Skip to content

prezi/changelog

Repository files navigation

changelog

Aims to answer the question "what's changed in the last hour?" in a distributed system. Very useful to have when the proverbial bad things hit the proverbial ventilation hardware. It can drastically drop mean time to recovery.

How?

changelog provides a simple REST API to post events to, and a web interface where you can quickly see and filter events. The basic idea is that you'll send any event that has even a remote chance of causing problems to this system. Later, when something goes wrong, you can quickly check what's changed in the last minutes / hours.

The web interface: Web UI

Sending events is simple:

curl http://changelog.awesomecompany.com/api/events \
  -X POST \
  -H 'Content-Type: application/json' \
  -d '{"criticality": 1, "unix_timestamp": 1395334488, "category": "misc", "description": "cli test"}'

All four JSON fields are required, and there are no other fields the server understands.

Clients implementing this logic:

Some ideas for events to send:

  • deployment, release
  • feature (switch|flag|toggle) changes
  • changes to chef / puppet / ansible / ...
  • changes to DNS configuration
  • cloud instances started, stopped

Getting started

Setup - Docker

A public image is available at prezi/changelog. It runs changelog on Gunicorn with Eventlet workers, and supports SQLite, MySQL and PostgreSQL. It's not completely production-ready out of the box, see below - you should think about at least SSL termination and authentication. You can set any of the configuration variables described below as environment variables on the container with the same name as the configuration variables.

To try it out with an in-container SQLite database:

sudo docker run -p 5000:5000 -it prezi/changelog

Setup - Manually

Prerequisites:

  • a relational database
  • python 3
  • virtualenv
  • optionally: a Sentry server for collecting exceptions; not that there'll be any :)
git clone https://github.com/prezi/changelog.git
cd changelog

# Install dependencies
./setup.sh

# In on shell, start the server
. virtualenv/bin/activate
python application.py

# In another shell, start the Webpack watcher
npm run watch

# When building for production, create a production frontend bundle
npm run build

Database support

changelog uses SQLAlchemy to support a wide range of databases. The client library bindings are not included in requirements.txt to make setup possible in the absence of client libraries on the system. If you want to use a database other than SQLite (you should), then you'll need to manually install the python bindings. Additional requirements files are provided to install known good versions of the bindings:

  • MySQL: pip install -r requirements-mysql.txt
  • PostgreSQL: pip install -r requirements-postgresql.txt

Configuration

Variables

Variable Description Default
ALCHEMY_URL SQL Alchemy connection string. sqlite:///changelog.db
LISTEN_HOST IP address where the application will listen when started with python application.py. 127.0.0.1 (0.0.0.0 in Docker)
LISTEN_PORT Port where the application will listen when started with python application.py. 5000
USE_SENTRY Send exceptions to Sentry? False
SENTRY_DSN Sentry DSN, used only if USE_SENTRY is True. None

The default configuration values defined are in settings.py.

With Docker

You can pass in any of these variables as an environment variable to the Docker container. Additional, docker-specific variables:

Variable Description Default
GUNICORN_WORKER_COUNT Number of worker processes started by Gunicorn. 2

For example, to use a PostgreSQL database:

sudo docker run -d --name changelog-db postgres
sudo docker run --rm --link changelog-db:db postgres psql -h db -U postgres -c 'CREATE DATABASE changelog'
sudo docker run -it -p 5000:5000 --link changelog-db:db -e ALCHEMY_URL=postgres://postgres@db/changelog prezi/changelog

Without Docker

You can set the environment variable CHANGELOG_SETTINGS_PATH to point to a python file. That file can set the values detailed below. The application prints the final configuration at startup to make debugging this easier (not that there's anything to debug, but configuration always needs debugging). The file pointed to by CHANGELOG_SETTINGS_PATH can, for example, look like this:

LISTEN_PORT = 8080
ALCHEMY_URL = 'sqlite:////opt/foo/bar/local/changelog.db'

Considerations for running in production

  • Running under a WSGI server is highly recommended. This project uses the Flask framework, see their documentation for running under Apache or Standalone WSGI containers (Uses Gunicorn with Eventlet workers in the provided Docker image)
  • No authentication is provided, you'll probably want to put some authenticating proxy in front of this application. Pull requests for adding authentication support are of course welcome.
  • Similarly, no HTTPS termination is provided. Ideally the WSGI container will take care of that.
  • Once you have a good number of events, enabling response compression (think mod_deflate) can speed things up.

Contributing

Feature requests are welcome, bug reports are especially welcome, and pull requests are super welcome. Please make sure to use a feature branch in your fork. Don't look too hard for tests, there are none - not for a lack of belief in tests, but because the application is almost trivial. Please make sure you also update the README to reflect your changes. While you're there you may as well add yourself to the Credits section too :)

Credits

  • Roy Rapoport (@royrapoport) for inspiring this tool with his talk at the SF Metrics Meetup
  • Zoltan Nagy (@abesto): initial implementation, current maintainer
  • Ryan Bowlby (@rbowlby)
  • Bálint Csergő (@deathowl) for adding SQLAlchemy support
  • Charles Hooper (@chooper)

Awesome tools used

These tools made it possible to write changelog in a weekend. Huge thanks.

  • Flask, the lightweight Python web framework and Flask-RESTful for making the REST api trivial to write
  • Bootstrap, for making the whole UI look not terrible and bootstrap-datetimepicker for making the date chooser pretty awesome
  • jQuery, because obviously
  • jQuery BBQ and jQuery hashchange for making the permalink simple to implement
  • Sentry for simple-to-use, awesome error reporting / collection / aggregation
  • cdnjs for hosting all the client-side libraries above
  • SQLAlchemy
  • Flask-SQLAlchemy SQLAlchemy support for Flask
  • Those that we take for granted: sqlite3, python, virtualenv, and the list goes on...

Changelog

Image gleefully copied from Android Police