Winning at Continuous Deployment with the Expand / Contract Pattern

Live demo of zero-downtime deployments.



You've done everything right: you have a continuous deployment pipeline, you have a commit stage with great unit tests, a pre-prod stage with fully automated deployment, robust end-to-end tests as well as some performance & security checks, and once all your quality gates are passed, your stream of changes march confidently towards production - but then somehow, someway, somewhy, production still blows up!

Continuous delivery is awesome, but the devil is in the details, and there's a decent chance you've been bitten by one of the biggest challenges - the safe management of contract changes.

Contracts can be obvious and public such as APIs (URL structure & schema), or obvious and private, such as database schemas, or non-obvious, such as the schema of in-flight messages in a queue, or dependency on specific timing expectations.

This talk demonstrates with real deployments the fundamental principle that underlies all safe contract changes - the 3 step model:

  1. expand - introduce the change in a backwards and forwards compatible way
  2. change - ensure all instances of the component and all dependants are using the new definition
  3. contract - retire the obsolete compatibility elements

The good news is, with continuous delivery we can often do all three steps in quick succession!

Shared At