fix(vector): propagate cancel in cleanup task; cover 403 + sweep-failure

Address round-1 review on #913 and the SonarCloud new_reliability_rating gate:

- credential_cleanup_task no longer catches the cancellation exception
  (Sonar python:S7497). A task-group cancel must propagate for structured-
  concurrency teardown; graceful shutdown still flows through shutdown_event,
  so the sleep no longer needs a cancel/break.
- Parametrize the scanner self-heal tests over 401 AND 403 (handled
  identically at both call sites) and add a test that a failing periodic
  sweep is logged non-fatally and does not crash the task.
- Log the stored-user count before the startup sweep (operability signal),
  add a debug line when the credential row was already gone, and document
  the at-most-one extra-401 convergence in _remove_stale_credential.

Refs Deck #198.

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

nextcloud_mcp_server/app.py

@@ -2076,6 +2076,15 @@ def get_app(transport: str = "streamable-http", enabled_apps: list[str] | None =
                 # drive an endless scanner re-spawn/401 loop (Deck #198). The
                 # credential_cleanup_task started below repeats it on a cadence.
                 try:
+                    # Log the cohort first: the sweep makes one OCS validation
+                    # call per stored user before readiness, so the count is the
+                    # operability signal if startup latency ever climbs.
+                    stored = await token_storage.get_all_app_password_user_ids()
+                    if stored:
+                        logger.info(
+                            "Running startup credential sweep for %s stored user(s)",
+                            len(stored),
+                        )
                     removed = await token_storage.cleanup_invalid_app_passwords(
                         nextcloud_host=nextcloud_host_for_sync
                     )

nextcloud_mcp_server/vector/oauth_sync.py

@@ -136,6 +136,11 @@ async def _remove_stale_credential(user_id: str, status_code: int) -> None:
     #198). Deleting the row drops the user out of
     ``get_all_app_password_user_ids()`` so the scanner is not recreated.
 
+    Convergence: if ``user_manager_task`` already snapshotted the user IDs for
+    the current poll before this deletion, it re-spawns the scanner once more on
+    the next cycle, which fails auth and deletes again — at most one extra 401
+    per manager poll interval, versus the unbounded loop before this fix.
+
     Best-effort: a storage failure here is logged, not raised — the periodic
     ``credential_cleanup_task`` sweep (and the next startup sweep) are backstops.
     """
@@ -147,6 +152,14 @@ async def _remove_stale_credential(user_id: str, status_code: int) -> None:
                 user_id,
                 status_code,
             )
+        else:
+            # Row already gone — raced with the periodic sweep or another
+            # scanner exit. Harmless; logged for diagnostics.
+            logger.debug(
+                "[BasicAuth] No stale app password to remove for %s (HTTP %s)",
+                user_id,
+                status_code,
+            )
     except Exception as e:
         logger.warning(
             "[BasicAuth] Failed to remove stale app password for %s: %s", user_id, e
@@ -473,12 +486,13 @@ async def credential_cleanup_task(
     task_status.started()
 
     while not shutdown_event.is_set():
-        # Sleep first — startup already swept; wake early on shutdown.
+        # Sleep first — startup already swept; wake early on shutdown. The
-        try:
+        # graceful path goes through shutdown_event (move_on_after returns once
-            with anyio.move_on_after(CREDENTIAL_CLEANUP_INTERVAL):
+        # teardown sets it), so we deliberately do NOT catch the cancellation
-                await shutdown_event.wait()
+        # exception: a task-group cancel must propagate for structured-
-        except anyio.get_cancelled_exc_class():
+        # concurrency teardown (re-swallowing it would breach anyio's contract).
-            break
+        with anyio.move_on_after(CREDENTIAL_CLEANUP_INTERVAL):
+            await shutdown_event.wait()
         if shutdown_event.is_set():
             break
 

tests/unit/vector/test_credential_self_heal.py

@@ -1,12 +1,14 @@
 """Unit tests: self-healing removal of stale app passwords (Deck #198).
 
 When a Nextcloud user is deleted/disabled, their stored app password keeps
-returning 401. Previously the scanner just stopped, leaving the credential in
+returning 401/403. Previously the scanner just stopped, leaving the credential in
 storage so ``user_manager_task`` re-spawned the scanner every poll interval — an
 endless re-spawn/401 loop. The scanner now deletes the credential on a hard auth
 failure, and a periodic ``credential_cleanup_task`` sweeps any that slip through.
 """
 
+import logging
+
 import anyio
 import httpx
 import pytest
@@ -16,10 +18,10 @@ from nextcloud_mcp_server.vector import oauth_sync
 pytestmark = pytest.mark.unit
 
 
-def _http_401() -> httpx.HTTPStatusError:
+def _http_error(status: int) -> httpx.HTTPStatusError:
     req = httpx.Request("GET", "https://cloud.example.org/ocs")
     return httpx.HTTPStatusError(
-        "unauth", request=req, response=httpx.Response(401, request=req)
+        "auth failure", request=req, response=httpx.Response(status, request=req)
     )
 
 
@@ -52,11 +54,12 @@ async def test_remove_stale_credential_swallows_storage_error(mocker):
     await oauth_sync._remove_stale_credential("ghost-user", 401)
 
 
-async def test_scanner_removes_credential_on_prevalidation_401(mocker):
+@pytest.mark.parametrize("status", [401, 403])
-    """A 401 validating creds deletes the credential and never enters the scan
+async def test_scanner_removes_credential_on_prevalidation_auth_failure(mocker, status):
+    """A 401/403 validating creds deletes the credential and never enters the scan
     loop — so ``user_manager_task`` won't see the user as provisioned again."""
     fake_client = mocker.AsyncMock()
-    fake_client.capabilities = mocker.AsyncMock(side_effect=_http_401())
+    fake_client.capabilities = mocker.AsyncMock(side_effect=_http_error(status))
     fake_client.close = mocker.AsyncMock()
     mocker.patch.object(
         oauth_sync,
@@ -82,8 +85,9 @@ async def test_scanner_removes_credential_on_prevalidation_401(mocker):
     storage.delete_app_password.assert_awaited_once_with("ghost-user")
 
 
-async def test_scanner_removes_credential_on_scan_loop_401(mocker):
+@pytest.mark.parametrize("status", [401, 403])
-    """A 401 raised while scanning (not pre-validation) also deletes the
+async def test_scanner_removes_credential_on_scan_loop_auth_failure(mocker, status):
+    """A 401/403 raised while scanning (not pre-validation) also deletes the
     credential before the scanner stops."""
     fake_client = mocker.AsyncMock()
     fake_client.capabilities = mocker.AsyncMock(return_value={})  # pre-validation ok
@@ -96,7 +100,7 @@ async def test_scanner_removes_credential_on_scan_loop_401(mocker):
     mocker.patch.object(
         oauth_sync,
         "scan_user_documents",
-        mocker.AsyncMock(side_effect=_http_401()),
+        mocker.AsyncMock(side_effect=_http_error(status)),
     )
     storage = mocker.MagicMock()
     storage.delete_app_password = mocker.AsyncMock(return_value=True)
@@ -137,3 +141,28 @@ async def test_credential_cleanup_task_sweeps_then_stops(mocker):
     storage.cleanup_invalid_app_passwords.assert_awaited_once_with(
         "https://cloud.example.org"
     )
+
+
+async def test_credential_cleanup_task_swallows_sweep_exception(mocker, caplog):
+    """A failing sweep is logged non-fatally and does not crash the task — the
+    loop survives to retry on the next cadence."""
+    mocker.patch.object(oauth_sync, "CREDENTIAL_CLEANUP_INTERVAL", 0)
+    shutdown = anyio.Event()
+    storage = mocker.MagicMock()
+
+    async def _boom(host):
+        shutdown.set()  # exit after this (failed) iteration
+        raise RuntimeError("sweep failed")
+
+    storage.cleanup_invalid_app_passwords = mocker.AsyncMock(side_effect=_boom)
+
+    with caplog.at_level(
+        logging.WARNING, logger="nextcloud_mcp_server.vector.oauth_sync"
+    ):
+        # Must return normally — the exception is swallowed, not propagated.
+        await oauth_sync.credential_cleanup_task(
+            storage, shutdown, "https://cloud.example.org"
+        )
+
+    storage.cleanup_invalid_app_passwords.assert_awaited_once()
+    assert "non-fatal" in caplog.text