Paul Fawkesley

Automating our Heroku deployments from Jenkins

At Sea Level Research we deploy to Heroku staging then production from Jenkins, with automatic database backups & migrations. Read on to find out how.

Last week we released an early version of an API that we’re hosting with Heroku. We’re an early-stage startup and this is a component of our first product. At this point in our startup life, Heroku is a great fit as we need to focus on features, not infrastructure. We are targeting corporate customers with a pretty niche product, so we aren’t likely to suffer from massive overnight scaling issues.

Automating Heroku

I love Heroku, but I always find the command line deployment stuff a bit fiddly. How do I turn on maintenance mode again? How do I backup the database before I run the migrations? And given that I’m fanatical about automation - especially when it comes to deployment - it made sense to tackle our process from day one. If you don’t believe in automating deployment you might get some amusement reading Knightmare: a devops cautionary tale

My wishlist of requirements went like this:

One simple shell script

It turns out this is reasonably straightforward using Heroku’s command line client. In an afternoon I wrote up the following shell script which fulfills most of the above:

https://github.com/sealevelresearch/sea-level-api/blob/master/scripts/jenkins_deploy_to_heroku.sh

The most interesting bit is probably the use of the pgbackups Heroku add-on. During a staging deployment, the script instructs the backups module to backup production, then get a URL to its backup. It resets the staging database - rather than just overwriting rows - to ensure the schemas exactly match, then restores the production backup into staging. This means we have confidence that the subsequent database migrations are running against the actual database.

Configuring Jenkins

It turns out quite a large chunk of the work came about from configuring our Jenkins box to have the necessary tools, like the heroku command line client. For this we use the excellent Ansible, and we’ve had to make a fair few lightweight modules now such as ansible-heroku-toolbelt as ansible-jenkins.

None of this is rocket science, but I do believe that taking the time to implement this clean, conservative & repeatable processes will be worth every penny six months down the line when I’m trying to push out a 2am bugfix for a major client…

Paul is CTO of Sea Level Research, a startup helping shipping ports to optimize their logistics.


Thoughts? Get in touch