Elixir SDK allows any valid Elixir application to be part of a Spawn network without the need for a sidecar attached.
## Installation
[Available in Hex](https://hex.pm/packages/spawn_sdk), the package can be installed
by adding `spawn_sdk` and `spawn_statestores_*` to your list of dependencies in `mix.exs`:
```elixir
def deps do
[
{:spawn_sdk, "~> 0.5.0"},
# You can uncomment one of those dependencies if you are going to use Persistent Actors
#{:spawn_statestores_mysql, "~> 0.5.0"},
#{:spawn_statestores_postgres, "~> 0.5.0"},
#{:spawn_statestores_mssql, "~> 0.5.0"},
#{:spawn_statestores_cockroachdb, "~> 0.5.0"},
#{:spawn_statestores_sqlite, "~> 0.5.0"},
]
end
```
### Deploy
Following steps below you will have a valid Elixir application to use in a Spawn cluster.
However, you will still need to generate a container image with your application so that you can use it together with the [Spawn Operator for Kubernetes](https://github.com/eigr/spawn/tree/main/spawn_operator/spawn_operator).
## How to use
After creating an elixir application project create the protobuf files for your business domain.
It is common practice to do this under the **priv/** folder. Let's demonstrate an example:
```protobuf
syntax = "proto3";
package io.eigr.spawn.example;
message MyState {
int32 value = 1;
}
message MyBusinessMessage {
int32 value = 1;
}
```
It is important to try to separate the type of message that must be stored as the actors' state from the type of messages
that will be exchanged between their actors' operations calls. In other words, the Actor's internal state is also represented
as a protobuf type, and it is a good practice to separate these types of messages from the others in its business domain.
In the above case `MyState` is the type protobuf that represents the state of the Actor that we will create later
while `MyBusiness` is the type of message that we will send and receive from this Actor.
Now that we have defined our input and output types as Protobuf types we will need to compile these files to generate their respective Elixir modules. An example of how to do this can be found [here](https://github.com/eigr/spawn/blob/main/spawn_sdk/spawn_sdk_example/compile-example-pb.sh)
> **_NOTE:_** You need to have installed the elixir plugin for protoc. More information on how to obtain and install the necessary tools can be found here [here](https://github.com/elixir-protobuf/protobuf#usage)
Now that the protobuf types have been created we can proceed with the code. Example definition of an Actor:
```elixir
defmodule SpawnSdkExample.Actors.MyActor do
use SpawnSdk.Actor,
name: "jose",
persistent: true,
state_type: Io.Eigr.Spawn.Example.MyState,
deactivate_timeout: 30_000,
snapshot_timeout: 2_000
require Logger
alias Io.Eigr.Spawn.Example.{MyState, MyBusinessMessage}
defact init(%Context{state: state} = ctx) do
Logger.info("[joe] Received InitRequest. Context: #{inspect(ctx)}")
%Value{}
|> Value.state(state)
|> Value.reply!()
end
defact sum(
%MyBusinessMessage{value: value} = data,
%Context{state: state} = ctx
) do
Logger.info("Received Request: #{inspect(data)}. Context: #{inspect(ctx)}")
new_value = if is_nil(state), do: value, else: (state.value || 0) + value
%Value{}
|> Value.of(%MyBusinessMessage{value: new_value}, %MyState{value: new_value})
|> Value.reply!()
end
end
```
In this example we are creating an actor in a Named/Eager way, that is, it is a known actor at compile time. We can also create Unnamed Dynamic/Lazy actors, that is, despite having its abstract behavior defined at compile time, a Lazy actor will only have a concrete instance when it is associated with an identifier/name at runtime. Below follows the same previous actor being defined as abstract.
We declare two actions that the Actor can do. An initialization action that will be called every time an Actor instance is created and an action that will be responsible for performing a simple sum.
Note Keep in mind that any Action that has the names present in the list below will behave as an initialization Action and will be called when the Actor is started (if there is more than one Action with one of these names, only one will be called).
Defaults inicialization Action names: "**init**", "**Init**", "**setup**", "**Setup**"
```elixir
defmodule SpawnSdkExample.Actors.AbstractActor do
use SpawnSdk.Actor,
abstract: true,
persistent: true,
state_type: Io.Eigr.Spawn.Example.MyState
require Logger
alias Io.Eigr.Spawn.Example.{MyState, MyBusinessMessage}
defact sum(
%MyBusinessMessage{value: value} = data,
%Context{state: state} = ctx
) do
Logger.info("Received Request: #{inspect(data)}. Context: #{inspect(ctx)}")
new_value = if is_nil(state), do: value, else: (state.value || 0) + value
%Value{}
|> Value.of(%MyBusinessMessage{value: new_value}, %MyState{value: new_value})
|> Value.reply!()
end
end
```
Notice that the only thing that has changed is the absence of the name argument definition and the abstract argument definition being set to true.
> **_NOTE:_** Can Elixir programmers think in terms of named vs abstract actors as more or less known at startup vs dynamically supervised/registered? That is, defining your actors directly in the supervision tree or using a Dynamic Supervisor for that.
## Side Effects
Actors can also emit side effects to other Actors as part of their response. See an example:
```elixir
defmodule SpawnSdkExample.Actors.AbstractActor do
use SpawnSdk.Actor,
abstract: true,
persistent: false,
state_type: Io.Eigr.Spawn.Example.MyState
require Logger
alias Io.Eigr.Spawn.Example.{MyState, MyBusinessMessage}
alias SpawnSdk.Flow.SideEffect
defact sum(%MyBusinessMessage{value: value} = data, %Context{state: state} = ctx) do
Logger.info("Received Request: #{inspect(data)}. Context: #{inspect(ctx)}")
new_value = if is_nil(state), do: value, else: (state.value || 0) + value
result = %MyBusinessMessage{value: new_value}
new_state = %MyState{value: new_value}
Value.of()
|> Value.value(result)
|> Value.state(new_state)
|> Value.effects(
# This returns a list of side effects. In this case containing only one effect. However, multiple effects can be chained together,
# just by calling the effect function as shown here.
# The delay means that it will be fired asynchronously after 5000 milliseconds (5 seconds)
# If only one effect is desired, you can also choose to use the to/3 function together with Value.effect().
# Example: Values.effect(SideEffect.to(name, func, payload))
SideEffect.of()
|> SideEffect.effect("joe", :sum, result, delay: 5_000, scheduled_to: ~U[2020-01-01 10:00:00.145Z])
# use delay or scheduled_to, not both
)
|> Value.reply!()
end
end
```
In the example above we see that the Actor joe will receive a request as a side effect from the Actor who issued this effect.
Side effects do not interfere with an actor's request-response flow. They will "always" be processed asynchronously and any response sent back from the Actor receiving the effect will be ignored by the effector.
## Pipe and Forward
Actors can also route some commands to other actors as part of their response. See an example:
```elixir
defmodule SpawnSdkExample.Actors.ForwardPipeActor do
use SpawnSdk.Actor,
name: "pipeforward",
abstract: false,
persistent: false,
state_type: Io.Eigr.Spawn.Example.MyState
require Logger
alias Io.Eigr.Spawn.Example.MyBusinessMessage
defact forward_example(%MyBusinessMessage{} = msg, _ctx) do
Logger.info("Received request with #{msg.value}")
Value.of()
|> Value.value(MyBusinessMessage.new(value: 999))
|> Value.forward(
Forward.to("second_actor", "sum_plus_one")
)
|> Value.void()
end
defact pipe_example(%MyBusinessMessage{} = msg, _ctx) do
Logger.info("Received request with #{msg.value}")
Value.of()
|> Value.value(MyBusinessMessage.new(value: 999))
|> Value.pipe(
Pipe.to("second_actor", "sum_plus_one")
)
|> Value.void()
end
end
defmodule SpawnSdkExample.Actors.SecondActorExample do
use SpawnSdk.Actor,
name: "second_actor",
abstract: false,
persistent: false,
state_type: Io.Eigr.Spawn.Example.MyState
require Logger
alias Io.Eigr.Spawn.Example.MyBusinessMessage
defact sum_plus_one(%MyBusinessMessage{} = msg, _ctx) do
Logger.info("Received request with #{msg.value}")
Value.of()
|> Value.value(MyBusinessMessage.new(value: msg.value + 1))
|> Value.void()
end
end
```
We are returning void in both examples so we dont care about what is being stored in the actor state.
In the case above, every time you call the `forward_example` the second_actor's `sum_plus_one` function will receive the value forwarded originally in the invocation as its input. The end result will be:
```elixir
iex> SpawnSdk.invoke("pipeforward", system: "spawn-system", command: "forward_example", payload: %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 1})
{:ok, %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 2}}
```
For the Pipe example, the the second_actor's `sum_plus_one` function will always receive `%MyBusinessMessage{value: 999}` due to getting the value from the previous specification in the `pipe_example` command, the end result will be:
```elixir
iex> SpawnSdk.invoke("pipeforward", system: "spawn-system", command: "pipe_example", payload: %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 1})
{:ok, %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 1000}}
```
## Broadcast
Actors can also send messages to a group of actors at once as an action callback. See the example below:
```elixir
defmodule Fleet.Actors.Driver do
use SpawnSdk.Actor,
abstract: true,
# Set ´driver´ channel for all actors of the same type (Fleet.Actors.Driver)
channel: "drivers",
state_type: Fleet.Domain.Driver
alias Fleet.Domain.{
Driver,
OfferRequest,
OfferResponse,
Point
}
require Logger
@brain_actor_channel "fleet-controllers"
defact update_position(%Point{} = position, %Context{state: %Driver{id: name} = driver} = ctx) do
Logger.info(
"Driver [#{name}] Received Update Position Event. Position: [#{inspect(position)}]. Context: #{inspect(ctx)}"
)
driver_state = %Driver{driver | position: position}
%Value{}
|> Value.of(driver_state, driver_state)
|> Value.broadcast(
Broadcast.to(
@brain_actor_channel,
"driver_position",
driver_state
)
)
|> Value.reply!()
end
end
```
In the case above, every time an Actor "driver" executes the update_position action it will send a message to all the actors participating in the channel called "fleet-controllers".
### Broadcast to External Subscribers
Sometimes you may want to send events out of ActorSystem using Phoenix.PubSub.
One way to do this is to take advantage of the same Broadcast infrastructure that Spawn offers you but indicating an external channel. Below is an example:
1. Create a Listener to receive the events using the `SpawnSdk.Channel.Subscriber` helper module.
```elixir
defmodule SpawnSdkExample.Subscriber do
@moduledoc """
This module exemplifies how to listen for pubsub events that were emitted by actors but that will be treated not by actors but as normal pubsub events.
This is particularly useful for integrations between Spawn and Phoenix LiveView.
"""
use GenServer
require logger
alias SpawnSdk.Channel.Subscriber
@impl true
define init(state) do
Subscriber.subscribe("external.channel")
{:ok, state}
end
@impl true
def handle_info({:receive, payload}, state) do
Logger.info("Received pubsub event #{inspect(payload)}")
{:noreply, state}
end
def start_link(args) do
GenServer.start_link(__MODULE__, args)
end
end
```
You need to match using the {:receive, payload} tuple in your handle_info.
> **_NOTE:_** By default SpawnSdk.Channel.Subscriber will subscribe to pubsub using the atom :actor_channel as an argument.
If you need to change this, just configure your configuration as follows:
***config.exs***
```elixir
config :spawn,
pubsub_group: :your_channel_group_here
```
2. SpawnSdk.System.Supervisor.
```elixir
defmodule SpawnSdkExample.Application do
@moduledoc false
use Application
@impl true
def start(_type, _args) do
children = [
{
SpawnSdk.System.Supervisor,
system: "spawn-system",
actors: [
SpawnSdkExample.Actors.JoeActor
],
extenal_subscribers: [
{SpawnSdkExample.Subscriber, []}
]
}
]
opts = [strategy: :one_for_one, name: SpawnSdkExample.Supervisor]
Supervisor.start_link(children, opts)
end
end
```
The important thing here is to use the external_subscribers attribute. As seen above :external_subscribers accepts a list of specs as a parameter.
3. Set your actor as you normally would and emit your broadcast events using Broadcast.to(channel, payload).
```elixir
defmodule SpawnSdkExample.Actors.JoeActor do
use SpawnSdk.Actor,
name: "joe",
state_type: Io.Eigr.Spawn.Example.MyState
require Logger
alias Io.Eigr.Spawn.Example.{MyState, MyBusinessMessage}
defact sum(%MyBusinessMessage{value: value} = data, %Context{state: state} = ctx) do
Logger.info("[joe] Received Request: #{inspect(data)}. Context: #{inspect(ctx)}")
new_value =
if is_nil(state) do
0 + value
else
(state.value || 0) + value
end
response = %MyBusinessMessage{value: new_value}
%Value{}
|> Value.of(response, %MyState{value: new_value})
|> Value.broadcast(Broadcast.to("my.channel", response))
|> Value.reply!()
end
end
```
## Timers
Actors can also declare Actions that act recursively as timers. See an example below:
```elixir
defmodule SpawnSdkExample.Actors.ClockActor do
use SpawnSdk.Actor,
name: "clock_actor",
state_type: Io.Eigr.Spawn.Example.MyState,
deactivate_timeout: 86_400_000
require Logger
alias Io.Eigr.Spawn.Example.MyState
@set_timer 15_000
defact clock(%Context{state: state} = ctx) do
Logger.info("[clock] Clock Actor Received Request. Context: #{inspect(ctx)}")
new_value = if is_nil(state), do: 0, else: state.value + 1
new_state = MyState.new(value: new_value)
Value.of()
|> Value.state(new_state)
|> Value.noreply!()
end
end
```
> **_NOTE:_** Timers Actions are ephemeral and only exist while the Actor is Enabled, ie running. Therefore Timers are not persistent and will not reactivate a timer's Actor after it is deactivated. Note that in the example above we set the value of deactivate timeout to an exceptionally high number, this is done to make the Actor remain active.
In the example above the ´clock´ action will be called every 15 seconds.
### Declaring the supervision tree
Once we define our actors we can now declare our supervision tree:
```elixir
defmodule SpawnSdkExample.Application do
@moduledoc false
use Application
@impl true
def start(_type, _args) do
children = [
{
SpawnSdk.System.Supervisor,
system: "spawn-system",
actors: [
SpawnSdkExample.Actors.MyActor,
SpawnSdkExample.Actors.AbstractActor
]
}
]
opts = [strategy: :one_for_one, name: SpawnSdkExample.Supervisor]
Supervisor.start_link(children, opts)
end
end
```
## Default Actions
Actors also have some standard actions that are not implemented by the user and that can be used as a way to get the state of an actor without the invocation requiring an extra trip to the host functions. You can think of them as a cache of their state, every time you invoke a default action on an actor it will return the value directly from the Sidecar process without this target process needing to invoke its equivalent host function.
Let's take an example. Suppose Actor Joe wants to know the current state of Actor Robert. What Joe can do is invoke Actor Robert's default action called get_state. This will make Actor Joe's sidecar find Actor Robert's sidecar somewhere in the cluster and Actor Robert's sidecar will return its own state directly to Joe without having to resort to your host function, this in turn will save you a called over the network and therefore this type of invocation is faster than invocations of user-defined actions usually are.
Any invocations to actions with the following names will follow this rule: "**get**",
"**Get**",
"**get_state**",
"**getState**",
"**GetState**"
> **_NOTE:_** You can override this behavior by defining your actor as an action with the same name as the default actions. In this case it will be the Action defined by you that will be called, implying perhaps another network roundtrip
## Running
To run the application via iex we can use the following command:
```
MIX_ENV=prod USER_FUNCTION_PORT=8092 PROXY_DATABASE_TYPE=mysql SPAWN_STATESTORE_KEY=3Jnb0hZiHIzHTOih7t2cTEPEpY98Tu1wvQkPfq/XwqE= iex --name spawn_a2@127.0.0.1 -S mix
```
> **_NOTE:_** This example uses the MySQL database as persistent storage for its actors. And it is also expected that you have previously created a database called eigr-functions-db in the MySQL instance.
The full example of this application can be found [here](https://github.com/eigr/spawn/tree/main/spawn_sdk/spawn_sdk_example).
## Client API Examples
To invoke Actors, use:
```elixir
iex> SpawnSdk.invoke("joe", system: "spawn-system", command: "sum", payload: %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 1})
{:ok, %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 12}}
```
You can invoke actor default functions like "get" to get its current state
```elixir
SpawnSdk.invoke("joe", system: "spawn-system", command: "get")
```
Spawning Actors:
```elixir
iex> SpawnSdk.spawn_actor("robert", system: "spawn-system", actor: SpawnSdkExample.Actors.AbstractActor)
{:ok, %{"robert" => SpawnSdkExample.Actors.AbstractActor}}
```
Invoke Spawned Actors:
```elixir
iex> SpawnSdk.invoke("robert", system: "spawn-system", command: "sum", payload: %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 1})
{:ok, %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 16}}
```
Invoke Actors in a lazy way without having to spawn them before:
```elixir
iex> SpawnSdk.invoke("robert_lazy", ref: SpawnSdkExample.Actors.AbstractActor, system: "spawn-system", command: "sum", payload: %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 1})
{:ok, %Io.Eigr.Spawn.Example.MyBusinessMessage{value: 1}}
```
Invoke Actors with a delay set in milliseconds:
```elixir
iex> SpawnSdk.invoke("joe", system: "spawn-system", command: "ping", delay: 5_000)
{:ok, :async}
```
Invoke Actors scheduled to a specific DateTime:
```elixir
iex> SpawnSdk.invoke("joe", system: "spawn-system", command: "ping", scheduled_to: ~U[2023-01-01 00:32:00.145Z])
{:ok, :async}
```
