Cloud Platform Translators

One idea from Gregor Hohpe’s Platform Strategy reframed platforms for me: a cloud platform is a translator. Developers say what they want in the platform’s language; the platform turns it into the cloud’s language. How it translates decides how much it hides and how much it can do.

Three kinds of translation

• Composition. Bundle several low-level resources into one higher-level construct. Infrastructure-as-code kits like the AWS CDK do this: one construct gives you a container cluster and a load balancer in front of it.
• Lookup. A logical resource becomes a physical one depending on context. “A relational database” turns into a managed database on whichever cloud and environment you’re deploying to. Platform orchestrators work this way.
• Model transformation. Expose a different model altogether, such as a data flow, and generate the cloud resources from it. This is the closest a platform gets to being a compiler, and the most powerful and the hardest to get right.

Declarative provisioning

Modern platforms are declarative: you describe the desired state rather than a sequence of commands. Deploying is then two steps: work out the difference between desired and actual state, then make the changes that close it.

That runs as a loop:

1. Observe changes to the specification and the environment.
2. Analyse them against the platform’s constraints and work out the actions.
3. Act, then go back to observing.

It’s the same loop Kubernetes controllers run, and once you see it you find it everywhere, from self-healing laptops to balancing in a handstand.

Related: Internal Developer Platform · Golden Paths · Platform Strategy