What the Kubernetes connector is

What it does, and why your cluster asks rather than orcakey.sh reaching in.

The connector is a small operator you run in your own cluster. You give it a service token and the name of a Kubernetes Secret, and it keeps that Secret filled with every key in the token's environment. It checks again every 60 seconds, so a key you add a week later reaches the Secret with no extra step.

It is shaped like Doppler's operator: one resource, one token, one Secret.

apiVersion: secrets.orcakey.sh/v1beta1
kind: OrcaKeySecret
metadata:
  name: checkout-prod
  namespace: orcakey-operator
spec:
  tokenSecret:
    name: checkout-prod-token
  managedSecret:
    name: checkout-env
    namespace: checkout

See Projecting an environment for the walkthrough.

The token decides the environment

A service token is scoped to exactly one environment, so the resource does not name one. The operator asks orcakey.sh who the token belongs to and reads that environment. A second environment means a second token and a second OrcaKeySecret.

Your cluster asks; orcakey.sh never reaches in

Most tools in this category take credentials for your cluster and push values into it. This one runs beside your workloads and asks for what it needs.

orcakey.sh stores no kubeconfig, holds no credential for your cluster, and needs no network path to your API server. The only credential involved is a service token at read access, and revoking it stops the flow on the operator’s next request.

The first fetch makes a column, holding every key

The first time an OrcaKeySecret fetches, the environment's page in orcakey.sh shows that namespace and Secret as a column, holding every key in the environment, with new keys arriving there too — what a resource with no scope has always received. Untick a key in that column and the next fetch stops sending it; the YAML never lists keys. This needs the connector at 1.1.0 or later. An older operator keeps syncing exactly as it does today, but its Secret has no column on the page — the page says so and names the version to upgrade to.

A Kubernetes Secret is a map, not a single value. An environment with DATABASE_URL, REDIS_URL and STRIPE_KEY becomes one Secret with those three entries, which a pod reads in one go:

envFrom:
  - secretRef:
      name: checkout-env

Add a key in orcakey.sh and, if the column still receives new keys, it appears. Remove a key and it goes, from every column that held it.

The Secret is a copy

Nothing in your cluster can change a secret in orcakey.sh. The operator has no write operation at all.

The operator owns the whole Secret. When the environment changes it overwrites the Secret with exactly the environment’s keys, so an edit made with kubectl lasts only until the next change in orcakey.sh. Change the value in orcakey.sh instead.

Idle costs nothing

Every 60 seconds the operator compares key names and change times, not values. It reads values only when something changed, so an environment nobody touches is read once and then left alone, and your audit trail records secrets that moved rather than timers that fired.