/tornado-production-skeleton

Simple example of a Tornado app in production
  1. Python 57.6%
  2. JavaScript 42.4%
Python JavaScript

Clone with HTTPS

Use Git or checkout with SVN using the web URL.

Open in Desktop Download ZIP
Find file
Switch branches/tags
Nothing to show
Latest commit c6d3515 Jul 11, 2012 @bdarnell Merge pull request #1 from stephenbalaban/master 
Documentation update
Permalink
Failed to load latest commit information.
chat Various changes to make imports work Feb 23, 2011
production Add comment about numprocs Feb 23, 2011
third_party Various changes to make imports work Feb 23, 2011
.gitignore Add .gitignore Feb 23, 2011
.gitmodules Add tornado submodule Feb 22, 2011
README.markdown Clarified how sitecustomize works (through the site module) and added… Jul 11, 2012
sitecustomize.py Various changes to make imports work Feb 23, 2011

README.markdown

Overview

This is a skeletal example of one way to run a Tornado application in production. It currently covers running the application under Supervisor. Future additions may include automating initial setup and deploying new code (e.g. with Fabric) and running multiple processes behind a proxy (e.g. nginx).

Code layout

chat directory

This is our application; it's just the chat demo from the Tornado distribution.

third_party directory

This directory contains third-party modules we depend on, such as Tornado. Including the module directly in the codebase ensures that updates to these modules can be deployed as a single unit with the application code rather than a separate step (note that this really only makes sense for pure-python modules; C modules like pycurl are probably best managed externally).

Tornado is included as a git submodule; this means that after cloning this repository you must run git submodule update --init, and re-run git submodule update when the version of tornado being used has changed.

sitecustomize.py

sitecustomize.py is the magic that makes the third_party directory work (together with .pth files like third_party/tornado.pth). Simply point your PYTHONPATH at the directory containing sitecustomize.py and third_party will be added to the path automatically. The magic that makes this work is the site module, see http://docs.python.org/library/site.html. The site module is imported when python is initialized, it appends site-specific paths to the module search path.

production directory

This directory contains scripts used for Supervisor and other production services.

Setup instructions

All of these steps should be run as the user the app will be running as (except installing Supervisor).

  1. Install supervisor (e.g. sudo easy_install supervisor).
  2. Create directories: mkdir -p ~/logs ~/apps/tornado-production-skeleton
  3. Make this directory available at ~/apps/tornado-production-skeleton/current (either copy it into that location or use a symlink)
  4. Set up supervisor files in ~. Note that supervisord.conf is a copy because it changes rarely and can be shared by multiple projects, while chat.supervisor is a symlink to the current code. cp ~/apps/tornado-production-skeleton/current/production/supervisord.conf ~; ln -s apps/tornado-production-skeleton/current/production/chat.supervisor ~
  5. Run supervisord (in your home directory)

Usage notes

  • After each reboot, supervisord will need to be restarted. (TODO: instructions on making this happen automatically)
  • Run supervisorctl update after any change to the supervisor config files. To restart the servers when the supervisor config files have not changed, use supervisorctl restart chat:*
  • Use supervisorctl status to check on the status of the process. Once the program starts its logs are written to ~/logs, but if it crashes on startup you'll have to ask supervisor for the output: supervisorctl tail chat:chat-8000 stderr