fix: await background sync task cancellation in lifespan shutdown (#456)

Signed-off-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: Paul Hernandez <phernandez@users.noreply.github.com>

Author: phernandez

src/basic_memory/api/app.py

@@ -59,6 +59,7 @@ async def lifespan(app: FastAPI):  # pragma: no cover
         app.state.sync_task = asyncio.create_task(initialize_file_sync(app_config))
     else:
         logger.info("Sync changes disabled. Skipping file sync service.")
+        app.state.sync_task = None
 
     # proceed with startup
     yield
@@ -67,6 +68,10 @@ async def lifespan(app: FastAPI):  # pragma: no cover
     if app.state.sync_task:
         logger.info("Stopping sync...")
         app.state.sync_task.cancel()  # pyright: ignore
+        try:
+            await app.state.sync_task
+        except asyncio.CancelledError:
+            logger.info("Sync task cancelled successfully")
 
     await db.shutdown_db()
 

test-int/mcp/test_lifespan_shutdown_sync_task_cancellation_integration.py

@@ -0,0 +1,70 @@
+"""
+Integration test for FastAPI lifespan shutdown behavior.
+
+This test verifies the asyncio cancellation pattern used by the API lifespan:
+when the background sync task is cancelled during shutdown, it must be *awaited*
+before database shutdown begins. This prevents "hang on exit" scenarios in
+`asyncio.run(...)` callers (e.g. CLI/MCP clients using httpx ASGITransport).
+"""
+
+import asyncio
+
+from httpx import ASGITransport, AsyncClient
+
+
+def test_lifespan_shutdown_awaits_sync_task_cancellation(app, monkeypatch):
+    """
+    Ensure lifespan shutdown awaits the cancelled background sync task.
+
+    Why this is deterministic:
+    - Cancelling a task does not make it "done" immediately; it becomes done only
+      once the event loop schedules it and it processes the CancelledError.
+    - In the buggy version, shutdown proceeded directly to db.shutdown_db()
+      immediately after calling cancel(), so at *entry* to shutdown_db the task
+      is still not done.
+    - In the fixed version, lifespan does `await sync_task` before shutdown_db,
+      so by the time shutdown_db is called, the task is done (cancelled).
+    """
+
+    # Import the *module* (not the package-level FastAPI `basic_memory.api.app` export)
+    # so monkeypatching affects the exact symbols referenced inside lifespan().
+    #
+    # Note: `basic_memory/api/__init__.py` re-exports `app`, so `import basic_memory.api.app`
+    # can resolve to the FastAPI instance rather than the `basic_memory.api.app` module.
+    import importlib
+
+    api_app_module = importlib.import_module("basic_memory.api.app")
+
+    # Keep startup cheap: we don't need real DB init for this ordering test.
+    async def _noop_initialize_app(_app_config):
+        return None
+
+    async def _fake_get_or_create_db(*_args, **_kwargs):
+        return object(), object()
+
+    monkeypatch.setattr(api_app_module, "initialize_app", _noop_initialize_app)
+    monkeypatch.setattr(api_app_module.db, "get_or_create_db", _fake_get_or_create_db)
+
+    # Make the sync task long-lived so it must be cancelled on shutdown.
+    async def _fake_initialize_file_sync(_app_config):
+        await asyncio.Event().wait()
+
+    monkeypatch.setattr(api_app_module, "initialize_file_sync", _fake_initialize_file_sync)
+
+    # Assert ordering: shutdown_db must be called only after the sync_task is done.
+    async def _assert_sync_task_done_before_db_shutdown():
+        assert api_app_module.app.state.sync_task is not None
+        assert api_app_module.app.state.sync_task.done()
+
+    monkeypatch.setattr(api_app_module.db, "shutdown_db", _assert_sync_task_done_before_db_shutdown)
+
+    async def _run_client_once():
+        async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as client:
+            # Any request is sufficient to trigger lifespan startup/shutdown.
+            await client.get("/__nonexistent__")
+
+    # Use asyncio.run to match the CLI/MCP execution model where loop teardown
+    # would hang if a background task is left running.
+    asyncio.run(_run_client_once())
+
+