Fsmx

A Finite-state machine implementation in Elixir, with opt-in Ecto friendliness.

Highlights:

  • Plays nicely with both bare Elixir structs and Ecto changesets
  • Ability to wrap transitions inside an Ecto.Multi for atomic updates
  • Guides you in the right direction when it comes to [side effects][a-note-on-side-effects]

Installation

Add dictator to your list of dependencies in mix.exs:

def deps do
  [
    {:fsmx, "~> 0.1.0"}
  ]
end

Usage

Simple state machine

defmodule App.StateMachine do
  defstruct [:state, :data]

  use Fsmx.Struct, transitions: %{
    "one" => ["two", "three"],
    "two" => ["three", "four"],
    "three" => "four"
  }
end

Use it via the Fsmx.transition/2 function:

struct = %App.StateMachine{state: "one", data: nil}

Fsmx.transition(struct, "two")
# {:ok, %App.StateMachine{state: "two"}}

Fsmx.transition(struct, "four")
# {:error, "invalid transition from one to four"}

Callbacks before transitions

You can implement a before_transition/3 callback to mutate the struct when before a transition happens. You only need to pattern-match on the scenarios you want to catch. No need to add a catch-all/do-nothing function at the end (the library already does that for you).

defmodule App.StateMachine do
  # ...

  def before_transition(struct, "two", _destination_state) do
    {:ok, %{struct | data: %{foo: :bar}}}
  end
end

Usage:

struct = %App.StateMachine{state: "two", data: nil}

Fsmx.transition(struct, "three")
# {:ok, %App.StateMachine{state: "three", data: %{foo: :bar}}

Validating transitions

The same before_transition/3 callback can be used to add custom validation logic, by returning an {:error, _} tuple when needed:

defmodule App.StateMachine do
  # ...


  def before_transition(%{data: nil}, _initial_state, "four") do
    {:error, "cannot reacth state four without data"}
  end
end

Usage:

struct = %App.StateMachine{state: "two", data: nil}

Fsmx.transition(struct, "four")
# {:error, "cannot react state four without data"}

Decoupling logic from data

Since logic can grow a lot, and fall out of scope in your structs/schemas, it's often useful to separate all that business logic into a separate module:

defmodule App.StateMachine do
  defstruct [:state]

  use Fsmx.Struct, fsm: App.Logic
end

defmodule App.BusinessLogic do
  use Fsmx.Fsm, transitions: %{
    "one" => ["two", "three"],
    "two" => ["three", "four"],
    "three" => "four"
  }

  # callbacks go here now
  def before_transition(struct, "two", _destination_state) do
    {:ok, %{struct | data: %{foo: :bar}}}
  end

  def before_transition(%{data: nil}, _initial_state, "four") do
    {:error, "cannot reacth state four without data"}
  end
end

Ecto support

Support for Ecto is built in, as long as ecto is in your mix.exs dependencies. With it, you get the ability to define state machines using Ecto schemas, and the Fsmx.Ecto module:

defmodule App.StateMachineSchema do
  use Ecto.Schema

  schema "state_machine" do
    field :state, :string, default: "one"
    field :data, :map
  end

  use Fsmx.Struct, transitions: %{
    "one" => ["two", "three"],
    "two" => ["three", "four"],
    "three" => "four"
  }
end

You can then mutate your state machine in one of two ways:

1. Transition changesets

Returns a changeset that mutates the :state field (or {:error, _} if the transition is invalid).

{:ok, schema} = %App.StateMachineSchema{state: "one"} |> Repo.insert()

Fsmx.transition_changeset(schema, "two")
# #Ecto.Changeset<changes: %{state: "two"}>

You can customize the changeset function, and again pattern match on specific transitions, and additional params:

defmodule App.StateMachineSchema do
  # ...

  # only include sent data on transitions from "one" to "two"
  def transition_changeset(changeset, "one", "two", params) do
    # changeset already includes a :state field change
    changeset
    |> cast(params, [:data])
    |> validate_required([:data])
  end

Usage:

{:ok, schema} = %App.StateMachineSchema{state: "one"} |> Repo.insert()

Fsmx.transition_changeset(schema, "two", %{"data"=> %{foo: :bar}})
# #Ecto.Changeset<changes: %{state: "two", data: %{foo: :bar}>

2. Transition with Ecto.Multi

Note: Please read a note on side effects first. Your future self will thank you.

If a state transition is part of a larger operation, and you want to guarantee atomicity of the whole operation, you can plug a state transition into an Ecto.Multi. The same changeset seen above will be used here:

{:ok, schema} = %App.StateMachineSchema{state: "one"} |> Repo.insert()

Ecto.Multi.new()
|> Fsmx.transition_multi(schema, "transition-id", "two", %{"data" => %{foo: :bar}})
|> Repo.transaction()

When using Ecto.Multi, you also get an additional after_transition_multi/3 callback, where you can append additional operations the resulting transaction, such as dealing with side effects (but again, please no that side effects are tricky)

defmodule App.StateMachineSchema do
  def after_transition_multi(schema, _from, "four") do
    Mailer.notify_admin(schema)
    |> Bamboo.deliver_later()

    {:ok, nil}
  end
end

Note that after_transition_multi/3 callbacks still run inside the database transaction, so be careful with expensive operations. In this example Bamboo.deliver_later/1 (from the awesome Bamboo package) doesn't spend time sending the actual email, it just spawns a task to do it asynchronously.

A note on side effects

Side effects are tricky. Database transactions are meant to guarantee atomicity, but side effects often touch beyond the database. Sending emails when a task is complete is a straight-forward example.

When you run side effects within an Ecto.Multi you need to be aware that, should the transaction later be rolled back, there's no way to un-send that email.

If the side effect is the last operation within your Ecto.Multi, you're probably 99% fine, which works for a lot of cases. But if you have more complex transactions, or if you do need 99.9999% consistency guarantees (because, let's face it, 100% is a pipe dream), then this simple library might not be for you.

Consider looking at Sage, for instance.

# this is *probably* fine
Ecto.Multi.new()
|> Fsmx.transition_multi(schema, "transition-id", "two", %{"data" => %{foo: :bar}})
|> Repo.transaction()

# this is dangerous, because your transition callback
# will run before the whole database transaction has run
Ecto.Multi.new()
|> Fsmx.transition_multi(schema, "transition-id", "two", %{"data" => %{foo: :bar}})
|> Ecto.Multi.update(:update, a_very_unreliable_changeset())
|> Repo.transaction()

Contributing

Feel free to contribute. Either by opening an issue, a Pull Request, or contacting the team directly

If you found a bug, please open an issue. You can also open a PR for bugs or new features. PRs will be reviewed and subject to our style guide and linters.

About

Fsmx is maintained by Subvisual.

<img alt="Subvisual logo" src="https://raw.githubusercontent.com/subvisual/guides/master/github/templates/subvisual_logo_with_name.png" width="350px" />