관리

사용자·그룹·서비스 토큰·OIDC 클라이언트를 관리하는 API입니다.

📄️ Reset Resource Metadata

Clear resource synchronization state so the next synchronization pass reprocesses resource metadata.

Name reservations that outlived their owner and cannot self-heal. Null only until the first sweep runs (or while detection is disabled) — a clean sweep still writes the marker, as ``status: ok`` with ``count: 0``. Act on ``reachable_count``; the rest are scoped to collections that no longer exist and cannot collide.

📄️ Release Name Fences

Release fenced name reservations. Admin only; explicit list, no wildcard. Every precondition is re-checked here rather than trusted from the marker — the marker is up to ``NAME_FENCE_DETECT_INTERVAL_SECONDS`` stale, and the same TOCTOU reasoning applies as to the D1b drop guards. A reservation is released only when it still exists, its owner is still unresolvable, and a non-``completed`` checkpoint is still holding it.

📄️ Get Deletion Guards

Delete guards whose holder died and which nothing will ever clear. Null only until the first sweep runs (or while detection is disabled) — a clean sweep still writes the marker, as ``status: ok`` with ``count: 0``. Act on ``blocking_count``: a ``collection`` guard refuses the DELETE of every child in that collection, and any other kind refuses its own while the resource still exists.

📄️ Release Deletion Guards

Clear stuck delete guards. Admin only; explicit list, no wildcard. Preconditions are re-checked here rather than trusted from the marker, the same TOCTOU reasoning as the D1b drop guards. The one fact taken from the marker is ``first_seen``, for guards too old to carry a ``start_time`` — it is the only record of their age, and it can only under-state it, so it never lets a young guard through. This frees the 409 only. The name reservation is deliberately left held, so no same-name resource can be created and then destroyed by a late cleanup — release those separately via ``/admin/name-fences/release`` once the physical objects are confirmed gone.

📄️ Retry Cleanups

The lossless counterpart of ``/name-fences/release``: instead of freeing the name and abandoning the physical residue, this re-runs the cleanup so the reservation is released on the normal all-steps-succeeded path. Targets any non-``completed`` checkpoint — including ``exhausted`` (re-arms the reclaim series at attempt 1) and the pre-reclaim backlog the scheduled pass skips as too old. A ``running`` checkpoint younger than an hour is refused (a live run may hold it); an older one is a dead mid-run worker, and re-dispatching it is safe because every step is an idempotent DROP.

📄️ Recover Ontology Bulk Delete

Re-dispatch a stored bulk-delete run that stopped in a partial or failed state, using the checkpoint saved for that run. Administrator authentication is required. A run that is still pending or running is refused, and repeating the call on an already re-dispatched run makes no further change.

📄️ List Users

List users after administrator authentication. Use `category` to limit results to a category such as `service_account`.

📄️ Create User

Create a new user. Requires admin authentication.

📄️ Get User Versions

Get user versions. Requires admin authentication.

📄️ Get User

Get a specific user. Requires admin authentication.

📄️ Update User

Update a user. Requires admin authentication.

📄️ Delete User

Permanently remove a user account. Administrator authentication is required, and the deleted account can no longer authenticate.

📄️ List User Permissions

List the permission grants a user directly holds. Requires admin. The inverse of ``GET /permissions`` (subjects for a resource): this returns the resources ``user_id`` was granted a role on — used to review a service account's access. Only directly-written grants are returned (model-implied and collection-inherited roles are excluded), so each is revocable.