Skip to content
SYS.DOCS // DOCS

Cluster Settings: Access & Security

Use Cluster Settings → Access to control who can see a cluster, which actions they can perform, and how credentials are issued.

The tab has two lists:

  • People: everyone who can open the cluster or holds a kubeconfig for it, once each, with their access and their kubeconfig. Your own kubeconfig is on your row.
  • Root access: the cluster’s SSH key, which an organization owner can download once, and the admin kubeconfig while a downloaded copy is still valid.

Access changes in People wait until you save them.

Edka combines organization roles with cluster-level developer access.

  • Organization owners and admins always have full access to every cluster in the organization.
  • Developers only get access to clusters they own or clusters where they have been granted access explicitly.
  • A cluster owner is the user assigned to a specific cluster. This is separate from the organization-wide Owner role.
  • The cluster owner always keeps full write access to that cluster.
  • The cluster owner can set developers’ access to that cluster in People.
  • Cluster owner access cannot be downgraded or removed from the Access tab.

Cluster-level permissions apply only to users with the organization role developer.

Access levelWhat it means
No accessThe cluster is hidden from the developer in cluster lists, routes, and cluster-scoped features. Any active personal kubeconfig for that cluster is revoked.
ReadThe developer can open the cluster and use read only features, but write actions are blocked. Their personal kubeconfig is issued with Kubernetes view access.
WriteThe developer can manage that cluster in Edka for standard day-2 operations. Their personal kubeconfig gets Kubernetes view across the cluster and edit in the namespaces listed for them.

In practice, read is for observing and troubleshooting, while write is for day-2 operations such as updating non-sensitive cluster settings, managing workloads, and editing secrets.

  • Developers’ access and kubeconfig namespaces can be managed by the cluster owner and by organization owners/admins.
  • Other people’s kubeconfigs can only be rotated, revoked, or restored by organization owners/admins.
  • Sensitive cluster actions are limited to the cluster owner and organization owners/admins. This includes enabling or disabling cluster protection, archiving or unarchiving a cluster, deleting a cluster, and setting, updating, or removing the cluster-specific Hetzner token.
  • Changing a developer from read to write, write to read, or no access, or changing their namespaces, revokes the developer’s current personal kubeconfig once you save. They must download a new kubeconfig after access is restored or changed.

If a developer has write access but is not the cluster owner, they can still perform normal operational tasks, but they cannot:

  • Change Cluster Protection
  • Archive or unarchive the cluster
  • Delete the cluster
  • Set, update, or remove the cluster-specific Hetzner API Token

These controls are available in Settings > General and Settings > Danger Zone.

People lists each person once:

  • Person: their name and email. Your row is marked (you).
  • Access: a No access, Read, and Write switch for each developer. The cluster owner shows Owner, and organization owners and admins show Organization owner or Organization admin. Someone who left the organization and still holds a kubeconfig shows Left the organization.
  • Kubeconfig: the kubeconfig version and the day it was downloaded, or not downloaded. A revoked kubeconfig stays in the list with the day it was revoked. Hover a day to see the time.
  • Rotate and Revoke for a kubeconfig in use, and Restore for a revoked one.

What you see depends on your role:

  • Organization owners and admins see everyone, with their kubeconfigs and the Rotate, Revoke, and Restore actions.
  • A developer who owns the cluster sees everyone and can change their access, but sees only their own kubeconfig.
  • Other developers see only their own row.

Developers you invite to the organization appear in the list.

As an organization owner or admin, use the actions at the end of a person’s row:

  • Rotate ends their current kubeconfig at once. They download the new version from this tab.
  • Revoke ends their kubeconfig and blocks downloads until access is restored.
  • Restore gives them a new kubeconfig after a revoke. It appears only while they can still open the cluster.

Rotate and Revoke apply to the latest credential version for a user.

Your own kubeconfig is on your row in People. Click Download kubeconfig to get it. Once you have one, the button reads Rotate and download, which ends your current kubeconfig and downloads a new one.

Behavior and restrictions:

  • Kubeconfig download is available only when the cluster is active.
  • If your kubeconfig is revoked, your row says Revoked and downloads are blocked. An organization owner or admin restores it with Restore on your row, and owners and admins restore their own the same way.
  • Rotating invalidates your current kubeconfig immediately.
  • Your kubeconfig scope follows your effective access:
    • organization owners/admins receive Kubernetes cluster-admin
    • developers with write receive Kubernetes view, and edit in their namespaces
    • developers with read receive Kubernetes view

A developer with Write access, and the cluster owner, gets a kubeconfig with Kubernetes view across the cluster and edit in the namespaces listed on their row, after Kubeconfig edits in. Without a namespace, the kubeconfig can only read. A namespace the cluster doesn’t have yet is created when the developer downloads their kubeconfig.

To list a namespace:

  1. Click Add namespace on the developer’s row.
  2. Type a name. The field suggests the cluster’s namespaces, and marks a name the cluster doesn’t have yet as new.
  3. Pick a suggestion or press Enter. The field stays open for the next namespace, and Escape closes it.

To remove a namespace, click the x on it. A developer can have up to 50 namespaces.

System namespaces (kube-system, kube-public, kube-node-lease, edka-system, edka-builds, arc-systems, system-upgrade) and the namespaces of Edka add-ons, such as cert-manager or monitoring, can’t be listed, because edit there reaches beyond the namespace.

Access and namespace changes are drafts until you save them. A changed row shows an orange mark. A bar at the bottom of the tab counts the unsaved changes, such as 2 unsaved changes, and holds Discard and Save. The bar appears only when something changed.

Saving ends the current kubeconfig of everyone whose access changes. Before you save, their Kubeconfig shows ends on save.

Every cluster has two credentials that Kubernetes RBAC can’t revoke:

  • the admin kubeconfig, a system:masters certificate that grants cluster-admin
  • the SSH key, which grants root on every node

Each stays valid until the cluster CA or the SSH key is rotated, so they are break-glass credentials:

  • Only organization owners can download them. Admins and developers use their personal kubeconfig.
  • Each download asks for a passkey, whether or not the organization requires step-up verification.
  • Each one downloads once.
  • An impersonation session can’t download them.
  • From the CLI, the sign-in needs write access.

The Root access list holds both. The SSH key row shows Download SSH key to owners until the key is downloaded. After the download, the row shows who downloaded the key and when, and a command to copy:

Terminal window
ssh -i ssh-key-<cluster-name>.pem root@<node-ip>

The Admin kubeconfig row appears once an owner has downloaded the admin kubeconfig, and stays until the cluster CA is rotated. It shows who downloaded it and when. The cluster overview marks a downloaded admin kubeconfig too.

Rotating the cluster CA ends every admin kubeconfig downloaded before it. Edka replaces the cluster’s certificate authorities on every node, and a certificate the previous ones signed no longer authenticates. A copy of the admin kubeconfig that left Edka gets Unauthorized from then on. The SSH key stays as it is.

  1. On the Access tab, find the Admin kubeconfig row under Root access. The row appears once an owner has downloaded the admin kubeconfig.
  2. Click Rotate CA, then Rotate CA again in the Rotate the cluster CA dialog.
  3. Verify with your passkey. A rotation asks for one every time.

The cluster shows Updating while the rotation runs. Every node restarts K3s twice, the masters one at a time and then the workers, and pods keep running. Once the cluster is Active again, the Admin kubeconfig row is gone, and an owner can download the new admin kubeconfig once.

Only owners can rotate the CA, on an active cluster. Satellite clusters have no CA in Edka to rotate. During an OS update round, rotate once the round finishes or after you cancel it under Settings → Maintenance.

  • A personal kubeconfig downloaded before the rotation keeps working with kubectl and other Go clients for one rotation. After a second rotation, it needs a new download.
  • Clients that verify certificates with OpenSSL, such as the Python Kubernetes client, Node.js tools, or curl, stop trusting the API server with a kubeconfig from before the rotation. Rotate your kubeconfig to get one with the new CA.
  • Kubernetes updates the kube-root-ca.crt ConfigMap in every namespace. A workload that read it at start and verifies with OpenSSL can’t verify the API server until it restarts. Workloads on Go clients keep working.
  • When Edka can’t reach every master and ready worker over SSH, the rotation stops before it changes anything, and the cluster’s activity says nothing was rotated.
  • A rotation that fails partway leaves the cluster on CAs every node trusts. Click Rotate CA again to finish it with the same CAs.
  • A worker that is offline for the whole rotation keeps a join token for the previous CA and can’t rejoin the cluster. Replace that worker.

Edka reaches a cluster it created with tokens it signs for that cluster. The Kubernetes API server on each control plane node accepts them because of Edka’s entry in its authentication config, /var/lib/rancher/k3s/server/edka/authentication-config.yaml. To cut Edka off, remove the entry on every control plane node:

Terminal window
cat > /var/lib/rancher/k3s/server/edka/authentication-config.yaml <<'EOF'
apiVersion: apiserver.config.k8s.io/v1
kind: AuthenticationConfiguration
jwt: []
EOF

The API server reloads the file by itself, without a restart. Once the cluster has refused Edka’s tokens for 10 minutes while they work on other clusters, Edka takes its access as removed:

  • Edka stops acting on the cluster: no deploys, add-on changes, console actions, node operations or OS update rounds. It uses no other credential in their place and does not put its entry back, and control plane nodes added later start without it.
  • The cluster stays in your account. Edka never marks it disconnected or removes it, and deleting it from Edka still works.
  • The cluster’s owner gets an email, and the cluster page shows Edka’s access removed.

To give Edka its access back, open the cluster and choose Restore access. Edka writes its entry back on each control plane node, one at a time, and restarts K3s on a node only when the drop-in that loads the config is gone.

Removing the entry leaves two credentials in place. Edka does not use them while its access is removed, but they keep working until you remove them too:

  • The cluster’s SSH key. Edka created the nodes with it. Print its public half with ssh-keygen -y -f ssh-key-<cluster-name>.pem and delete that line from /root/.ssh/authorized_keys on every node.
  • The admin certificate. Edka holds the kubeconfig K3s wrote at install. Rotating the cluster’s certificate authority makes it stop working; follow rotating the K3s CA.

On an archived cluster, the Access tab disables access and namespace changes, kubeconfig rotation, Revoke and Restore, the SSH key download, and CA rotation.