Wednesday, May 28, 2014

Continuous Integration that isn't

I have seen people who run Jenkins and claim that they have realized CI.

Actually, one of those people was me, a couple years down the line.
Before you ask, no, it wasn't in an agile team.

Configuring a Jenkins is easy, and getting the Jenkins to pull a repo, create a build and deploy it on a server also is. But that is not Continuous Integration!

So, here is my story:
We were in a customer project and pretty much nobody had heard of CI, only one guy had an idea "Why should we manually deliver software to the testers? There are tools out there, let's do CI"

And so we did.
The Jenkins was up and running. Whenever the team manager pressed the button, the software got deployed to the Test Environment and the downtime for a new deployment was reduced from a couple hours to less than 5 minutes. As the testers knew when the "Deploy" button was being pressed, usually Friday EOB, test was absolutely not affected by any downtimes.
A big benefit!

However, usually the first thing happening after the deployment - something didn't work.
Like, for instance, the web server. Or the Messaging Queue. Or the database. Or the business processes. Or anything else that used to work in the past.

Lesson learned

Continuous Integration is so much more than automating the build/deployment chain and reducing outages to a couple minutes.
CI shouldn't result in outages in the first place. You can use techniques like Parallel Deployment to attain Zero Downtime for patches.

Also, you haven't understood CI until you have tons of other techniques in place.
CI that works on a button press is missing the point: CI should be continuous, not scheduled and manual.

  • If you are delivering dozens of new features with each build, your CI has a very slim chance of locating the error source. Make sure your CI is integrated in a way that each feature has at least 1 build on it's own.
  • If you don't have unit test coverage, CI isn't even worth being called such. Move towards high unit test coverage before bothering with CI.
  • If you don't have automated regression and smoke tests, CI is more likely to cause harm than help. Invest into test coverage and link the automated tests to the CI server. 
  • If you don't have a rapid feedback cycle into your development process, CI has no benefit. Make sure the developer who committed the failing build gets informed and acts within minutes.
  • If you aren't acting immediately on failed builds or errors in the deployment, that's not CI, it's a mess!
    STOP and fix if the tests fail. Don't proceed to code further based on harmful code!
  • If you are spending a full week on manual integration tests, you may have a CI tool, but you don't have CI!
    Create automated integration tests that can be run as part of the CI process. If you can't eliminate manual components, rethink your approach!

CI isn't about having a tool - it's about having the processes and engineering practices that allow you to deliver rapidly and often!
Real CI comes with a mindset "Let's rather have one too many than one too few deployments".

No comments:

Post a Comment