fix dynamic config

Signed-off-by: Rui Zhang <[email protected]>

Author: zerofishnoodles

src/vllm_router/service_discovery.py

@@ -226,6 +226,7 @@ def __init__(
         self.engines_id = [str(uuid.uuid4()) for i in range(0, len(urls))]
         self.added_timestamp = int(time.time())
         self.unhealthy_endpoint_hashes = []
+        self._running = True
         if static_backend_health_checks:
             self.start_health_check_task()
         self.prefill_model_labels = prefill_model_labels
@@ -250,10 +251,10 @@ def get_unhealthy_endpoint_hashes(self) -> list[str]:
         return unhealthy_endpoints
 
     async def check_model_health(self):
-        while True:
+        while self._running:
             try:
                 self.unhealthy_endpoint_hashes = self.get_unhealthy_endpoint_hashes()
-                time.sleep(60)
+                await asyncio.sleep(60)  # Use async sleep instead of time.sleep
             except Exception as e:
                 logger.error(e)
 
@@ -340,6 +341,17 @@ async def initialize_client_sessions(self) -> None:
                         timeout=aiohttp.ClientTimeout(total=None),
                     )
 
+    def close(self):
+        """
+        Close the service discovery module and clean up health check resources.
+        """
+        self._running = False
+        if hasattr(self, "loop") and self.loop is not None:
+            self.loop.call_soon_threadsafe(self.loop.stop)
+            if hasattr(self, "thread") and self.thread is not None:
+                self.thread.join(timeout=5.0)
+            self.loop.close()
+
 
 class K8sPodIPServiceDiscovery(ServiceDiscovery):
     def __init__(