defmodule Flippant do @moduledoc """ Feature toggling for Elixir applications. Flippant defines features in terms of `actors`, `groups`, and `rules`: * **Actors** - Typically an actor is a `%User{}` or some other persistent struct that identifies who is using your application. * **Groups** - Groups identify and qualify actors. For example, the `admins` group would identify actors that are admins, while `beta-testers` may identify a few actors that are testing a feature. It is entirely up to you to define groups in your application. * **Rules** - Rules bind groups with individual features. These are evaluated against actors to see if a feature should be enabled. Let's walk through setting up a few groups and rules. ### Groups First, a group that nobody can belong to. This is useful for disabling a feature without deleting it. Groups are registered with a `name` and an evalutation `function`. In this case the name of our group is "nobody", and the function always returns `false`: Flippant.register("nobody", fn(_actor, _values) -> false end) Now the opposite, a group that everybody can belong to: Flippant.register("everybody", fn(_actor, _values) -> true end) To be more specific and define staff only features we define a "staff" group: Flippant.register("staff", fn %User{staff?: staff?}, _values -> staff? end) Lastly, we'll roll out a feature out to a percentage of the actors. It expects a list of integers between `1` and `10`. If the user's id modulo `10` is in the list, then the feature is enabled: Flippant.register("adopters", fn _actor, [] -> false %User{id: id}, samples -> rem(id, 10) in samples end) With some core groups defined we can set up some rules now. ### Rules Rules are comprised of a name, a group, and an optional set of values. Starting with a simple example that builds on the groups we have already created, we'll enable the "search" feature: # Any staff can use the "search" feature Flippant.enable("search", "staff") # 30% of "adopters" can use the "search" feature as well Flippant.enable("search", "adopters", [0, 1, 2]) Because rules are built of binaries and simple data they can be defined or refined at runtime. In fact, this is a crucial part of feature toggling. Rules can be added, removed or modified at runtime. # Turn search off for adopters Flippant.disable("search", "adopters") # On second thought, enable it again for 10% of users Flippant.enable("search", "adopters", [3]) With a set of groups and rules defined we can check whether a feature is enabled for a particular actor: staff_user = %User{id: 1, staff?: true} early_user = %User{id: 2, staff?: false} later_user = %User{id: 3, staff?: false} Flippant.enabled?("search", staff_user) #=> true, staff Flippant.enabled?("search", early_user) #=> false, not an adopter Flippant.enabled?("search", later_user) #=> true, is an adopter If an actor qualifies for multiple groups and *any* of the rules evaluate to true that feature will be enabled for them. Think of the "nobody" and "everybody" groups that were defined earlier: Flippant.enable("search", "everybody") Flippant.enable("search", "nobody") Flippant.enabled?("search", %User{}) #=> true ### Breakdown Evaluating rules requires a round trip to the database. Clearly, with a lot of rules it is inefficient to evaluate each one individually. The `breakdown/1` function helps with this scenario: Flippant.enable("search", "staff") Flippant.enable("delete", "everybody") Flippant.enable("invite", "nobody") Flippant.breakdown(%User{id: 1, staff?: true}) #=> %{"search" => true, "delete" => true, "invite" => false} The breakdown is a simple map of binary keys to boolean values. This is particularly useful for single page applications where you can serialize the breakdown on boot or send it back from an endpoint as JSON. ### Adapters Feature rules are stored in adapters. Flippant comes with a few base adapters: * `Flippant.Adapters.Memory` - An in-memory adapter, ideal for testing (see below). * `Flippant.Adapters.Postgres` - A postgrex powered PostgreSQL adapter. * `Flippant.Adapters.Redis` - A redix powered Redis adapter. For adapter specific options reference the `start_link/1` function of each. Some adapters, notably the `Postgres` adapter, may require setup before they can be used. To simplify the setup process you can run `Flippant.setup()`, or see the adapters documentation for migration details. ### Testing Testing is simplest with the `Memory` adapter. Within `config/test.exs` override the `:adapter`: config :flippant, adapter: Flippant.Adapters.Memory The memory adapter will be cleared whenever the application is restarted, or it can be cleared between test runs using `Flippant.clear(:features)`. ### Defining Groups on Application Start Group definitions are stored in a process, which requires the Flippant application to be started. That means they can't be defined within a configuration file and should instead be linked from `Application.start/2`. You can make `Flippant.register/2` calls directly from the application module, or put them into a separate module and start it as a temporary worker. Here we're starting a temporary worker with the rest of an application: defmodule MyApp do use Application def start(_type, _args) do import Supervisor.Spec, warn: false children = [ worker(MyApp.Flippant, [], restart: :temporary) ] opts = [strategy: :one_for_one, name: MyApp.Supervisor] Supervisor.start_link(children, opts) end end Note that the worker is defined with `restart: :temporary`. Now, define the `MyApp.Flippant` module: defmodule MyApp.Flippant do def start_link do Flippant.register("everybody", &everybody?/2) Flippant.register("nobody", &nobody?/2) Flippant.register("staff", &staff?/2) :ignore end def everybody?(_, _), do: true def nobody?(_, _), do: false def staff?(%User{staff?: staff?}, _), do: staff? end ### Backups and Portability The `dump/1` and `load/1` functions are handy for storing feature backups on disk. The backup may be used to transfer features between database servers, or even between adapters. For example, if you've decided to move away from using Redis and would like to switch to Postgres instead, you could transfer the data with a few commands: # Dump from the Redis instance Flippant.dump("flippant.dump") # Restart the application Application.stop(:flippant) Application.put_env(:flippant, :adapter, Flippant.Adapter.Postgres) Application.ensure_started(:flippant) # Load to the postgres instance Flippant.load("flippant.dump") """ use Supervisor alias Flippant.Config @doc """ Start a Flippant process linked to the current process. """ @spec start_link([{:name, module()}]) :: Supervisor.on_start() def start_link(opts \\ []) do opts = Keyword.put_new(opts, :name, __MODULE__) conf = Config.new(opts) :ok = Config.put(opts[:name], conf) Supervisor.start_link(__MODULE__, conf, name: opts[:name]) end @impl true def init(conf) do Supervisor.init([{conf.adapter, conf.adapter_opts}], strategy: :one_for_one) end # Adapter @doc """ Retrieve the `pid` of the configured adapter process. This will return `nil` if the adapter hasn't been started. """ @spec adapter(name :: atom()) :: pid | nil def adapter(name \\ __MODULE__) do %{adapter: adapter} = Config.get(name) Process.whereis(adapter) end @doc """ Add a new feature without any rules. Adding a feature does not enable it for any groups, that can be done using `enable/2` or `enable/3`. ## Examples Flippant.add("search") #=> :ok """ @spec add(name :: atom(), feature :: binary()) :: :ok def add(name \\ __MODULE__, feature) when is_binary(feature) do name |> adapter() |> GenServer.cast({:add, normalize(feature)}) end @doc """ Generate a mapping of all features and associated rules. Breakdown without any arguments defaults to `:all`, and will list all registered features along with their group and value metadata. It is the only way to retrieve a snapshot of all the features in the system. The operation is optimized for round-trip efficiency. Alternatively, breakdown takes a single `actor` argument, typically a `%User{}` struct or some other entity. It generates a map outlining which features are enabled for the actor. ## Examples Assuming the groups `awesome`, `heinous`, and `radical`, and the features `search`, `delete` and `invite` are enabled, the breakdown would look like: Flippant.breakdown() #=> %{"search" => %{"awesome" => [], "heinous" => []}, "delete" => %{"radical" => []}, "invite" => %{"heinous" => []}} Getting the breakdown for a particular actor: actor = %User{ id: 1, awesome?: true, radical?: false} Flippant.breakdown(actor) #=> %{"delete" => true, "search" => false} """ @spec breakdown(actor :: map | struct | :all) :: map def breakdown(actor \\ :all), do: breakdown(__MODULE__, actor) @spec breakdown(name :: atom(), actor :: map | struct | :all) :: map def breakdown(name, actor) do name |> adapter() |> GenServer.call({:breakdown, actor}) end @doc """ Purge registered features. This is particularly useful in testing when you want to reset to a clean slate after a test. ## Examples Flippant.clear() #=> :ok """ @spec clear(name :: atom()) :: :ok def clear(name \\ __MODULE__) do name |> adapter() |> GenServer.cast(:clear) end @doc """ Disable a feature for a particular group. The feature is kept, but any rules for that group are removed. ## Examples Disable the `search` feature for the `adopters` group: Flippant.disable("search", "adopters") #=> :ok Alternatively, individual values may be disabled for a group. This is useful when a group should stay enabled and only a single value (i.e. user id) needs to be removed. Disable `search` feature for a user in the `adopters` group: Flippant.disable("search", "adopters", [123]) #=> :ok """ @spec disable(binary, binary) :: :ok def disable(feature, group, values \\ []) when is_binary(feature) and is_binary(group) and is_list(values) do disable(__MODULE__, feature, group, values) end @spec disable(name :: atom(), binary, binary) :: :ok def disable(name, feature, group, values) when is_binary(feature) and is_binary(group) and is_list(values) do name |> adapter() |> GenServer.cast({:remove, normalize(feature), group, values}) end @doc """ Dump the full feature breakdown to a file. The `dump/1` command aggregates all features using `breakdown/0`, encodes them as json, and writes the result to a file on disk. Dumps are portable between adapters, so a dump may be subsequently used to load the data into another adapter. ## Examples Dump a daily backup: Flippant.dump((Date.utc_today() |> Date.to_string()) <> ".dump") #=> :ok """ @spec dump(name :: atom(), binary()) :: :ok | {:error, File.posix()} def dump(name \\ __MODULE__, path) when is_binary(path) do dumped = name |> adapter() |> GenServer.call({:breakdown, :all}) |> Jason.encode!() File.write(path, dumped) end @doc """ Enable a feature for a particular group. Features can be enabled for a group along with a set of values. The values will be passed along to the group's registered function when determining whether a feature is enabled for a particular actor. Values are useful when limiting a feature to a subset of actors by `id` or some other distinguishing factor. Value serialization can be customized by using an alternate module implementing the `Flippant.Serializer` behaviour. ## Examples Enable the `search` feature for the `radical` group, without any specific values: Flippant.enable("search", "radical") #=> :ok Assuming the group `awesome` checks whether an actor's id is in the list of values, you would enable the `search` feature for actors 1, 2 and 3 like this: Flippant.enable("search", "awesome", [1, 2, 3]) #=> :ok """ @spec enable(binary, binary, list(any)) :: :ok def enable(feature, group, values \\ []) do enable(__MODULE__, feature, group, values) end @spec enable(name :: atom(), binary, binary, list(any)) :: :ok def enable(name, feature, group, values) when is_binary(feature) and is_binary(group) do name |> adapter() |> GenServer.cast({:add, normalize(feature), {group, values}}) end @doc """ Check if a particular feature is enabled for an actor. If the actor belongs to any groups that have access to the feature then it will be enabled. ## Examples Flippant.enabled?("search", actor) #=> false """ @spec enabled?(name :: atom(), binary, map | struct) :: boolean def enabled?(name \\ __MODULE__, feature, actor) when is_binary(feature) do name |> adapter() |> GenServer.call({:enabled?, normalize(feature), actor}) end @doc """ Check whether a given feature has been registered. If a `group` is provided it will check whether the feature has any rules for that group. ## Examples Flippant.exists?("search") #=> false Flippant.add("search") Flippant.exists?("search") #=> true """ @spec exists?(binary(), binary() | :any) :: boolean() def exists?(feature, group \\ :any) when is_binary(feature) do exists?(__MODULE__, feature, group) end @spec exists?(name :: atom(), binary(), binary() | :any) :: boolean() def exists?(name, feature, group) when is_binary(feature) do name |> adapter() |> GenServer.call({:exists?, normalize(feature), group}) end @doc """ List all known features or only features enabled for a particular group. ## Examples Given the features `search` and `delete`: Flippant.features() #=> ["search", "delete"] Flippant.features(:all) #=> ["search", "delete"] If the `search` feature were only enabled for the `awesome` group: Flippant.features("awesome") #=> ["search"] """ @spec features(:all | binary()) :: list(binary()) def features(group \\ :all) do features(__MODULE__, group) end @spec features(name :: atom(), :all | binary()) :: list(binary()) def features(name, group) do name |> adapter() |> GenServer.call({:features, group}) end @doc """ Restore all features from a dump file. Dumped features may be restored in full using the `load/1` function. During the load process the file will be decoded as json. Loading happens atomically, but it does _not_ clear out any existing features. To have a clean restore you'll need to run `clear/1` first. ## Examples Restore a dump into a clean environment: Flippant.clear(:features) #=> :ok Flippant.load("backup.dump") #=> :ok """ @spec load(name :: atom(), binary()) :: :ok | {:error, File.posix() | binary()} def load(name \\ __MODULE__, path) when is_binary(path) do with {:ok, data} <- File.read(path) do name |> adapter() |> GenServer.cast({:restore, Jason.decode!(data)}) end end @doc """ Rename an existing feature. If the new feature name already exists it will overwritten and all of the rules will be replaced. ## Examples Flippant.rename("search", "super-search") :ok """ @spec rename(name :: atom(), binary, binary) :: :ok def rename(name \\ __MODULE__, old_name, new_name) when is_binary(old_name) and is_binary(new_name) do name |> adapter() |> GenServer.cast({:rename, normalize(old_name), normalize(new_name)}) end @doc """ Fully remove a feature for all groups. ## Examples Flippant.remove("search") :ok """ @spec remove(name :: atom(), binary) :: :ok def remove(name \\ __MODULE__, feature) when is_binary(feature) do name |> adapter() |> GenServer.cast({:remove, normalize(feature)}) end @doc """ Prepare the adapter for usage. For adapters that don't require any setup this is a no-op. For other adapters, such as Postgres, which require a schema/table to operate this will create the necessary table. ## Examples Flippant.setup() :ok """ @spec setup(name :: atom()) :: :ok def setup(name \\ __MODULE__) do name |> adapter() |> GenServer.cast(:setup) end @doc false def update_config(name \\ __MODULE__, key, value) do conf = name |> Config.get() |> Map.put(key, value) Config.put(name, conf) end defp normalize(value) when is_binary(value) do value |> String.downcase() |> String.trim() end end