Configs

A config is the reusable, versioned set of settings a workload runs with — kept separate from the image so the same build can run many ways.

What a config holds

env — environment variables, literal or secret references.
resources — requests and limits (250m CPU, 256Mi memory).
scaling — a fixed replica count, or autoscaling between a floor and a ceiling.
routes — how the component is exposed (routes & ingress).
secret refs — pointers to secrets, never the values themselves.

Layering & reuse

Configs are versioned and reusable across stages. A stage resolves an ordered chain of configs — a shared base plus per-stage overrides — where later layers win. So prod inherits everything from your base and changes only what it must, and the diff between environments is exactly that override, nothing more.

baseprod override

components:
  web:
    resources: { cpuRequest: 250m, memRequest: 256Mi }
    scaling: { mode: static, replicas: 2 }
    env: { LOG_LEVEL: info }

components:
  web:
    scaling: { mode: auto, replicas: 3, max: 10 }
    env: { LOG_LEVEL: warn }

Resolved for prod, web runs 3–10 replicas at LOG_LEVEL=warn, keeping the base's resource requests — no duplication, no drift.