Git deployment rollback workflow

Question

More and more server-side file deployments are handled using git. It's nice and there are plenty of guides available how to setup your deployment workflow with git, rsync and others.

However, I'd like to ask what's the cleanest way to set deployment rollbacks, so that

• Every time you deploy, you record the latest state before the deployment (no need to manually read through logs to find the commit)

• What git commands use to rollback to the prior (recorded) state in the case deployment has unforeseen consequences

The scope of the question is Linux servers, shell scripting and command line git.

Answer 1

Note that there is no general solution to this problem. I would propose two solutions.

First one requires usage of Fabric and some deep thinking how to handle whole deployment process. For a Django site I maintain, I wrote a fabric script that deploys staging on every git commmit. Deploying from staging to production is then a simple fabric command that copies all the files to a new folder (increments a version by 1), for example from production/v55/ to production/v56/ (ok, it also does backups and runs migrations). If anything goes wrong, rollback command restores backups, and starts production environment from folder production/v55. Less talk, more code: https://github.com/kiberpipa/Intranet/blob/master/fabfile.py

The second option requires more reading and has a bigger learning curve, but also provides cleaner solution. As Lenin suggested to use framework with declarative configuration, I would propose to go a step further and learn a Linux distribution with declarative configuration - http://nixos.org/. NixOS has built in capabilities for distributed software deployment (including rollbacks) and also tools to deploy stuff from your machine https://github.com/NixOS/nixops. See also thesis on Distributed Software Deployment, which covers also your questions (being part of a much bigger problem): http://www.st.ewi.tudelft.nl/~sander/index.php/phdthesis

Answer 2

Please have a look at Capistrano and Chef which require ruby/ror support. But are great deployment tools. Python's fabric is also an awesome tool.