Setup Multi User

SUSE® Rancher Prime Continuous Delivery uses Kubernetes RBAC where possible.

One addition on top of RBAC is the GitRepoRestriction resource, which can be used to control GitRepo resources in a namespace.

A multi-user fleet setup looks like this:

  • tenants don’t share namespaces, each tenant has one or more namespaces on the upstream cluster, where they can create GitRepo resources

  • tenants can’t deploy cluster wide resources and are limited to a set of namespaces on downstream clusters

  • clusters are in a separate namespace

Shared Clusters
important information

The isolation of tenants is not complete and relies on Kubernetes RBAC to be set up correctly. Without manual setup from an operator tenants can still deploy cluster wide resources. Even with the available SUSE® Rancher Prime Continuous Delivery restrictions, users are only restricted to namespaces, but namespaces don’t provide much isolation on their own. E.g. they can still consume as many resources as they like.

However, the existing SUSE® Rancher Prime Continuous Delivery restrictions allow users to share clusters, and deploy resources without conflicts.

Example User

This would create a user 'fleetuser', who can only manage GitRepo resources in the 'project1' namespace.

kubectl create serviceaccount fleetuser
kubectl create namespace project1
kubectl create -n project1 role fleetuser --verb=get --verb=list --verb=create --verb=delete --resource=gitrepos.fleet.cattle.io
kubectl create -n project1 rolebinding fleetuser --serviceaccount=default:fleetuser --role=fleetuser

If we want to give access to multiple namespaces, we can use a single cluster role with two role bindings:

kubectl create clusterrole fleetuser --verb=get --verb=list --verb=create --verb=delete --resource=gitrepos.fleet.cattle.io
kubectl create -n project1 rolebinding fleetuser --serviceaccount=default:fleetuser --clusterrole=fleetuser
kubectl create -n project2 rolebinding fleetuser --serviceaccount=default:fleetuser --clusterrole=fleetuser

This makes sure, tenants can’t interfere with GitRepo resources from other tenants, since they don’t have access to their namespaces.

Allow Access to Clusters

This assumes all GitRepos created by 'fleetuser' have the team: one label. Different labels could be used, to select different cluster namespaces.

In each of the user’s namespaces, as an admin create a BundleNamespaceMapping.

kind: BundleNamespaceMapping
apiVersion: fleet.cattle.io/v1alpha1
metadata:
  name: mapping
  namespace: project1

# Bundles to match by label.
# The labels are defined in the fleet.yaml # labels field or from the
# GitRepo metadata.labels field
bundleSelector:
  matchLabels:
    team: one
    # or target one repo
    #fleet.cattle.io/repo-name: simpleapp

# Namespaces, containing clusters, to match by label
namespaceSelector:
  matchLabels:
    kubernetes.io/metadata.name: fleet-default
    # the label is on the namespace
    #workspace: prod

The target section in the GitRepo resource can be used to deploy only to a subset of the matched clusters.

Restricting Access to Downstream Clusters

Admins can further restrict tenants by creating a GitRepoRestriction in each of their namespaces.

kind: GitRepoRestriction
apiVersion: fleet.cattle.io/v1alpha1
metadata:
  name: restriction
  namespace: project1

allowedTargetNamespaces:
  - project1simpleapp

This will deny the creation of cluster wide resources, which may interfere with other tenants and limit the deployment to the 'project1simpleapp' namespace.

An Example GitRepo Resource

A GitRepo resource created by a tenant, without admin access could look like this:

kind: GitRepo
apiVersion: fleet.cattle.io/v1alpha1
metadata:
  name: simpleapp
  namespace: project1
  labels:
    team: one

spec:
  repo: https://github.com/rancher/fleet-examples
  paths:
  - bundle-diffs

  targetNamespace: project1simpleapp

  # do not match the upstream/local cluster, won't work
  targets:
  - name: dev
    clusterSelector:
      matchLabels:
        env: dev

This includes the team: one label and and the required targetNamespace.

Together with the previous BundleNamespaceMapping it would target all clusters with a env: dev label in the 'fleet-default' namespace.

BundleNamespaceMappings do not work with local clusters, so make sure not to target them.