Daniel Doubrovkine bio photo

Daniel Doubrovkine

aka dB., @awscloud, former CTO @artsy, +@vestris, NYC

Email Twitter LinkedIn Github Strava
Creative Commons License


We’ve switched from CIJoe to Jenkins. I guess we’ve “graduated” to Jenkins, but in retrospect we should have just used that in the first place. I think CIJoe is fine, we did thousands of builds with it, thank you CIJoe. But it’s actually easier to setup Jenkins and it’s much lighter than what I anticipated. As someone said, there’s no need to have CI written in Ruby to build Ruby projects.

This is what I see after spending a day with Jenkins.

  1. A lot less manual configuration, running as a service out-of-the-box.
  2. Multiple projects without multiple services on different ports.
  3. Console output during build with live progress update.
  4. Built-in support for multiple build nodes.

Jenkins is a lot closer to my BuildTracker project at Microsoft – we were doing all the things Jenkins is doing, ten years ago, with build labs that had hundreds of servers. Buildtracker is a “build lab out of the box” and I hear that it’s live and well and widely used in Redmond. Maybe I’ll contribute to Jenkins some plugins around features we built then.

Long story …

I couldn’t get CIJoe to use RVM and went down the rabbit hole of a global Ruby installation and running CIJoe as root. I managed to get CIJoe to auto-start on boot using a script from this post, but I was still having trouble getting repeatable builds. I am sure all these issues are resolvable, but it smelled wrong to me, especially since RVM is built to fix most of those problems. After fighting it for half a day I decided to give JenkinsCI a try. An apt-get later I had it running. From the start Jenkins runs as a user, jenkins, which can have a regular RVM setup without any headaches – nice!

We used to commit CIJoe runner configuration into Git (like this). It doesn’t look quite possible with Jenkins unless you commit the entire build script. I think we’ll move the remaining steps into a Rake task and execute that with a Jenkins Rake plugin.

Typing too fast is bad for you. When adding a command-line task to Jenkins on *nix, make sure you choose Execute shell and not Execute Windows batch command. The latter is the first choice. I had at least three face-to-palm moments.

Jenkins eats shell commands. You can’t do loops, if-s or other bash-like constructs that you might be used to. For example, I was trying to ignore failures on adding a remote in a build step, unsuccessfully. I tried command; exit 0 and all kinds of other tricks at no avail. You have to keep things simple – in my case I ended up changing a git push command from git push heroku master (that’s why I needed to add a heroku remote) to git push git@heroku.com:app-name.git.

Jenkins with Git and Github tags and builds on detached states (if you do git status on the workspace you’ll find yourself on no branch). Hence you can’t push anything remotely from here. Our builds deploy to staging, so I had to change the build to push the currently built git commit.

git push git@heroku.com:app-name.git $GIT_COMMIT:master

My wish list for Jenkins today includes the following. I am sure it will grow. If you want to code some open-source stuff, I’ll be happy to help you in any way I can in exchange.

  • Colorize ANSI console output
  • Let me setup a project where I can choose which revision to build. This could make functionality similar to heroku-bartender.

I am Jenkins noob, so looking forward to your comments.