breaking(BA-5653): drop sessions/kernels access_key columns by jopemachine · Pull Request #11040 · lablup/backend.ai

jopemachine · 2026-04-14T06:59:43Z

📚 Stack

main
- refactor(BA-5709): add main_access_key resolver helpers #11041 — refactor(BA-5709): add main_access_key resolver helpers
  - refactor(BA-5710): thread main_access_key through UserPermission #11043 — refactor(BA-5710): thread main_access_key through UserPermission
    - refactor(BA-5711): rename SessionData user_uuid to owner_id #11045 — refactor(BA-5711): rename SessionData user_uuid to owner_id
      - refactor(BA-5650): add owner_id helpers and rename session data types [1/4] #11046 — refactor(BA-5712): switch session repository to owner_id
        
        refactor(BA-5713): collapse scheduler signatures to owner_id #11047 — refactor(BA-5713): collapse scheduler signatures to owner_id
        
        refactor(BA-5650): propagate owner_id through scheduler and sokovan [2/4] #11048 — refactor(BA-5714): propagate owner_id rename into sokovan
        
        refactor(BA-5715): resolve owner_id via current_user() in services #11049 — refactor(BA-5715): resolve owner_id via current_user() in services
        
        refactor(BA-5650): unify session ownership to owner_id and main_access_key [1/2] #11050 — breaking(BA-5716): drop owner_access_key from REST v1 session API
        
        refactor(BA-5650): update tests and remaining ORM references [2/2] #11051 — refactor(BA-5717): test and remaining ORM updates (final tip)
        👉 breaking(BA-5653): drop sessions/kernels access_key columns #11040 ← you are here — breaking(BA-5653): drop sessions/kernels access_key columns

Merge from top to bottom. Intermediate slices may not build standalone; the final refactor tip is #11051, and #11040 is the schema-drop follow-up on top of it.

Summary

Breaking schema change. Alembic migration 8c1d2e3f4a5b_drop_session_kernel_access_key drops sessions.access_key and kernels.access_key columns. Downstream callers resolve the owner's main_access_key from the users table. user_uuid column stays. ORM drop of the access_key mapped column and call-site migration follow in subsequent commits on this branch.

Resolves BA-5653. Part of epic BA-5650.

📚 Documentation preview 📚: https://sorna--11040.org.readthedocs.build/en/11040/

📚 Documentation preview 📚: https://sorna-ko--11040.org.readthedocs.build/ko/11040/

BA-5650 Series: Split Rationale

Overall goal: migrate the session owner identifier from access_key (keypair) to owner_id (user UUID), and drop the sessions.access_key / kernels.access_key columns.

Split criteria: layer + dependency order. Bottom-up (DB helpers → service → API) so the destructive column drop can land safely at the end.

Order	Issue	Layer	Role
1	BA-5709	Foundation	Add `get_main_access_key_by_id` and related resolver helpers (everything else depends on this)
2	BA-5710	RBAC / Permission	Rename `UserPermission.user_uuid → owner_id`; add `main_access_key` field
3	BA-5711	Data	Rename `SessionData.user_uuid → owner_id`; Row adapters; GQL node
4	BA-5712	Repository	Collapse `SessionRepository`, `SessionDBSource`, creators signatures
5	BA-5713	Scheduler	predicates / drf / options; scheduler repo; stream/events shims
6	BA-5714	Sokovan	allocation / lifecycle / workload, launcher, controller, executor
7	BA-5715	Service	Drop `owner_id` from 21 read/control Actions; resolve via `current_user()`
8	BA-5716	API (v1)	Remove `owner_access_key` from REST v1 DTOs (breaking)
9	BA-5717	Tests / Cleanup	Test suite updates; remaining ORM / gql_legacy touch-ups
10	BA-5653	Schema	Alembic migration — rename/drop columns (destructive, must land last)

Why this split

Dependencies: the BA-5709 helpers are required before 3–7 can recover access_key from owner_id. BA-5653 (destructive) only runs once every reader has migrated off the dropped column.
Review size: a single-shot change would touch hundreds of files. Slicing by layer keeps each PR focused on one concern.
Rollback safety: steps 1–7 are no-op on external behavior, step 8 is the API breaking change, step 10 is the schema drop. Each step is independently revertible.

Copilot

Pull request overview

This PR introduces a breaking schema change to remove redundant access_key snapshot columns from the sessions and kernels tables, aligning persistence with the newer approach of resolving the owner’s main_access_key via the users table.

Changes:

Add an Alembic migration to drop sessions.access_key and kernels.access_key (with best-effort index cleanup).
Regenerate/update GraphQL schema reference docs (enum ordering change only).
Add a breaking-change note documenting the dropped columns.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
`src/ai/backend/manager/models/alembic/versions/8c1d2e3f4a5b_drop_session_kernel_access_key.py`	Migration to drop `access_key` columns from `sessions`/`kernels`.
`docs/manager/graphql-reference/v2-schema.graphql`	Documentation-only schema output change (enum value ordering).
`docs/manager/graphql-reference/supergraph.graphql`	Documentation-only supergraph output change (enum value ordering).
`changes/11040.breaking.md`	Release-note entry for the breaking schema change.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Clean up the ``changes/BA-5650-{D,E,F}.misc.md`` files that were superseded when each slice's news fragment was renamed to the assigned PR number (e.g. ``changes/11046.enhance.md``). These stragglers made it onto downstream branches during the cascade rebases.

- Tests for SessionEnqueueData / KernelEnqueueData / TerminatingSessionData now use ``main_access_key`` and ``owner_id`` (renamed in slice F/G). - ``SessionAdapter._session_data_to_node`` reads ``data.owner_id`` and drops the obsolete ``data.access_key``; the SessionMetadata GQL DTO's ``access_key`` field is now an empty string until the read-time resolver lands. - ``cache_invalidation`` uses ``info.access_key`` (the field name on SessionTransitionInfo). - ``ShutdownServiceAction``/``GetContainerLogsAction``/``RenameSessionAction`` no longer accept ``owner_access_key``; drop the kwarg in the GQL adapter. - ``ModelServingRepository.get_session_by_id`` drops the now-removed positional ``owner_access_key`` arg from ``SessionRow.get_session``. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

REST v1 session endpoints no longer accept owner_access_key. The delegation field is replaced with owner_id (user UUID) and is honored only on the three session-creation endpoints (create_from_template / create_from_params / create_cluster). Read and control endpoints always act as the authenticated caller. - common/dto/manager/session/request.py: drop the owner_access_key field from Create/Destroy/Restart/GetContainerLogs/GetStatusHistory request DTOs; add owner_id to the three creation DTOs. - api/rest/session/handler.py: remove all 26 resolve_access_key_scope calls, drop the AuthProcessors dependency, and renumber log format placeholders that dropped the owner argument. - api/rest/v2/session/handler.py: drop the redundant user_ctx / access_key arguments from shutdown_service / get_logs / update. - api/adapters/session.py: update adapter call sites accordingly. - api/rest/tree.py: drop the auth= argument from SessionHandler(). Test updates for the corresponding DTO assertions and component fixtures are included.

Final slice of the BA-5650 stack split. Contains: - Remaining ORM touch-ups: models/endpoint/row.py, models/keypair/row.py, repositories/scheduler/{repository,db_source}.py, api/adapters/vfolder.py, api/gql_legacy/endpoint.py, api/gql_legacy/routing.py. - Test updates that depend on the action/service/DTO renames already landed in earlier slices: adapter/session, scheduler repositories, session lifecycle/service, sokovan scheduler suite, compute_sessions handler, dependency injection tests. - Autouse ``_user_context`` fixture under tests/unit/manager/services/session/conftest.py so service tests work without the auth middleware.

Co-authored-by: octodog <mu001@lablup.com>

This slice was carrying unrelated work that crept in during the original PR split (prometheus client, valkey route health, deployment auto_activate, web ssl_enabled, auth client_type_id, gql_legacy routing health_status). All such files are reverted to the slice-H base. Also fixes: - Use session.user.main_access_key (loaded via selectinload) instead of session.main_access_key in scheduler db_source — SessionRow has no main_access_key attribute. - UserService delegate path now resolves target_main_access_key via UserRepository.get_main_access_key_by_id and forwards it to delegate_endpoint_ownership. - Drop stale changes/BA-5650-H.breaking.md (slice H ships 11050.breaking.md). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Resolve conflict markers in scheduler db_source by keeping the user-table-joined ``main_access_key`` lookup (single source of truth) rather than the obsolete ``session.access_key`` snapshot. - Drop duplicate ``resolve_main_access_keys`` definition introduced by the rebase. - Use ``spec.access_key`` (SessionCreationSpec) where applicable — the spec dataclass keeps ``access_key`` for now. - Test fixture ``_make_session_data`` now passes ``owner_id`` instead of relying on attribute assignment after construction. - Use ``main_access_key`` on PendingSessionData in provisioner test. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

``SessionRepository`` and the underlying ``SessionDBSource`` now take ``owner_id: UUID`` on every method that previously accepted ``owner_access_key: AccessKey``. Affects: - ``get_session_validated`` - ``match_sessions`` - ``update_session_name`` - ``find_dependency_sessions`` / ``_find_dependent_sessions`` - ``get_target_session_ids`` - ``get_session_with_group`` The matching ``dependency_graph`` helpers and ``creators`` are updated in lockstep. Service-layer callers still pass ``owner_access_key`` temporarily; they will be migrated in the next slice.

Rename access_key -> main_access_key on sokovan data types (SessionAllocation, PreparedSessionData, SessionDataForPull, SessionDataForStart, SessionWorkload) and update every sokovan caller accordingly. Affects: - sokovan/data/{allocation,lifecycle,workload}.py - sokovan/scheduler/handlers/lifecycle/* - sokovan/scheduler/handlers/maintenance/sweep_sessions.py - sokovan/scheduler/provisioner/{provisioner,sequencers,validators}/* - sokovan/scheduler/launcher/launcher.py - sokovan/scheduler/post_processors/cache_invalidation.py - sokovan/scheduler/fair_share/aggregator.py - sokovan/scheduling_controller/{preparers,scheduling_controller}.py - sokovan/deployment/{executor,route}.py No external behavior change.