One of my favorite things about heroku is their simplicity to get up and running on a new app. I’ve recently been trying to replicate parts of their setup that I really like on my AWS servers. So I wanted to share where I’m at so far, and to start off we need a blank EC2 ubuntu instance. In this example, I’m also using RDS, but a quick edit of the connection string and you can use whatever you like. As this example is using nginx and gunicorn to serve a flask application, you might need to make changes to serve pyramid or django applications. (Not many though)

To start off let’s get our new host added to our ssh config file. I’m on a Mac but in any *nix OS, your ssh config is in ~/.ssh/config. Open that file in your favorite editor. vim ~/.ssh/config

Host hostname
  User ubuntu
  IdentityFile ~/.ssh/key.pem

Now we can connect to the instance with a simple name and not worry about keys. For example:

ssh hostname

##Base OS and Environment Setup On first things first let’s get all our packages up to date on the instance. So start an SSH session as shown in the command above, then type:

sudo apt-get update && sudo apt-get upgrade

This updates the package listing and resposity, and then installs any packages that have pending updates. This is not always safe due to OS release cycles; however, on Ubuntu 12.04.01+ it’s safe until we move to the next LTS of Ubuntu.

Now we’re ready to start getting our python, ubuntu, and web serving applications installed. We’re also going to upgrade distribute from pip since MySQL-python requires a newer version. I like to use virtualenvwrapper to setup my environments.

sudo apt-get install python-dev make python-pip nginx libevent-dev git libmysqlclient-dev
sudo easy_install -U distribute

Onward to setting up the tools to run our Flask application. We’ll be using gunicorn to serve the application, watchdog to watch for file changes, and supervisor to manage the processes. Some people prefer to use virtualenvs, but my endgoal is to use OpsWorks to spin up and down at will.

sudo pip install gunicorn
sudo pip install watchdog
sudo pip install supervisor

In my case I’m gonna be serving via SSL, so time to setup the certificates. I always put my certificates in /opt/ssl. The contains the entire certificate chain, and contains the private key used with the certificate.

sudo chgrp admin /opt
sudo chmod 775 /opt
mkdir /opt/ssl
vim /opt/ssl/
vim /opt/ssl/

##Setup Git Hosting Now let’s setup a git repo to recieve our application. This setups a bare git repo that we can push data into. This is what enables us to deploy our application via git. The directory created last is where our app is going to be copied to for the active running copy.

mkdir appname.git
cd appname.git
git init --bare
mkdir /opt/appname

Next, we’re gonna setup a git post-receive hook. This will copy every push to the git repo over to our /opt/appname directory.

vim hooks/post-receive

Put in the following:

GIT_WORK_TREE=/opt/appname git checkout -f

Now make it executable:

chmod +x hooks/post-receive

##Setup Nginx Now we need to build an nginx configuration file for our applications. sudo vim /etc/nginx/sites-available/appname I’m gonna step through this file a bit at a time to explain it. First we’re going to setup the reference to our gunicorn server. We’re going to be running it on localhost port 5000.

upstream appname {
 #server unix:/tmp/gunicorn.sock fail_timeout=0;
 # For a TCP configuration:
 server fail_timeout=0;

Next I want to setup a listener to redirect all HTTP requests to HTTPS:

server {
       listen 80;
       rewrite ^ https://$server_name$request_uri? permanent;

Next we’re going to setup the ssl listener.

server {
 listen 443 ssl;

 client_max_body_size 4G;

 keepalive_timeout 300;

 ssl_certificate /opt/appname/;
 ssl_certificate_key /opt/appname/;
 ssl_protocols SSLv3 TLSv1 TLSv1.1 TLSv1.2;
 ssl_ciphers HIGH:!aNULL:!MD5;

We’re going to setup a directory where we can put temporary files to override the gunicorn proxy. Very useful for scheduled maintenance.

 # path for static files
 root /opt/appname/nginx-static;

 location / {
     # checks for static file, if not found proxy to app
     try_files $uri @proxy_to_app;

Now let’s setup the actually proxying piece for gunicorn. The proxy_pass directive must match the upstream name we defined earlier.

 location @proxy_to_app {
     proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
     proxy_set_header Host $http_host;
     proxy_redirect off;

     proxy_pass http://appname;

Lastly, we need to define a location for server based error pages.

 error_page 500 502 503 504 /500.html;
 location = /500.html {
     root /opt/appname/nginx-static;

The complete nginx conf file as a gist is available here. We need to turn off the default site, and enable the one for our app.

sudo ln -s /etc/nginx/sites-available/appname /etc/nginx/sites-enabled/appname

##Setup Gunicorn Next let’s setup an app config for gunicorn. This will bind gunicorn to the proper interface and port. It also sets up logs in the /opt/appname directory. It creates workers based on the number of processors, and is great for autoscaling on Amazon instances. We also set the timeout to 5 minutes. This file goes in /opt/appname/

import multiprocessing

bind = ''
accesslog = 'access.log'
errorlog = 'error.log'
workers = multiprocessing.cpu_count() * 2 + 1
timeout = 300

##Setup Supervisord Now we need to switch to the root user and setup supervisord. I’ve setup a gist with a common supervisord init script. This was lifted from a stackoverflow post. We need to make the script executable and then update the system rc.d process to utilize the scripts.

sudo su
curl > /etc/init.d/supervisord
cd /etc/init.d
sudo chmod +x supervisord
sudo update-rc.d supervisord defaults

Next we need to setup a conf file for supervisord that will point to our gunicorn process and setup watchmedo to restart the process when we push new files. First let’s start with the general configuration values for supervisord.

file=/tmp/supervisor.sock ; (the path to the socket file)
logfile=/tmp/supervisord.log ; (main log file;default $CWD/supervisord.log)
logfile_maxbytes=50MB ; (max main logfile bytes b4 rotation;default 50MB)
logfile_backups=10 ; (num of main logfile rotation backups;default 10)
loglevel=info ; (log level;default info; others: debug,warn,trace)
pidfile=/var/run/ ; (supervisord pidfile;default
nodaemon=false ; (start in foreground if true;default false)
minfds=1024 ; (min. avail startup file descriptors;default 1024)
minprocs=200 ; (min. avail process descriptors;default 200)
; the below section must remain in the config file for RPC
; (supervisorctl/web interface) to work, additional interfaces may be
; added by defining them in separate rpcinterface: sections
supervisor.rpcinterface_factory = supervisor.rpcinterface:make_main_rpcinterface
serverurl=unix:///tmp/supervisor.sock ; use a unix:// URL for a unix socket
; The below sample program section shows all possible program subsection values,
; create one or more 'real' program: sections to be able to control them under
; supervisor.

Now we need to setup the gunicorn directive. We specify the command to execute gunicorn, and the process name to watch for. Next we specify the directory we want to run the command in, and then the user we want the process to run on. Followed by directives about whether or not we excute the command on server start, and if it dies do we want to restart it. Finally, we specify environment variables in key/value pairs that we use in our app environment.

command=/usr/local/bin/gunicorn appname:app -c /opt/appname/

Finally we setup watchmedo to monitor file changes. It watches /opt/appname recursively for any file changes, and will restart gunicorn via supervisord.

command=/usr/local/bin/watchmedo shell-command --patterns "*.py;*.txt" --recursive --command='/
usr/local/bin/supervisorctl restart gunicorn' /opt/appname

##Push to Git Now back on our desktop let’s push our app via git to the server:

git remote add prod ubuntu@hostname:appname.git
git push prod master

You should now see files under /opt/appname relating to your application on the EC2 instance. Now lets install all the requirements. Some people like to add this to the watchmedo directive in /etc/supervisord.conf.

sudo pip install -r /opt/appname/requirements.txt

Perform any application setup you need. For example, create the database or if you have any background processes that run as part of the application, you’ll want to make sure you put any required environment variables such as database connection strings in the /etc/environment file.

Restart the server with the follow:

sudo shutdown -r now

Now check that supervisord started our processes correctly.

sudo supervisorctl status

We can also check both the gunicorn and supervisord logs in /tmp

Let me know what problems you run into!

Additional Django Resources - Django Gunicorn and Nginx Setup - Graphite on Ubuntu 12.04 LTS – Part II: gunicorn, nginx and supervisord

comments powered by Disqus