@@ -429,7 +429,9 @@ def update_servers(self, server_works: List[LightningWork]):
429429 AutoScaler uses this method to increase/decrease the number of works.
430430 """
431431 old_servers = set (self ._servers )
432- server_urls : List [str ] = [server .url for server in server_works if server .url ]
432+ server_urls : List [str ] = [
433+ f"http://{ server ._internal_ip } :{ server .port } " for server in server_works if server ._internal_ip
434+ ]
433435 new_servers = set (server_urls )
434436
435437 if new_servers == old_servers :
@@ -458,12 +460,22 @@ def send_request_to_update_servers(self, servers: List[str]):
458460 detail = "Invalid authentication credentials" ,
459461 headers = {"WWW-Authenticate" : "Basic" },
460462 ) from e
463+
464+ if not self ._internal_ip :
465+ return
466+
461467 headers = {
462468 "accept" : "application/json" ,
463469 "username" : USERNAME ,
464470 "Authorization" : AUTHORIZATION_TYPE + " " + data ,
465471 }
466- response = requests .put (f"{ self .url } /system/update-servers" , json = servers , headers = headers , timeout = 10 )
472+
473+ response = requests .put (
474+ f"http://{ self ._internal_ip } :{ self .port } /system/update-servers" ,
475+ json = servers ,
476+ headers = headers ,
477+ timeout = 10 ,
478+ )
467479 response .raise_for_status ()
468480
469481 @staticmethod
@@ -735,7 +747,11 @@ def scale(self, replicas: int, metrics: dict) -> int:
735747 @property
736748 def num_pending_requests (self ) -> int :
737749 """Fetches the number of pending requests via load balancer."""
738- return int (requests .get (f"{ self .load_balancer .url } /num-requests" ).json ())
750+ if not self .load_balancer ._internal_ip :
751+ return 0
752+ return int (
753+ requests .get (f"http://{ self .load_balancer ._internal_ip } :{ self .load_balancer .port } /num-requests" ).json ()
754+ )
739755
740756 @property
741757 def num_pending_works (self ) -> int :
0 commit comments