Cover image for Generated Module As A Guard

Generated Module As A Guard

mudasobwa profile image Aleksei Matiushkin Updated on ・3 min read

Imagine the application that receives some data from the external source. For the sake of an example let’s assume the data is currency rates stream.

The application has a set of rules to filter the incoming data stream. Let’s say we have a list of currencies we are interested in, and we want only the currencies from this list to pass through. Also, sometimes we receive invalid rates (nobody is perfect, our rates provider is not an exception.) So we maintain a long-lived validator that ensures that the rate in the stream looks fine for us and only then we allow the machinery to process it. Otherwise, we just ignore it.

Naïve Approach

The naïve approach to handle this use case would be to maintain a map, containing currency pairs as keys and rules as values, and apply rules to the incoming rates to check whether the rate is of interest or not. Rules would be simple maps specifying the acceptable interval for the rate as min and max values. Something like this (for the sake of an example let's assume rates are coming as maps already, for instance from RabbitMQ or like):

defmodule Validator do
  use Agent

  def start_link,
    do: Agent.start_link(fn -> %{} end, name: __MODULE__)

  def update_rules(currency_pair, rules),
    do: Agent.update(__MODULE__, &(Map.put(&1, currency_pair, rules))

  def valid?(%{currency_pair: pair, rate: rate}) do
    with %{} = rules <- Agent.get(__MODULE__, & &1),
         rule when not is_nil(rule) <- rules[pair],
         r when r > rule.min and r < rule.max  <- rate do
      {:ok, r},
      _ -> :error

This is good, and this works. But can we improve the performance?

Pattern matching

Instead of looking up the map with rules, we might simply generate the module, that will have one function valid? with as many clauses as we have rules. These clauses will directly pattern match the input and return {:ok, rate} tuple (the existence of this clause guarantees that the rate is good.) The last clause will accept all the non-matched garbage to return :error.

Sounds smart?—Indeed. Let’s implement it. Imagine we still have this map with rules.

defmodule Validator do
  def instance!(rules, merge \\ true) do
    mod = Module.concat(["Validator", "Instance"])

    current_rules =
      if Code.ensure_compiled?(mod) do
        rules = if merge, do: apply(mod, :rules, []), else: %{}
        %{} # no previous rules

    Module.create(mod, ast(rules, current_rules), Macro.Env.location(__ENV__))

  defp ast(rules, current_rules) do
    rules = Map.merge(current_rules, rules) # the latter takes precedence
      quote do
        def valid?(_), do: :error
        def rules(), do: unquote(Macro.escape(rules))
      end |
      Enum.map(rules, fn {pair, %{min: min, max: max}} ->
        quote do
          def valid?(%{currency_pair: unquote(pair), rate: rate})
                when rate > unquote(min) and rate < unquote(max),
            do: {:ok, rate}
    ] |> :lists.reverse()

When we need to update our rules, we just call Validator.instance!/1 and receive back the module named Validator.Instance. It has several clauses for valid?/1 function. Let’s see this in action

iex|1  Validator.instance!(%{"USDEUR" => %{min: 1.0, max: 2.0}})
{:module, Validator.Instance,
 <<70, 79, 82, 49, 0, 0, 4, 244, 66, 69, 65, 77, 65, 116, 85, 56, 0, 0, 0, 167,
   0, 0, 0, 17, 25, 69, 108, 105, 120, 105, 114, 46, 86, 97, 108, 105, 100, 97,
   116, 111, 114, 46, 73, 110, 115, 116, 97, ...>>, [valid?: 1, valid?: 1]}

iex|2  Validator.Instance.valid?(%{currency_pair: "USDEUR", rate: 1.5})
{:ok, 1.5}

iex|3  Validator.Instance.valid?(%{currency_pair: "USDEUR", rate: 0.5})

iex|4  Validator.Instance.valid?(%{currency_pair: "USDGBP", rate: 1.5})

Exactly what we needed, and blazingly fast.

Please note, that by default the rules are being merged into the existing rules (if this is a subsequent call to Validator.instance!/2 and the module with the ruleset does already exist. To completely renew the set of rules, one might pass false as the second parameter in call to Validator.instance!/2.

Further Improvement

If the amount of incoming rates is big enough, like thousands per a second, we might use Flow on GenStage to validate them in bulks. That would be probably the topic of the next writing on the subject.

UPD I implemented everything I wanted and packaged the code. Here is a follow-up writing describing Exvalibur package:

Happy generating!

Posted on Oct 30 '18 by:

mudasobwa profile

Aleksei Matiushkin


NB! I am _not_ a member of #DEVCommunity. → I like: Elixir, Erlang, Ruby, R, C, COBOL. → I hate: Apple, JS, Rails, haters. → I am more functional, than object oriented.


markdown guide