Skip to content

Access

The workstation

A member of a namespace (Getting started) logs in with their identity, through the cluster's identity broker (on auth. and the cluster's zone): a security key (a passkey) where the cluster offers one, another provider where its admins added it. The cluster serves a script that prepares a workstation: it installs kubectl (the cluster's version, from dl.k8s.io, its checksum verified) and kubelogin (kubectl's OpenID Connect plugin, https://github.com/int128/kubelogin), adds the cluster to the kubeconfig and makes it the current context, then signs in.

curl -fsS https://cloud.dampetra.com/setup.sh | sh

Written by the cluster for itself: it asks nothing. Linux or macOS (x86_64 or ARM), with curl and python3. At the end a browser opens on the identity provider; once signed in, the script shows the identity and groups the cluster sees (kubectl auth whoami). Run it again any time, e.g. after the cluster changed: it touches nothing else of the kubeconfig.

Options go after sh -s -- (curl -fsS https://cloud.dampetra.com/setup.sh | sh -s -- --skip-login-check):

Option
--proxy <url> the HTTP(S) proxy reaching the cluster and its identity provider (in the kubeconfig too); curl needs it from the first command (HTTPS_PROXY, curl -x)
--no-proxy <hosts> hosts reached without the proxy while setting up
--kubeconfig <file> the kubeconfig to configure (default: kubectl's, $KUBECONFIG or ~/.kube/config)
--kubectl-path <file> where kubectl is, or goes (default ~/.local/bin/kubectl); kubelogin goes beside it
--kubelogin-arg <argument> an argument for kubelogin, repeatable: --grant-type=device-code without a browser (an agent, a server: a code to enter on another device)
--skip-login-check no sign-in at the end
--help the options

A kubectl already there may be one minor version away from the cluster's at most (kubectl's skew policy): the script stops otherwise, and says what to do. The kubeconfig alone, without the script: https://cloud.dampetra.com/kubeconfig (the user runs kubectl oidc-login, which needs kubelogin installed as kubectl-oidc_login). Nothing secret in either.

The cluster knows a member as oidc: and their email (the request for a namespace names that email); their groups, when the provider gives some, as oidc: and the group. A member has the rights of their role in their namespaces only (admin: everything there but the quota and the namespace itself, roles and bindings included; deployer: the workloads), and none elsewhere. kubectl auth whoami shows what the cluster sees.

What being authenticated adds

These pages are public: they say how the cluster works, not what runs in it. Through the API, each reader sees what their rights allow:

  • Every authenticated user: these pages, kubectl -n kube-public get configmap ethere-docs -o jsonpath='{.data.index\.md}' (each page is a key: network.md, llms.txt…).
  • A namespace's members: its own page, kubectl -n my-app get configmap ethere-docs: its allowed names, its quota, its members, the Flux objects to start from.
  • Secrets are never in a page: a page names the Secret holding one, which only those allowed read.