Five copies of the same truth
In an organisation with no registry, an application's address is written in the server configuration, in DNS, in the monitoring system, in a spreadsheet and in somebody's head. The day it changes, three of the five places get updated. The other two fail weeks later, with nobody knowing why.
The registry as single source
The alternative is a single registry, readable by machines and by people, holding what everybody needs to know about each application:
- Its name, its owner and the phase it is in.
- Where it is deployed in each environment.
- Which ports it uses, allocated without collisions.
- What addresses it is published at.
- What secrets it needs, by name, never by value.
Everything else is derived from the registry: the deployment configuration, the network routes, what the monitoring system watches, the documentation and the cost allocation. Nobody copies a value by hand into somewhere else.
How it stays healthy
- It is edited as code: in a repository, with review, with history. Every change has an author and a reason.
- It is validated automatically: duplicate ports, duplicate addresses or applications with no owner fail validation before reaching any environment.
- It has a single writer per kind of data: whoever registers an application is the one who records it; the processes only read.
- It is small and stable: data that changes rarely. What changes every minute (metrics, live status) is not kept here.
A registry like that says what secrets each application needs. What it must never say is their value. That has its own place.
How we apply it: Platform Engineering.