Skip to content

Issue #24 — prepare_db.py AKS-fanout dispatch moved to a service

Motivation

Issue #24 (split oversized files violating SRP) acceptance criterion #2 asks to move the cloud / data-plane logic out of the POST /api/storage/prepare-db route. The _try_dispatch_aks_mode helper was a ~470-line function inside the route module that mixed HTTP validation, AKS cluster-health / node-readiness / kubelet-RBAC pre-flight, NCBI key listing, metadata start-marker writes, the per-(account, db) lock, and the Celery dispatch — most of it tightly coupled to HTTPException. A previous pass (2026-06-06) deliberately deferred the full extraction to "its own focused PR with a domain-error / result-object boundary"; this change is that PR.

User-facing change

None. This is a pure structural refactor. The HTTP contract is byte-identical: same status codes, same 409 detail objects (code: aks_unavailable / kubelet_rbac_missing), same success response dict, same mode=auto server-side fall-through.

API / IaC diff summary

  • New api/services/storage/prepare_db_aks_dispatch.py:
  • try_dispatch_aks_mode(...) -> dict | None — the moved body. Returns the success response dict on dispatch, None to fall through to server-side.
  • AksDispatchError(status_code, detail) — domain error raised instead of HTTPException; the module never imports FastAPI.
  • Input-format validation that previously used common._check (which raises HTTPException) is inlined to raise AksDispatchError(400, …) with the same message format. The NCBI snapshot / key-listing / taxonomy helpers are resolved through the common module so existing test monkeypatch seams keep working.
  • api/routes/storage/prepare_db.py (1,527 → 1,066 lines):
  • Removed the _try_dispatch_aks_mode body; the route now calls try_dispatch_aks_mode(...) and translates AksDispatchError into HTTPException(exc.status_code, exc.detail).
  • Updated the context header (_try_dispatch_aks_mode dropped from the entry points; AKS dispatch boundary documented). Dropped the now-unused time import. The prepare_db, prepare_db_cancel, prepare_db_delete server-side paths and the re-exported _poll_copy_completion / _update_metadata / _is_stale_prepare_marker / _prepare_db_lock surface (used by the prepare_db_via_aks task, recover-prepare-db.py, and tests) are unchanged.
  • api/tests/test_prepare_db_aks_route.py: _baseline_patches now also patches api.routes.storage.common._resolve_latest_dir (the moved code resolves the NCBI snapshot through common); the route-module patch is kept for the server-side path.

No new dependency. No route, response, or env-var contract change.

Validation evidence

  • uv run ruff check api → All checks passed.
  • uv run pytest -q api/tests → 3669 passed, 3 skipped.
  • Focused: test_prepare_db_aks_route.py, test_prepare_db_hardening.py, test_prepare_db_routes.py, test_prepare_db_delete_route.py, test_storage_shared_taxonomy.py, test_prepare_db_aks_task.py → 56 passed.

Remaining issue #24 work (still open)

  • SettingsPanel sections still over ~600 lines: TelemetrySection (626), PublicHttpsSection (663), DiagnosticsSection (745), VnetPeeringSection (749).
  • Priority 2 frontend: ClusterBento.tsx (948) and web/src/api/blast.ts (811) not yet split.