Open Source API Gateway in Elixir
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


Rackla is an open source framework for building API gateways. When we say API gateway, we mean to proxy and potentially enhance the communication by transforming the data sent over HTTP between servers and clients (such as browsers). The communication can be enhanced by throwing away unnecessary data, concatenating multiple requests into a single request or converting the data between different formats. Another possibility is to change existing APIs so that they work in a different way, or merge any amount of existing APIs into a single new one.

API Gateway

With Rackla, you can asynchronously execute multiple HTTP-requests and transform them in any way you want. The results, encapsulated in the Rackla type, can be transformed with well known functions such as map, flat_map and reduce. By using the pipe operator in Elixir, you can express your new API end-points as pipelines which start with requests that are piped in to transforming functions and finally piped into a response.

Rackla builds on Plug in order to expose new end-points and communicate with clients over HTTP. Internally, it uses Hackney to make HTTP requests and Poison for dealing with JSON. A big thank you to everyone involved in these projects!

The documentation is also available online on HexDocs.

(This project was initially created as part of my masters's thesis: Optimising clients with API gateways.)

Skeleton implementation - ready to use

You can clone the Rackla Skeleton project in order to get a complete working API gateway with runnable example end-points and tests. The skeleton project contains everything you need to easily get started - it contains all "infrastructure" needed to easily expose (and use) your end-points, deploy your API gateway to a cloud service such as Heroku or build a Docker image.

Using Rackla as a library

In mix.exs, add :rackla and :cowboy as dependencies:

defp deps do
    {:rackla, "~> 1.2"},
    {:cowboy, "~> 1.0"} # Or your web server of choice (which works with Plug)

and add :rackla and :cowboy to applications:

def application do
  [applications: [:logger, :rackla, :cowboy]]

You should read Plug's documentation about defining a router and setting up supervised handlers. You can always look at the Rackla Skeleton project for inspiration about how to set everything up.

(Remix is a nice library you can use to get hot reloading during development.)

Example usages

OpenWeatherMap API (JSON)

OpenWeatherMap has started requiring that you sign up and get an API key, the example below does not reflect that.

OpenWeatherMap has an API with the following end-point that we're going to use:,SE. That end-point lets us specify one city to retrieve weather data from, defined by: ?q=Malmo,SE (found at the end of the URL).

This will return the weather data for Malmö in Sweden:

       "description":"scattered clouds",

What we're interested in is the temp value stored in main, and the name field. The goal with the new end-point in our API gateway is to take an arbitrary amount of cities separated by | in the query string, such as: /temperature?malmo,se|halmstad,se|copenhagen,dk|san francisco,us|stockholm,se and return a list of temperatures for these cities.

We want our JSON response to look like this:

      "San Francisco":285.087

And this is how the code in Rackla will look like (explained below):

get "/temperature" do
  temperature_extractor = fn(http_response) ->
    case http_response do
      {:error, reason} ->
        "HTTP request failed because: #{reason}"
      ok_response ->
        case Poison.decode(ok_response) do
          {:ok, json_decoded} ->
            Map.put(%{}, json_decoded["name"], json_decoded["main"]["temp"])

          {:error, reason} ->
            "Failed to decode response because: #{inspect(reason)}"

  |> String.split("|")
  |> Rackla.request
  |> Rackla.response(json: true, compress: true)

Let us walk through the code and explain what is happening here. First we define our endpoint /temperature as we would normally do in Plug. Then we define a function which we'll use in the function in the pipeline, let's get back to it later and first look at the pipeline defined at the bottom of our end-point.

Here's what the pipeline will do:

  • Get the query string from conn (implicitly provided by Plug), in this example it will be the string: "malmo,se|halmstad,se|copenhagen,dk|san francisco,us|stockholm,se".
  • Split the string on | to get a list instead: ["malmo,se", "halmstad,se", "copenhagen,dk", "san francisco,us", "stockholm,se"].
  • Map over the list and convert the city strings into a list of URLs to the OpenWeatherMap API.
  • Request all URLs, this will return a Rackla type which will contain (when ready) the response or an :error tuple on failure for each URL.
  • Map over the results using our function temperature_extractor (explained below).
  • Respond to the client. We use the options :json to encode our response in JSON format and set the appropriate headers (this will take the Elixir map type returned in temperature_extractor and convert it to a JSON map and put all responses in a JSON list). We can also use :compress in order to compress the result with gzip compression (when :compress is true, Rackla will check the request headers to make sure that the client accepts gzip - you can also set it to :force to always respond with gzip).

So let's walk through what the function temperature_extractor does. First of all, we pattern match to make sure that our HTTP request hasn't failed. If it has failed, we simply return a string with the reason for the failure. If our HTTP request has succeed, we try to decode it from JSON format using the library Poison. If the decoding is successful, we create a new Elixir map containing the name and the temperature for the response. This will be, for example, %{"Malmo" => 289.751} in one of the responses. The response function will later be able to encode this Elixir map into a JSON map automatically.

Done! That's all we need to do to make it work!

Instagram (Base64 encoded images)

In this example, we will instead of providing an API, actually serve an entire HTML page that we can view in our browser. While serving a full HTML page is not really Rackla's main goal, this example illustrates how we can make recursive requests and work with chunked responses.

get "/instagram" do
  "<!doctype html><html lang=\"en\"><head></head><body>"
  |> Rackla.just
  |> Rackla.response

  "" <> conn.query_string
  |> Rackla.request
  |> Rackla.flat_map(fn(response) ->
    case response do
      {:error, error} ->

      _ ->
        case Poison.decode(response) do
          {:ok, json} ->
            |> Map.get("data")
            |> Rackla.request
            |> ->
              case img_data do
                {:error, error} ->

                _ ->
                  "<img src=\"data:image/jpeg;base64,#{Base.encode64(img_data)}\" height=\"150px\" width=\"150px\">"

          {:error, _} ->
  |> Rackla.response

  |> Rackla.just
  |> Rackla.response

Once again, let's go through the code to see what is happening. We start by exposing the end-point /instagram as we normally do in Plug. Then we define the first of three pipelines. We store some HTML code in a string, convert it into a Rackla type with the function just and use response to send it to the client.

After we've responded with the HTML code, we can move on to the big middle pipeline. In it, we will call the an Instagram API end-point - the actual response can be seen here: We have to pass an access token to the Instagram API so we let the user supply it via the query string in the browser and add it to the Instagram URL. We call request with the URL and then use flat_map. The reason for using flat_map is because it gives us an easy way to create new requests based on the responses from previous requests.

If we get a response from the Instagram API, we decode it from JSON format to an Elixir data structure with Poison. We then extract the list stored in the key "data" in the Instagram response. This will give us a list of items from our Instagram feed. We then map over these items and extract the URL for the images in standard resolution which will give us a list of URLs pointing to images. We can now pipe this list in to the request function to fetch all these images.

Now, we map over the results - the results will now be binary image data! We can take this binary image data, Base64 encode it and place it inside a HTML image tag. By doing so, the browser can render the response chunks as images directly on our page. In the the "outer" pipeline, we end it with the response function which will send the HTML image tags to the client, in this case the browser. (It is important to notice that response is only used in the outer pipeline and not in the inner pipeline created inside flat_map).

Finally, we create the last pipeline which will send the closing HTML tags.

What is cool about this approach is that the image requests are executed concurrently. This means that the image HTML tags will be sent to the client as soon as they are available (as soon as we get a response from any of the image requests). We only make one request to our API gateway from the client and we only send one response to the client from our API gateway but the images will be sent in chunks so they will show up one after another in the browser just like if we requested them individually.

We will also notice, in this example, that the order is nondeterministic - meaning that we will (most likely) get a different order in which the images are sent every time we refresh the page. If we wanted to preserve the ordering of the images, we could either send the ordering with the chunks and let the client code render the images on the correct position - or we could set the :sync option to true in response which would then wait for the responses and then send them in the appropriate order.

Basic authentication

Sending the username and password as headers:

get "/auth-example" do
  url = ""
  headers = %{"Authorization" => "Basic #{Base.encode64("username:password")}"}

  %Rackla.Request{url: url, headers: headers}
  |> request
  |> response

Adding them as part of the URL:

get "/auth-example" do
  username = "my_username"
  password = "my_password"
  |> request
  |> response

Using a SOCKS5 Proxy or HTTP (Connect) Tunnel

For more detailed information about using proxies, see the documentation for Rackla.Proxy.

SOCKS5 using request setting

%Rackla.Request{url: "", options: %{proxy: %Rackla.Proxy{type: :socks5, host: "localhost", port: 8080}}} 
|> Rackla.request 
|> Rackla.collect

HTTP Tunnel using global setting

|> Rackla.request(proxy: %Rackla.Proxy{type: :connect, host: "localhost", port: 8080}) 
|> Rackla.collect

Adding processing time to a JSON response

get "/resp-time" do
  current_millis = fn() ->
    {mega, sec, micro} = :os.timestamp
    (mega*1000000 + sec)*1000 + round(micro/1000)

  urls = ["", ",se"]

  start_time = current_millis.()

  |> request
  |> map(fn(http_response) ->

    end_time = current_millis.() - start_time

    case http_response do
      {:error, reason} ->
        "HTTP request failed because: #{reason} in time #{end_time}"

      ok_response ->
        case Poison.decode(ok_response) do
          {:ok, json_decoded} ->
            Map.put(json_decoded, "response_time", end_time)

          {:error, reason} ->
            "Failed to decode response because: #{inspect(reason)} in time #{end_time}"
  |> response(json: true)

Reverse proxy / Request forwarding

get "/test/a-simple-request-proxy" do
  # You should check for errors
  {:ok, the_request} = incoming_request()
  |> Map.put(:url, "")
  |> request
  |> response

More examples

A collection of smaller example end-points can be found in found lib/rackla/router.ex which illustrates additional techniques that can be used in Rackla.

The Rackla type

Rackla is also the name of the type used in all of Rackla's functions. Internally, it consists of a list of Elixir processes which communicate with message passing according to a protocol defined inside Rackla (these processes can themselves contain even more nested Rackla types). The Rackla type should never be modified directly!

The Rackla type is created with request which converts one or many HTTP requests to a single Rackla type. You can also encapsulate normal Elixir types in a Rackla type with the functions just or just_list.

Most functions, like map, flat_map and reduce, defined in Rackla will take a Rackla type and return a new Rackla type.

The response function converts the Rackla type to a HTTP response which is sent to the client by utilizing Plug. You can also convert a Rackla type into "normal" Elixir types with the function collect.

It is important to note that once a Rackla type has been used, it is no longer valid:

a = Rackla.just(1)
b = a |> + 1))
# a is now "dead" and can't be used anymore

Under normal circumstances, the Rackla type should be "invisible". Think of it as the box which the data is transported inside and that all the functions you use automatically opens the box, takes out the value for you and then puts it in a new box when you're done with it.

(Or simply think of it as a monad if you're comfortable with that.)

Function overview

The documentation is also available online.


Takes a single string (URL) or a Rackla.Request struct and executes a HTTP request to the defined server. You can, by using the Rackla.Request struct, specify more advanced options for your request such as which HTTP verb to use but also individual connection timeout limits etc. You can also call this function with a list of strings or Rackla.Request structs in order to perform multiple requests concurrently.

This function will return a Rackla type which will contain the results from the request(s) once available or an :error tuple in case of failures such non-responding servers or DNS lookup failures. Per default, on success, it will only contain the response payload but the entire response can be used by setting the option :full to true.


  • :full - If set to true, the Rackla type will contain a Rackla.Response struct with the status code, headers and body (payload), default: false.
  • :connect_timeout - Connection timeout limit in milliseconds, default: 5_000.
  • :receive_timeout - Receive timeout limit in milliseconds, default: 5_000.
  • :insecure - If set to true, SSL certificates will not be checked, default: false.

If you specify any options in a Rackla.Request struct, these will overwrite the options passed to the request function for that specific request.


Returns a new Rackla type, where each encapsulated item is the result of invoking fun on each corresponding encapsulated item.


Rackla.just_list([1,2,3]) |> -> x * 2 end) |> Rackla.collect
[2, 4, 6]


Takes a Rackla type, applies the specified function to each of the elements encapsulated in it and returns a new Rackla type with the results. The given function must return a Rackla type.

This function is useful when you want to create a new request pipeline based on the results of a previous request. In those cases, you can use Rackla.flat_map to access the response from a request and call Rackla.request inside the function since Rackla.request returns a Rackla type.


Rackla.just_list([1,2,3]) |> Rackla.flat_map(fn(x) -> Rackla.just(x * 2) end) |> Rackla.collect
[2, 4, 6]


Invokes fun for each element in the Rackla type passing that element and the accumulator acc as arguments. funs return value is stored in acc. The first element of the collection is used as the initial value of acc (you can also use Rackla.reduce/3 and specify your own accumulator). Returns the accumulated value inside a Rackla type.


Rackla.just_list([1,2,3]) |> Rackla.reduce(fn (x, acc) -> x + acc end) |> Rackla.collect


Takes any type an encapsulates it in a Rackla type.


Rackla.just([1,2,3]) |>
[1, 2, 3]


Takes a list of and encapsulates each of the containing elements separately in a Rackla type.


Rackla.just_list([1,2,3]) |>


Returns the element encapsulated inside a Rackla type, or a list of elements in case the Rackla type contains many elements.


Rackla.just_list([1,2,3]) |> Rackla.collect


Returns a new Rackla type by joining the encapsulated elements from two Rackla types.


Rackla.join(Rackla.just(1), Rackla.just(2)) |> Rackla.collect
[1, 2]


Converts a Rackla type to a HTTP response and send it to the client by using Plug.Conn. The Plug.Conn will be taken implicitly by looking for a variable named conn. If you want to specify which Plug.Conn to use, you can use Rackla.response_conn.

Strings will be sent as is to the client. If the Rackla type contains any other type such as a list, it will be converted into a string by using inspect on it. You can also convert Elixir data types to JSON format by setting the option :json to true.

Using this macro is the same as writing: conn = response_conn(rackla, conn, options)


  • :compress - Compresses the response by applying a gzip compression to it. When this option is used, the entire response has to be sent in one chunk. You can't reuse the conn to send any more data after Rackla.response with :compress set to true has been invoked. When set to true, Rackla will check the request header content-encoding to make sure the client accepts gzip responses. If you want to respond with gzip without checking the request headers, you can set :compress to :force.
  • :json - If set to true, the encapsulated elements will be converted into a JSON encoded string before they are sent to the client. This will also set the header "Content-Type" to the appropriate "application/json; charset=utf-8".


Convert an incoming request (from Plug) to a Rackla.Request. If options is specified, it will be added to the Rackla.Request. For valid options, see documentation for Rackla.Request.

Returns either {:ok, Rackla.Request} or {:error, reason} as per :gen_tcp.recv/2.

The Plug.Conn will be taken implicitly by looking for a variable named conn. If you want to specify which Plug.Conn to use, you can use Rackla.incoming_request_conn.

Using this macro is the same as writing: conn = incoming_request_conn(conn, options)

From Plug.Conn documentation: Because the request body can be of any size, reading the body will only work once, as Plug will not cache the result of these operations. If you need to access the body multiple times, it is your responsibility to store it. Finally keep in mind some plugs like Plug.Parsers may read the body, so the body may be unavailable after being accessed by such plugs.


Rackla source code is released under Apache 2 License. Check LICENSE file for more information.