refactor: govern background job lifecycle

backend/application.py

@@ -128,14 +128,19 @@ class DashboardService(
             profile_supplier=self._resolved_llm_profile,
         )
         self.screener.ensure_builtin_strategies()
-        self._background_stop = threading.Event()
+
-        self._background_thread = self.jobs.start_scheduler(
+    def start_background_jobs(self) -> threading.Thread:
+        return self.jobs.start_scheduler(
             self._background_refresh_tick,
-            self._background_stop,
             interval_seconds=5,
             initial_delay_seconds=3,
         )
 
+    def stop_background_jobs(self, timeout_seconds: float = 5) -> bool:
+        scheduler_stopped = self.jobs.stop_scheduler(timeout_seconds)
+        workers_stopped = self.jobs.wait_for_idle(timeout_seconds)
+        return scheduler_stopped and workers_stopped
+
 
     def _load_system_credentials(self, environment: dict[str, str]) -> dict[str, Any]:
         encrypted = self.database.get_system_setting("credentials")

backend/bootstrap/runtime.py

@@ -16,12 +16,13 @@ def main(handler_class: type[Any] | None = None, service: Any | None = None) ->
     parser.add_argument("--port", type=int, default=8765)
     args = parser.parse_args()
     server = ThreadingHTTPServer((args.host, args.port), handler_class)
-    print(f"Xiaobai Review Web is running at http://{args.host}:{args.port}")
-    print("Press Ctrl+C to stop.")
     try:
+        service.start_background_jobs()
+        print(f"Xiaobai Review Web is running at http://{args.host}:{args.port}")
+        print("Press Ctrl+C to stop.")
         server.serve_forever()
     except KeyboardInterrupt:
         pass
     finally:
-        service._background_stop.set()
+        service.stop_background_jobs()
         server.server_close()

backend/jobs/runner.py

@@ -18,6 +18,9 @@ class InProcessJobRunner:
         self.repository = repository
         self._locks: dict[str, threading.Lock] = {}
         self._locks_guard = threading.Lock()
+        self._scheduler_guard = threading.Lock()
+        self._scheduler_stop = threading.Event()
+        self._scheduler_thread: threading.Thread | None = None
 
     def submit(
         self, job_id: str, idempotency_key: str, action: JobAction,
@@ -52,27 +55,42 @@ class InProcessJobRunner:
         return True
 
     def start_scheduler(
-        self, callback: Callable[[], None], stop_event: threading.Event,
+        self, callback: Callable[[], None], interval_seconds: float,
-        interval_seconds: float, initial_delay_seconds: float = 0,
+        initial_delay_seconds: float = 0,
     ) -> threading.Thread:
-        def schedule_loop() -> None:
+        with self._scheduler_guard:
-            if stop_event.wait(initial_delay_seconds):
+            current = self._scheduler_thread
-                return
+            if current is not None and current.is_alive():
-            while not stop_event.is_set():
+                return current
-                try:
+            self._scheduler_stop.clear()
-                    callback()
-                except Exception:
-                    # Submitted jobs persist their own failures; the scheduler must stay alive.
-                    pass
-                stop_event.wait(interval_seconds)
 
-        thread = threading.Thread(
+            def schedule_loop() -> None:
-            target=schedule_loop,
+                if self._scheduler_stop.wait(initial_delay_seconds):
-            name="background-job-scheduler",
+                    return
-            daemon=True,
+                while not self._scheduler_stop.is_set():
-        )
+                    try:
-        thread.start()
+                        callback()
-        return thread
+                    except Exception:
+                        # Submitted jobs persist failures; the scheduler must stay alive.
+                        pass
+                    self._scheduler_stop.wait(interval_seconds)
+
+            thread = threading.Thread(
+                target=schedule_loop,
+                name="background-job-scheduler",
+                daemon=True,
+            )
+            self._scheduler_thread = thread
+            thread.start()
+            return thread
+
+    def stop_scheduler(self, timeout_seconds: float = 5) -> bool:
+        with self._scheduler_guard:
+            thread = self._scheduler_thread
+            self._scheduler_stop.set()
+            if thread is not None and thread is not threading.current_thread():
+                thread.join(max(0, timeout_seconds))
+            return thread is None or not thread.is_alive()
 
     def wait_for_idle(self, timeout_seconds: float = 5) -> bool:
         deadline = time.monotonic() + max(0, timeout_seconds)

config/architecture-inventory.json

@@ -374,8 +374,8 @@
     },
     {
       "path": "backend/application.py",
-      "bytes": 47552,
+      "bytes": 47769,
-      "lines": 1087
+      "lines": 1092
     },
     {
       "path": "frontend/styles/theme.css",

tests/test_background_lifecycle.py

@@ -0,0 +1,65 @@
+from __future__ import annotations
+
+import ast
+import sys
+import unittest
+from pathlib import Path
+from unittest.mock import patch
+
+from backend.bootstrap import runtime
+
+
+class BackgroundLifecycleTests(unittest.TestCase):
+    def test_service_construction_does_not_start_the_scheduler(self) -> None:
+        source = (Path(__file__).parents[1] / "backend" / "application.py").read_text(
+            encoding="utf-8"
+        )
+        tree = ast.parse(source)
+        service_class = next(
+            node for node in tree.body
+            if isinstance(node, ast.ClassDef) and node.name == "DashboardService"
+        )
+        constructor = next(
+            node for node in service_class.body
+            if isinstance(node, ast.FunctionDef) and node.name == "__init__"
+        )
+        called_methods = {
+            node.func.attr for node in ast.walk(constructor)
+            if isinstance(node, ast.Call) and isinstance(node.func, ast.Attribute)
+        }
+        self.assertNotIn("start_scheduler", called_methods)
+
+    def test_runtime_starts_jobs_after_bind_and_stops_before_close(self) -> None:
+        events: list[str] = []
+
+        class Service:
+            def start_background_jobs(self) -> None:
+                events.append("start-jobs")
+
+            def stop_background_jobs(self) -> None:
+                events.append("stop-jobs")
+
+        class Server:
+            def __init__(self, address: tuple[str, int], handler: object) -> None:
+                events.append("bind")
+
+            def serve_forever(self) -> None:
+                events.append("serve")
+                raise KeyboardInterrupt
+
+            def server_close(self) -> None:
+                events.append("close")
+
+        with (
+            patch.object(runtime, "ThreadingHTTPServer", Server),
+            patch.object(sys, "argv", ["server.py", "--port", "8797"]),
+        ):
+            runtime.main(object, Service())
+
+        self.assertEqual(
+            events, ["bind", "start-jobs", "serve", "stop-jobs", "close"]
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_job_runner.py

@@ -68,6 +68,29 @@ class JobRunnerTests(unittest.TestCase):
         )
         self.assertEqual(self.repository.recent(1)[0]["status"], "failed")
 
+    def test_scheduler_start_is_idempotent_and_stop_waits_for_exit(self) -> None:
+        first = self.runner.start_scheduler(
+            lambda: None, interval_seconds=60, initial_delay_seconds=60
+        )
+        second = self.runner.start_scheduler(
+            lambda: None, interval_seconds=60, initial_delay_seconds=60
+        )
+        self.assertIs(first, second)
+        self.assertTrue(first.is_alive())
+        self.assertTrue(self.runner.stop_scheduler())
+        self.assertFalse(first.is_alive())
+
+    def test_scheduler_can_restart_after_an_orderly_stop(self) -> None:
+        first = self.runner.start_scheduler(
+            lambda: None, interval_seconds=60, initial_delay_seconds=60
+        )
+        self.assertTrue(self.runner.stop_scheduler())
+        second = self.runner.start_scheduler(
+            lambda: None, interval_seconds=60, initial_delay_seconds=60
+        )
+        self.assertIsNot(first, second)
+        self.assertTrue(self.runner.stop_scheduler())
+
 
 if __name__ == "__main__":
     unittest.main()

tools/run_preservation_runtime.py

@@ -32,17 +32,22 @@ def main() -> None:
     from server import RequestHandler, SERVICE
 
     server = ThreadingHTTPServer(("127.0.0.1", args.port), RequestHandler)
-    print(
-        f"Preservation runtime is running at http://127.0.0.1:{args.port} "
-        f"with database {SERVICE.database.path}",
-        flush=True,
-    )
     try:
+        if hasattr(SERVICE, "start_background_jobs"):
+            SERVICE.start_background_jobs()
+        print(
+            f"Preservation runtime is running at http://127.0.0.1:{args.port} "
+            f"with database {SERVICE.database.path}",
+            flush=True,
+        )
         server.serve_forever()
     except KeyboardInterrupt:
         pass
     finally:
-        SERVICE._background_stop.set()
+        if hasattr(SERVICE, "stop_background_jobs"):
+            SERVICE.stop_background_jobs()
+        else:
+            SERVICE._background_stop.set()
         server.server_close()