Skip to content

Dedicated Ceph Admin instance

Use a separate instance when Ceph RGW administration needs its own network and secret boundary. BucketReef has only three deployment profiles: full, admin and user. A dedicated instance uses full with explicit runtime switches; there is no separate security mode or implicit isolation guarantee.

The recommended isolated deployment has its own PostgreSQL database, UI/API JWT and credential key rings, RGW identity, administrator bootstrap, and restricted ingress. Sharing the database and credential ring would retain access to the same secrets and would provide only surface/network separation.

Disable Ceph Admin on the main admin instance with RUNTIME_SURFACES=["admin","storage_ops"] and do not store the dedicated Ceph Admin key there. Admin Ops and Supervision remain separate credentials.

Runtime configuration

DEPLOYMENT_PROFILE=full
RUNTIME_SURFACES=["ceph_admin"]
SCHEDULED_JOBS_ENABLED=false
WEBHOOK_WORKER_ENABLED=false
BUCKET_MIGRATION_WORKER_ENABLED=false

The local surface list exposes only Ceph Admin. Enable ceph_admin_enabled for the dedicated installation through Admin before narrowing its surfaces, or set it in the deployment manifest.

Configure endpoints with the deployment manifest. Each Ceph entry specifies ceph_admin_allowed: true and separate ceph_admin_access_key and ceph_admin_secret_key from an externally configured RGW user with admin=true and system=false. BucketReef validates that pair without creating or modifying the Ceph Admin user. For example:

[{
  "name": "Ceph Admin",
  "endpoint_url": "https://rgw.example.com",
  "provider": "ceph",
  "service_identity_mode": "managed",
  "ceph_admin_allowed": true,
  "ceph_admin_access_key": "CEPH_ADMIN_ACCESS_KEY",
  "ceph_admin_secret_key": "CEPH_ADMIN_SECRET_KEY"
}]

This example uses Ceph Admin alone: no Admin Ops, Runtime or Supervision pair is required in either managed or external service mode. Storage/Traffic metrics show an unavailable state without Supervision; usage composition uses Ceph Admin and remains accessible. Enabling Metrics/Usage in the form still requires Supervision detection.

Ceph Admin validation does not require Admin Ops credentials. If Runtime/Supervision are also needed, provide their external pairs or Admin Ops credentials with users=read,write;accounts=read for managed provisioning. Any additional stored secrets require the same network/database isolation as the Ceph Admin identity. Startup reconciles persisted endpoints even if they were not changed by ENV synchronization. Disabling the global feature blocks access and preserves the stored Ceph Admin keys and endpoint authorization; it does not revoke RGW users.

ENV endpoint metadata are synchronized into this instance's isolated database and remain environment-managed/read-only. Back up the database and its credential ring together, independently of the main deployment.

Docker Compose

Put the runtime switches above, the isolated DATABASE_URL and new key rings in an operator-owned .env.ceph-admin. Also configure its own public origin, hosts, WebAuthn origin/RP ID, published port and TLS/reverse-proxy boundary.

docker compose --project-name bucketreef-ceph-admin \
  --env-file .env.ceph-admin -f compose.yaml \
  up -d --wait backend frontend

docker compose --project-name bucketreef-ceph-admin \
  --env-file .env.ceph-admin -f compose.yaml \
  exec backend python -m app.scripts.issue_first_admin_bootstrap

Start only backend and frontend; do not start the scheduler. The main Administration project still uses compose.admin.yaml, with RUNTIME_SURFACES=["admin","storage_ops"] in its env file.

Helm

Use a separate release with an operator-owned values file containing:

deploymentProfile: full
runtimeSurfaces: [ceph_admin]
backend:
  databaseType: postgresql
  persistence:
    enabled: false
  existingSecret: bucketreef-ceph-admin-auth
  env:
    SCHEDULED_JOBS_ENABLED: "false"
    WEBHOOK_WORKER_ENABLED: "false"
    BUCKET_MIGRATION_WORKER_ENABLED: "false"
costCronJob:
  enabled: false
healthcheckCronJob:
  enabled: false
quotaMonitorCronJob:
  enabled: false
usageHistoryCronJob:
  enabled: false
notificationRetentionCronJob:
  enabled: false

Complete this file with the instance's ingress/TLS, exact trusted proxy CIDRs, origins and strict NetworkPolicy selectors/egress. The dedicated Secret contains its isolated database-url, ui-jwt-keys, api-jwt-keys and credential-keys. With scheduled jobs disabled, internal-cron-token is not consumed. Rendering fails if a built-in CronJob is enabled while scheduled-job endpoints are off.

helm upgrade --install bucketreef-ceph-admin \
  oci://ghcr.io/ksperis/charts/bucketreef --version X.Y.Z \
  --values production-ceph-admin.yaml

On the main admin release set runtimeSurfaces: [admin, storage_ops]; other profile-controlled switches remain fixed. Configure switches in backend.env, not extraEnv.

Verification

Issue the normal bootstrap token against the empty isolated database. After login, the only workspace should be Ceph Admin. Enroll a passkey from **Profile

Security**, then enable Require passkeys for administrators.

Before exposing the ingress, verify the runtime routes, run python -m app.scripts.check_production_readiness, check the restricted network path and ensure the privileged RGW identity is absent from other instances. An intentional Ceph Admin-only runtime does not warn about disabled jobs.

The removed profiles and overlays have no compatibility aliases. Before upgrade:

  • Replace admin-no-ceph-admin with admin and explicitly disable Ceph Admin.
  • Replace ceph-admin-high-security with full and the switches above; remove CEPH_ADMIN_HIGH_SECURITY_MODE and the old Compose/Helm overlay references.
  • Release tooling and external installers that validate an exact bundle file list must accept the remaining base/admin/user Compose files before distributing a new bundle.
  • Preserve the existing database and key rings during this configuration change. Moving a shared-state installation to an isolated database is a separate migration and requires its own administrator bootstrap.