Skip to content

Getting started

From an application that has never run here, new or with manifests written for another cluster, to the same application served on its name. Each step names the page with the details.

1. Read the choices

Ethere's choices lists what this cluster decides that another leaves open. An application with existing manifests usually changes three things: its Ingress becomes an HTTPRoute, its Services and servers move to IPv6, and it declares its network policies.

2. Ask for a namespace

Namespaces are granted by the cluster's admins, in the cluster's declaration: a git repository they review. The request is an entry of its tenants/values.yaml:

tenants:
  my-app:                                   # the tenant: a project
    namespaces:
      my-app: {hostnames: [my-app.cloud.example.com]}   # each namespace, the names its routes may take
      my-app-staging: {hostnames: [staging.my-app.cloud.example.com]}
    members:
      - {user: alice@example.com, role: admin, namespaces: [my-app, my-app-staging]}
      - {serviceAccount: ci, role: deployer, namespaces: [my-app-staging]}
  • Names: under the cluster's zone (here cloud.example.com: this cluster's is on its own page), or under a domain of the project's own: its DNS is then delegated to the cluster, which the admins set up (a *. entry allows every name under it).
  • Members: a person by the identity they log in with (Access), role: admin (everything in the namespace) or role: deployer (the workloads, not the roles); a CI as a service account of the namespace.
  • Quotas: the admins add a quota per namespace (a ResourceQuota's hard).

Once the change is merged, the cluster applies it within minutes: the namespaces, their quotas, the members' bindings, a service account flux, and a page for each namespace (kubectl -n <namespace> get configmap ethere-docs). Removing a namespace from the declaration never deletes its data.

3. Get access

A kubeconfig, then kubectl -n my-app get pods (Access).

4. Deploy

  1. Images: built by the project, pushed to a registry the cluster can pull from (Images).
  2. A chart (or plain manifests) for the application: a Deployment, an IPv6 Service, an HTTPRoute on one of the namespace's names (Exposing), its network policies (Network), volumes if it keeps data (Storage).
  3. Flux objects in the namespace: a GitRepository (with its deploy key) and a HelmRelease naming the service account flux (Deploying).

From then on, a commit to the application's repository is a deployment.

5. Keep a note in the application's repository

A few lines in its AGENTS.md (or CLAUDE.md): this cluster's documentation (its llms.txt), its namespaces, and its own commands (how it builds and pushes its images, which file holds the deployed version).