fix(vector-sync): address PR review — dict guard, symmetric backstop, metrics

- purge route: 400 (not 500) on a valid-JSON non-object body
- scanner: backstop-purge admin-disabled note/news_item/deck_card points
  (their deletion-tracking lives inside the skipped scan_* fns), mirroring the
  files path; gated on a concrete allow-set so fail-open never deletes
- processor: record_ingest_dropped("admin_disabled") so consent-skipped index
  tasks are observable/alertable
- app.py: list /api/v1/vector-sync/purge in the endpoints log line
- capabilities: drop empty-string doc types; return frozenset throughout
- purge: document the count-before-delete approximation
- tests: non-object body -> 400, ProvisioningRequiredError -> 428, cache TTL
  expiry refetch, and the scanner consent backstop

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/unit/test_capabilities.py

@@ -2,6 +2,8 @@
 
 from __future__ import annotations
 
+from unittest.mock import AsyncMock
+
 import nextcloud_mcp_server.capabilities as cap
 from nextcloud_mcp_server.capabilities import (
     _parse_enabled_doc_types,
@@ -84,22 +86,19 @@ def test_is_doc_type_allowed_empty_set_blocks_all():
 # ---------------------------------------------------------------------------
 
 
-class _FakeClient:
-    def __init__(self, payload=None, raises: Exception | None = None):
-        self._payload = payload
-        self._raises = raises
-        self.calls = 0
-
-    async def capabilities(self):
-        self.calls += 1
-        if self._raises is not None:
-            raise self._raises
-        return self._payload
+def _client(payload=None, raises: Exception | None = None) -> AsyncMock:
+    """An object with an async ``capabilities()`` method (AsyncMock-backed)."""
+    m = AsyncMock()
+    if raises is not None:
+        m.capabilities.side_effect = raises
+    else:
+        m.capabilities.return_value = payload
+    return m
 
 
 async def test_allowed_doc_types_parses_and_caches():
     clear_cache()
-    client = _FakeClient(_payload(["note", "file"]))
+    client = _client(_payload(["note", "file"]))
 
     first = await allowed_doc_types(client, "alice")
     second = await allowed_doc_types(client, "alice")
@@ -107,38 +106,53 @@ async def test_allowed_doc_types_parses_and_caches():
     assert first == frozenset({"note", "file"})
     assert second == frozenset({"note", "file"})
     # Second call served from the cache — only one OCS round-trip.
-    assert client.calls == 1
+    assert client.capabilities.await_count == 1
 
 
 async def test_allowed_doc_types_missing_block_returns_none():
     clear_cache()
-    client = _FakeClient({"ocs": {"data": {"capabilities": {}}}})
+    client = _client({"ocs": {"data": {"capabilities": {}}}})
     assert await allowed_doc_types(client, "bob") is None
 
 
 async def test_allowed_doc_types_fail_open_not_cached():
     clear_cache()
-    client = _FakeClient(raises=RuntimeError("ocs down"))
+    client = _client(raises=RuntimeError("ocs down"))
 
     assert await allowed_doc_types(client, "carol") is None
     # Failures are not cached — the next call retries the OCS lookup.
     assert await allowed_doc_types(client, "carol") is None
-    assert client.calls == 2
+    assert client.capabilities.await_count == 2
 
 
 async def test_allowed_doc_types_cache_is_per_user():
     clear_cache()
-    alice = _FakeClient(_payload(["note"]))
-    bob = _FakeClient(_payload(["file"]))
+    alice = _client(_payload(["note"]))
+    bob = _client(_payload(["file"]))
 
     assert await allowed_doc_types(alice, "alice") == frozenset({"note"})
     assert await allowed_doc_types(bob, "bob") == frozenset({"file"})
 
 
+async def test_allowed_doc_types_refetches_after_ttl(monkeypatch):
+    clear_cache()
+    client = _client(_payload(["note"]))
+
+    # Drive the module clock so the second call lands past the TTL window.
+    clock = {"now": 1000.0}
+    monkeypatch.setattr(cap.time, "monotonic", lambda: clock["now"])
+
+    await allowed_doc_types(client, "erin")
+    clock["now"] += cap._CACHE_TTL_SECONDS + 1
+    await allowed_doc_types(client, "erin")
+
+    assert client.capabilities.await_count == 2
+
+
 async def test_clear_cache_forces_refetch():
     clear_cache()
-    client = _FakeClient(_payload(["note"]))
+    client = _client(_payload(["note"]))
     await allowed_doc_types(client, "dave")
     cap.clear_cache()
     await allowed_doc_types(client, "dave")
-    assert client.calls == 2
+    assert client.capabilities.await_count == 2

tests/unit/test_vector_sync_purge_route.py

@@ -13,6 +13,7 @@ from starlette.routing import Route
 from starlette.testclient import TestClient
 
 from nextcloud_mcp_server.api.vector_sync import purge_doc_types_route
+from nextcloud_mcp_server.auth.scope_authorization import ProvisioningRequiredError
 
 pytestmark = pytest.mark.unit
 
@@ -72,7 +73,7 @@ def _patch_purge(mocker, result=None):
     )
 
 
-async def test_unauthorized_when_token_invalid(mocker):
+def test_unauthorized_when_token_invalid(mocker):
     mocker.patch(
         "nextcloud_mcp_server.api.vector_sync.validate_token_and_get_user",
         new=AsyncMock(side_effect=ValueError("bad token")),
@@ -86,7 +87,7 @@ async def test_unauthorized_when_token_invalid(mocker):
     purge.assert_not_called()
 
 
-async def test_bad_request_when_doc_types_not_list(mocker):
+def test_bad_request_when_doc_types_not_list(mocker):
     _patch_token(mocker)
     purge = _patch_purge(mocker)
 
@@ -97,7 +98,7 @@ async def test_bad_request_when_doc_types_not_list(mocker):
     purge.assert_not_called()
 
 
-async def test_forbidden_when_not_admin(mocker):
+def test_forbidden_when_not_admin(mocker):
     _patch_token(mocker, "bob")
     _patch_basic_auth(mocker, "bob")
     _patch_outbound_client(mocker)
@@ -111,7 +112,7 @@ async def test_forbidden_when_not_admin(mocker):
     purge.assert_not_called()
 
 
-async def test_empty_doc_types_is_noop(mocker):
+def test_empty_doc_types_is_noop(mocker):
     _patch_token(mocker)
     purge = _patch_purge(mocker)
 
@@ -123,7 +124,7 @@ async def test_empty_doc_types_is_noop(mocker):
     purge.assert_not_called()
 
 
-async def test_admin_purge_happy_path(mocker):
+def test_admin_purge_happy_path(mocker):
     _patch_token(mocker, "admin")
     _patch_basic_auth(mocker, "admin")
     _patch_outbound_client(mocker)
@@ -136,3 +137,30 @@ async def test_admin_purge_happy_path(mocker):
     assert resp.status_code == 200
     assert resp.json() == {"purged": {"file": 12}}
     purge.assert_awaited_once_with(["file"])
+
+
+def test_bad_request_when_body_not_object(mocker):
+    # A valid JSON non-object (e.g. a list) must 400, not 500.
+    _patch_token(mocker)
+    purge = _patch_purge(mocker)
+
+    client = TestClient(_build_app())
+    resp = client.post("/api/v1/vector-sync/purge", json=[1, 2, 3])
+
+    assert resp.status_code == 400
+    purge.assert_not_called()
+
+
+def test_provisioning_required_returns_428(mocker):
+    _patch_token(mocker, "admin")
+    mocker.patch(
+        "nextcloud_mcp_server.api.vector_sync.get_basic_auth_for_user",
+        new=AsyncMock(side_effect=ProvisioningRequiredError("not provisioned")),
+    )
+    purge = _patch_purge(mocker)
+
+    client = TestClient(_build_app())
+    resp = client.post("/api/v1/vector-sync/purge", json={"doc_types": ["file"]})
+
+    assert resp.status_code == 428
+    purge.assert_not_called()

tests/unit/vector/test_purge.py

@@ -20,10 +20,12 @@ def _patch_qdrant(monkeypatch, *, counts: dict[str, int], delete_raises=None):
     def _doc_type_of(flt):
         return flt.must[0].match.value
 
-    async def fake_count(*, collection_name, count_filter, exact):
+    # Sync side_effects: AsyncMock awaits the call and returns the value, so the
+    # helpers don't need to be coroutines themselves.
+    def fake_count(*, collection_name, count_filter, exact):
         return SimpleNamespace(count=counts.get(_doc_type_of(count_filter), 0))
 
-    async def fake_delete(*, collection_name, points_selector):
+    def fake_delete(*, collection_name, points_selector):
         dt = _doc_type_of(points_selector)
         if delete_raises and dt in delete_raises:
             raise RuntimeError(f"delete failed for {dt}")
@@ -31,10 +33,9 @@ def _patch_qdrant(monkeypatch, *, counts: dict[str, int], delete_raises=None):
     client.count.side_effect = fake_count
     client.delete.side_effect = fake_delete
 
-    async def fake_get_qdrant_client():
-        return client
-
-    monkeypatch.setattr(purge_module, "get_qdrant_client", fake_get_qdrant_client)
+    monkeypatch.setattr(
+        purge_module, "get_qdrant_client", AsyncMock(return_value=client)
+    )
     monkeypatch.setattr(
         purge_module,
         "get_settings",

tests/unit/vector/test_scanner_consent_backstop.py

@@ -0,0 +1,81 @@
+"""Unit tests for the scanner's admin-consent backstop deletion.
+
+When an admin disables a text source (note/news_item/deck_card), the scanner
+skips its scan_* function, so the in-function deletion-tracking never runs. The
+backstop enqueues deletes for any indexed points of the disabled type, mirroring
+the files path — but only on a concrete allow-set (never on fail-open None).
+"""
+
+from __future__ import annotations
+
+from types import SimpleNamespace
+from typing import cast
+from unittest.mock import AsyncMock
+
+from nextcloud_mcp_server.vector import scanner as scanner_module
+from nextcloud_mcp_server.vector.queue.ports import TaskProducer
+from nextcloud_mcp_server.vector.scanner import _enqueue_deletes_for_disabled_types
+
+
+def _producer(send: AsyncMock) -> TaskProducer:
+    """A minimal stand-in for the TaskProducer protocol (only ``send`` is used)."""
+    return cast(TaskProducer, SimpleNamespace(send=send))
+
+
+def _patch_qdrant(monkeypatch, points_by_type: dict[str, list[str]]):
+    client = AsyncMock()
+
+    def fake_scroll(
+        *, collection_name, scroll_filter, with_payload, with_vectors, limit, offset
+    ):
+        # must=[user_id, doc_type] — doc_type is the second condition.
+        doc_type = scroll_filter.must[1].match.value
+        points = [
+            SimpleNamespace(payload={"doc_id": doc_id})
+            for doc_id in points_by_type.get(doc_type, [])
+        ]
+        return (points, None)
+
+    client.scroll.side_effect = fake_scroll
+    monkeypatch.setattr(
+        scanner_module, "get_qdrant_client", AsyncMock(return_value=client)
+    )
+    monkeypatch.setattr(
+        scanner_module,
+        "get_settings",
+        lambda: SimpleNamespace(get_collection_name=lambda: "c"),
+    )
+
+
+async def test_enqueues_deletes_for_disabled_text_type(monkeypatch):
+    _patch_qdrant(monkeypatch, {"note": ["n1", "n2"], "deck_card": ["d1"]})
+    sent: list = []
+    stream = _producer(AsyncMock(side_effect=lambda t: sent.append(t)))
+
+    # note disabled; news_item + deck_card still allowed.
+    allowed = frozenset({"file", "news_item", "deck_card"})
+    queued = await _enqueue_deletes_for_disabled_types("alice", stream, allowed, 1)
+
+    assert queued == 2
+    assert {t.doc_id for t in sent} == {"n1", "n2"}
+    assert all(t.operation == "delete" and t.doc_type == "note" for t in sent)
+
+
+async def test_noop_when_allowed_is_none(monkeypatch):
+    # Fail-open: a transient capability read must never trigger deletion.
+    send = AsyncMock()
+    queued = await _enqueue_deletes_for_disabled_types(
+        "alice", _producer(send), None, 1
+    )
+    assert queued == 0
+    send.assert_not_called()
+
+
+async def test_noop_when_all_text_types_allowed(monkeypatch):
+    send = AsyncMock()
+    allowed = frozenset({"note", "news_item", "deck_card", "file"})
+    queued = await _enqueue_deletes_for_disabled_types(
+        "alice", _producer(send), allowed, 1
+    )
+    assert queued == 0
+    send.assert_not_called()