Restrict the ClickHouse operator to a named set of namespaces instead of granting it permissions across the whole cluster.
By default the operator is granted its permissions through a single ClusterRole, so it can act on the resources it manages in every namespace. That is a poor fit when the operator runs in a cluster you share with other workloads. Namespace scoping grants most of those permissions through a Role in each namespace you nominate instead.
Prerequisites
- Follow the deployment tutorial for your platform until you are ready to deploy the operator.
- The namespaces your ClickHouse clusters will live in, decided up front (this guide uses
ns-default-xx-01andns-default-xx-02) - Permission to install or upgrade the operator Helm release
Enabling Namespace Scoping in the Operator
Two values control this:
operator.watchNamespaceslists the namespaces the operator watches. AClickHouseClusterin any other namespace is never reconciled.operator.rbac.namespaceScopedgrants the operator’s permissions on namespaced resources through aRolein each of those namespaces, instead of one cluster-wideClusterRole.
Set both. The operator still needs a small ClusterRole, because some of the resources it reads are not namespaced:
| Scope | Resources |
|---|---|
| Cluster-wide | namespaces, persistentvolumes, and the licenses custom resource |
| Each watched namespace | statefulsets, pods, services, configmaps, secrets, persistentvolumeclaims, poddisruptionbudgets, events, and the namespaced ClickHouse custom resources |
Operator namespace, by default clickhouse-operator-system |
the same namespaced set, because the operator reads its license secret there |
The operator always watches its own namespace in addition to the list you supply, so you do not need to include it yourself.
Steps
Decide Which Namespaces the Operator Watches
Each ClickHouse cluster needs its own namespace, so note down every namespace you intend to deploy into.
Adding a cluster later means adding its namespace to the list and upgrading the operator release first. Plan for the namespaces you expect to need, including any you will use for compute-compute separation.
Enable Namespace-Scoped RBAC
Add both values to your operator values file, operator-values.yaml:
operator:
watchNamespaces:
- ns-default-xx-01
- ns-default-xx-02
rbac:
namespaceScoped: trueApply them by upgrading the operator release, keeping your existing values:
helm upgrade clickhouse-operator \
oci://<your-registry>/helm/clickhouse-operator-helm \
--version <version> \
--namespace clickhouse-operator-system \
--values operator-values.yamlThe namespaces must exist before the operator release is upgraded, because the chart creates a Role and RoleBinding in each one.
Verify the Permissions
Confirm the operator is watching the namespaces you listed. Its own namespace appears alongside them:
kubectl -n clickhouse-operator-system logs deploy/clickhouse-operator-clickhouse-operator-helm | grep "informer cache is scoped"Check that a Role exists in each watched namespace:
kubectl -n ns-default-xx-01 get role,rolebindingConfirm what the operator can do, which resolves every binding that applies to it:
kubectl auth can-i --list -n ns-default-xx-01 \
--as=system:serviceaccount:clickhouse-operator-system:clickhouse-operator-clickhouse-operator-helmRunning the same command against a namespace you did not nominate should show only the cluster-wide resources from the table above.
Supplying Your Own ServiceAccount or RBAC
If your organisation provisions identities and permissions outside the chart, you can take over either part.
To run the operator as a ServiceAccount you manage, create it first and name it. A name is required, otherwise the chart would bind to an account it did not create:
operator:
serviceAccount:
create: false
name: my-operator-saTo manage the operator’s permissions yourself, stop the chart creating them:
operator:
rbac:
create: falseFor a worked example, see Minimal Operator RBAC.
Limitations
- The watch list is fixed at deploy time. Adding a namespace requires an operator release upgrade, so it is not suitable for namespaces created on demand.
- The operator still needs the cluster-wide permissions listed above. A deployment with no cluster-scoped permissions at all is currently not supported.