Access Tiers
Each manifest that grants human access uses the same three-tier permissions block. This page states what each tier can and cannot do in Google Cloud.
| Grant | readviewers | devcontributors | adminadministrators |
|---|---|---|---|
| Infrastructure control-plane read | Yes | Yes | Yes |
| Spanner, Cloud SQL and BigQuery control-plane read | Yes | Yes | Yes |
| Cloud Storage object read (get, list) | Yes | Yes | Yes |
| Spanner data read (select, read) | Yes | Yes | Yes |
| BigQuery data read (query) | Yes | Yes | Yes |
| Identity Platform user read | Yes | Yes | Yes |
| Cloud Storage object create (add only) | No | Yes | Yes |
| Logging, mesh and edge configuration read | No | Yes | Yes |
| Cloud Storage object overwrite and delete | No | No | Yes |
| Spanner DML (insert, update, delete) | No | No | Yes |
| Cloud SQL connect and IAM login | No | No | Yes |
| BigQuery load and insert | No | No | Yes |
| Identity Platform user create, update, delete | No | No | Yes |
| Schema or DDL changes, anywhereno tier holds it | No | No | No |
| Secret payloads, IAM policy, impersonationno tier holds it | No | No | No |
- The tiers are cumulative: each tier has all the grants of the tier before it.
- The control plane is read-only in all three tiers.
- A person in more than one tier gets the widest role once.
- Grants apply per project, not on the whole tenant folder.
- Identity-Aware Proxy is an option for customer-facing traffic.
The tiers are cumulative. They differ on the data plane only. The control plane is read-only in all three tiers. Infrastructure comes from manifests, and a direct change in the console is drift that the engine removes on the next run.
The permission sets are the same on Infrastream Cloud and on Private Cloud. Only the organization that holds the role differs.
Names
| Manifest field | Common name | Custom role |
|---|---|---|
permissions.administrators | admin | managed_admin |
permissions.contributors | dev | managed_editor |
permissions.viewers | read | managed_viewer |
Name OrganizationUser and OrganizationUserGroup manifests. Never write raw email addresses.
spec:
permissions:
administrators:
groups: [platform-admins]
contributors:
groups: [payments-team]
members: [alice]
viewers:
groups: [finance-readers]
What each tier grants
| Capability | read | dev | admin |
|---|---|---|---|
| Infrastructure control-plane read | Yes | Yes | Yes |
| Spanner, Cloud SQL and BigQuery control-plane read | Yes | Yes | Yes |
Cloud Storage object read (get, list) | Yes | Yes | Yes |
Spanner data read (select, read) | Yes | Yes | Yes |
| BigQuery data read (query) | Yes | Yes | Yes |
| Identity Platform user read | Yes | Yes | Yes |
| Cloud Storage object create (add only) | No | Yes | Yes |
| Logging, mesh and edge configuration read | No | Yes | Yes |
| Cloud Storage object overwrite and delete | No | No | Yes |
| Spanner DML (insert, update, delete) | No | No | Yes |
| Cloud SQL connect and IAM login | No | No | Yes |
| BigQuery load and insert | No | No | Yes |
| Identity Platform user create, update, delete | No | No | Yes |
| Schema or DDL changes, anywhere | No | No | No |
| Secret payloads, IAM policy, impersonation | No | No | No |
- dev writes only add objects. In Cloud Storage,
storage.objects.createwithoutstorage.objects.deleteis append-only. Spanner and BigQuery have no equivalent. A database write there is full create, update and delete. So dev gets no database write. - admin controls contents, not structure. An admin can write every row and object. An admin cannot alter a table, drop a database or change a bucket policy, because manifests declare those.
- read sees application data, never credentials. A reader can read a Spanner row or a Cloud Storage object. A reader never reaches a secret payload, a Kubernetes Secret, a KMS key, an identity-provider client secret, a VM serial port or a service-account key. Bulk-extraction APIs are admin-only: Cloud SQL backup export, the BigQuery Storage Read API, and Spanner
partitionQueryandpartitionRead.
What no tier can do
No tier has these permissions. Admin does not have them either.
| Withheld | What it would expose |
|---|---|
secretmanager.versions.access | Secret payloads |
container.secrets.* | Kubernetes Secrets through GKE RBAC |
container.pods.exec | A command in a pod, which runs as the service account of the workload |
firebaseauth.configs.getHashConfig | Password-hash parameters |
firebaseauth.configs.getSecret | OAuth client secrets of upstream identity providers |
iam.serviceAccountKeys.create and .get | Long-lived service-account credentials |
cloudkms.cryptoKeyVersions.useToDecrypt | Anything wrapped with a tenant CMEK |
cloudsql.sslCerts.create | A client certificate, which is a database credential |
compute.instances.getSerialPortOutput | VM startup secrets |
Every *.setIamPolicy and every form of service-account impersonation are also withheld. Because of this, no tier can promote itself to the next tier.
Control-plane changes are also withheld, because they would be drift:
compute.*,run.*andcontainer.*create, update and delete- Spanner
updateDdlanddrop - BigQuery table and dataset changes
cloudsql.instances.updateand.deletestorage.buckets.updatelogging.sinks.*serviceusage.services.enableorgpolicy.*resourcemanager.projects.delete,.updateand.move
Two of these protect the audit trail. storage.buckets.update would let an admin turn off the object versions that make admin deletes recoverable. logging.sinks.* and logging.logs.delete would let a tier that writes data destroy the log of that write.
How grants resolve
- The highest tier wins. If a principal appears in more than one tier, it gets the role of the widest tier once.
- Grants are per project. A binding applies at the project, not on the tenant folder. The folder carries only
roles/resourcemanager.folderViewer, which lets people navigate the hierarchy. - A tier with no role grants nothing. If you set no role for a tier, the platform skips those principals. They do not fall back to a wider tier.
Outside the tiers
The tiers cover Google Cloud IAM. Two systems decide access by their own rules:
- Kubernetes RBAC. Dev and admin hold
container.clusters.connect. It opens a session with the API server of the cluster. RBAC bindings decide what that session can do, and this model does not control them. No tier holdscontainer.secrets.*orcontainer.pods.execon the Google Cloud side. - PostgreSQL roles. The Cloud SQL connect permission of admin gives a connection. Database roles and grants decide what you can do after that. See
Database.
Related
- The Security Framework
- Manage Access
Project,EnvironmentandOrganizationalUnitcarry apermissionsblock.