Multi-tenancy and tiered access in Data
How BullSequana AI isolates Data workspaces and applies viewer, editor, administrator, and pipeline roles.
BullSequana AI provides tenant isolation and team-scoped access throughout Data. In Developer Workspace, each Kubeflow profile maps to an isolated workspace namespace, and the platform creates distinct identities for human access and pipeline execution.
Tenant and team boundaries
A tenant represents one customer or organization. Teams organize departments or working groups inside that tenant. Workspace membership, notebooks, pipelines, artifacts, and service access follow this hierarchy.
Users discover and operate only the resources allowed by their tenant, team, and assigned role. Provider-level administration remains separate from tenant and workspace administration.
A tenant's Data services are its Lakekeeper project and default warehouse, its Data authorization store and metastore integration, and Compute. The platform operator sets them up for a tenant in the platform configuration. A tenant created in the Portal does not get them automatically; see Tenant Data services. Until they are set up, the tenant's profile namespaces carry no metastore project, and Compute lists no namespace for the tenant's members.
Managed-ingestion and catalog access
Data ingestion in Studio (access and sharing) uses tenant-scoped functional permissions plus grants on connections and ingestions. Data Consumer, Data Builder and Data Steward roles cover common tasks; credential replacement is separate. Ownership and viewer/editor/invoker/operator/grant-manager relationships determine access to individual resources. These are not the Kubeflow namespace tiers below, and platform or tenant administration does not grant unrestricted access to private ingestion resources.
Table discovery in the data catalog needs the data.catalog.read permission in the active tenant. Data Consumer grants it, and so do Data Builder and Data Steward, which include Data Consumer. The authorization service then limits the result to the tables the caller can discover.
Workspace access tiers
BullSequana AI replaces Kubeflow's single collaborator role with four identities in every profile namespace:
| Access tier | Service account | Product behavior |
|---|---|---|
| Viewer | default-viewer | Reads notebooks, pipelines, and logs without changing workspace resources. |
| Editor | default-editor | Creates notebooks and runs pipelines without unrestricted namespace administration. |
| Administrator | default-admin | Manages the workspace namespace and its membership-bound resources. |
| Pipeline runner | pipeline-runner | Executes Argo Workflows independently of any human user's access tier. |
This tiered model gives workspace owners a least-privilege alternative to granting every collaborator the same permissions.
Automatic policy enforcement
The platform generates the required ServiceAccounts and RoleBindings whenever a Kubeflow profile namespace is created. Kyverno policies reconcile the access tiers automatically and also apply them to existing profile namespaces.
Administrators assign users to the appropriate tier; they do not create Kubernetes identities or bindings by hand. The generated resources remain aligned with the profile namespace throughout its lifecycle.
Spark workload identity
In a profile namespace linked to the tenant's Data services, each member with the editor or administrator tier gets a personal Spark ServiceAccount, named u- followed by their Keycloak user ID. The profile bucket provisioner creates it and registers it in Lakekeeper with that user's own catalog grants. Spark clusters and Compute jobs run their pods as this identity, so the Iceberg catalog authorizes each user's Spark work as that user. Admission policies stop a pod from running as another user's identity.
Creating a Spark cluster or job needs the editor or administrator tier of the profile namespace as well as a Spark role. Members with only the viewer tier get no Spark identity and cannot create Spark workloads. See Run Spark workloads.
Separate pipeline identity
Pipeline runs use pipeline-runner rather than inheriting the permissions of the person who submitted the run. This separation keeps automated execution stable when team membership changes and prevents a pipeline from receiving administrator access simply because its author has that role.
The same identity is used consistently by the Kubeflow Pipelines and Argo Workflows integration.
What administrators control
Tenant and workspace administrators control:
- which users belong to the tenant and team
- which workspace profile they can access
- whether their role is viewer, editor, or administrator
- which data, secrets, and compute resources the workspace can use
BullSequana AI creates and enforces the corresponding workspace identities and Kubernetes permissions.
A Data platform user belongs to exactly one tenant. A person who works for several tenants has a separate user in each tenant; see Manage members.