Create and deploy your Erlang / Cowboy Application on Heroku in 30 minutes


Erlang is a programming language used to build massively scalable soft real-time systems with requirements on high availability.

Cowboy is a small, fast and modular HTTP server written in Erlang.

Heroku is a cloud application platform which allows you to deploy and scale your own application at pleasure.

In this post, we will write an Erlang/Cowboy Web Application from scratch, deploying it on Heroku. Since Heroku offers a free tier for deploying web apps, this will cost you nothing. Also, the entire process should take no more than half an hour.



Create the skeleton of an Erlang app using Rebar

Rebar is the de-facto standard build-tool for Erlang projects.

Fetch rebar from Github and bootstrap it:

$ git clone
$ cd rebar
$ ./bootstrap
$ cd ..

Initialize a new git repository and use rebar to create the skeleton for a new Erlang app. I decided to call my application erlblog. Call your application differently, replacing every occurrence of erlblog with your favourite application name in the instructions below. Please note that this is not optional, since two applications cannot have the same name on Heroku and you don’t dare to clash with my own application.

$ git init erlblog
$ cd erlblog
$ cp ../rebar/rebar .
$ ./rebar create-app appid=erlblog

Commit what you have done in git:

$ git add rebar src
$ git commit -m "Add rebar skeleton"

Create a Heroku application

Login into Heroku using the heroku command from the terminal.

$ heroku login

Use your Heroku email and password to login.

Now create a new Heroku app, using the Erlang buildpack from @archaelus:

$ heroku create erlblog --stack cedar --buildpack

Use Cowboy to create a simple web app

Add the Cowboy web server as a rebar dependency:

$ cat rebar.config

{deps, [
        {cowboy, "0.8.4", {git, "", {tag, "0.8.4"}}}

Add cowboy to the list of applications in your .app.src file. Also, set the http_port environment variable to 8080 (see next paragraphs).

$ cat src/

{application, erlblog,
  {description, ""},
  {vsn, "1"},
  {registered, []},
  {applications, [
  {mod, { erlblog_app, []}},
  {env, [{http_port, 8080}]}

Modify the start/2 function from the erlblog_app module so that Cowboy starts a pool of acceptors when the erlblog application is started. Configure the Cowboy dispatcher with a single dispatching rule, routing all requests to ’/’ to the erlblog_handler (see below).

Heroku assigns random ports to your application and uses the OS environment variable $PORT to inform you about the port on which your web server should listen to. Therefore, in the following code we read that environment variable, defaulting to port 8080 in case the environment variable is not specified. This is useful, for example, if you want to try your web server locally before deploying it on Heroku.

$ cat src/erlblog_app.erl



%% Application callbacks
-export([start/2, stop/1]).

-define(C_ACCEPTORS,  100).
%% ===================================================================
%% Application callbacks
%% ===================================================================

start(_StartType, _StartArgs) ->
    Routes    = routes(),
    Dispatch  = cowboy_router:compile(Routes),
    Port      = port(),
    TransOpts = [{port, Port}],
    ProtoOpts = [{env, [{dispatch, Dispatch}]}],
    {ok, _}   = cowboy:start_http(http, ?C_ACCEPTORS, TransOpts, ProtoOpts),

stop(_State) ->

%% ===================================================================
%% Internal functions
%% ===================================================================
routes() ->
     {'_', [
            {"/", erlblog_handler, []}

port() ->
    case os:getenv("PORT") of
        false ->
            {ok, Port} = application:get_env(http_port),
        Other ->

Let’s now implement a basic HTTP Cowboy handler, which simply replies with a 200 status code and a notorious welcoming message to any incoming request:

$ cat src/erlblog_handler.erl



init(_Transport, Req, []) ->
    {ok, Req, undefined}.

handle(Req, State) ->
    {ok, Req2} = cowboy_req:reply(200, [], <<"Hello world!">>, Req),
    {ok, Req2, State}.

terminate(_Reason, _Req, _State) ->

Finally, let’s create an interface module which will be responsible for starting your erlblog application together with all its dependencies.

$ cat src/erlblog.erl



start() ->
    ok = application:start(crypto),
    ok = application:start(ranch),
    ok = application:start(cowboy),
    ok = application:start(erlblog).

Compile and run your application locally

Compile the erlblog application using rebar:

$ ./rebar get-deps compile

Start the application and verify that everything works as expected:

$ erl -pa ebin deps/*/ebin -s erlblog

From the Erlang shell, type:

1> application:which_applications().

The erlblog application should be included in the output.

Finally, point your browser to:


And verify that the string “Hello World!” is there.

You can use Ctrl-G q to exit the Erlang shell.

If everything works as expected, commit everything to git:

$ git add rebar.config src
git ci -m "Include Cowboy skeleton"

Configure your Heroku app

You need to tell Heroku that you’re going to deploy an Erlang Application. To do so, you need to create a Procfile file, containing your start-up script:

$ cat Procfile

web: erl -pa ebin deps/*/ebin -noshell -noinput -s erlblog

Commit your changes to git:

$ git add Procfile
$ git ci -m "Add Procfile"

You also want to specify that your application requires Erlang R15B01:

$ cat .preferred_otp_version


Commit your changes to git:

$ git add .preferred_otp_version
$ git ci -m "Specify R15B01 as Erlang version"

Deploy your Erlang application on Heroku

That’s the beautiful part:

$ git push heroku master

You should now be able to access the erlblog application at:

If something does not work as expected, you might want to verify the logs for your Heroku app:

$ heroku logs

Profile your deployed Erlang application

Let’s now verify how many requests our erlblog application can handle. Please note that to run the steps below, you need ab and gnuplot installed on your machine.

Using ApacheBench, perform 5000 HTTP requests against your new web server, using 20 concurrent requests. Store the output in the gnuplot.dat file.

$ ab -n 5000 -c 20 -g gnuplot.dat

This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd,
Licensed to The Apache Software Foundation,

Benchmarking (be patient)
Completed 500 requests
Completed 1000 requests
Completed 1500 requests
Completed 2000 requests
Completed 2500 requests
Completed 3000 requests
Completed 3500 requests
Completed 4000 requests
Completed 4500 requests
Completed 5000 requests
Finished 5000 requests

Server Software:        Cowboy
Server Hostname:
Server Port:            80

Document Path:          /
Document Length:        12 bytes

Concurrency Level:      20
Time taken for tests:   71.473 seconds
Complete requests:      5000
Failed requests:        0
Write errors:           0
Total transferred:      615000 bytes
HTML transferred:       60000 bytes
Requests per second:    69.96 [#/sec] (mean)
Time per request:       285.891 [ms] (mean)
Time per request:       14.295 [ms] (mean, across all concurrent requests)
Transfer rate:          8.40 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:      121  130  46.4    127    1317
Processing:   126  154  60.0    140    1176
Waiting:      125  153  60.0    139    1162
Total:        248  284  75.7    268    1495

Percentage of the requests served within a certain time (ms)
  50%    268
  66%    275
  75%    282
  80%    286
  90%    307
  95%    343
  98%    441
  99%    617
 100%   1495 (longest request)

Using our Heroku free tier (1 single dyno worker), our Cowboy Web Server managed to complete all 5000 requests, allowing ~70 requests per second. 90% of the requests have been served in about 300 ms. Of course, such a good result has been possible only because we surely have been hitting some kind of cache in the Heroku servers. Still, not bad for a free hosting solution for a simple Erlang applications.

We can visualize the above results using Gnuplot:

$ gnuplot

gnuplot> plot "gnuplot.dat" using 9 smooth sbezier with lines title "Cowboy Heroku Benchmarking"

The complete source code for the erlblog application is available here.