[serve] Fix issue http proxy downscaling issues

dashboard/client/src/components/StatusChip.tsx

@@ -78,6 +78,7 @@ const colorMap = {
     [ServeSystemActorStatus.HEALTHY]: green,
     [ServeSystemActorStatus.UNHEALTHY]: red,
     [ServeSystemActorStatus.STARTING]: orange,
+    [ServeSystemActorStatus.DRAINING]: blueGrey,
   },
   serveController: {
     [ServeSystemActorStatus.HEALTHY]: green,

dashboard/client/src/pages/serve/hook/useServeApplications.ts

@@ -13,6 +13,7 @@ const SERVE_HTTP_PROXY_STATUS_SORT_ORDER: Record<
   [ServeSystemActorStatus.UNHEALTHY]: 0,
   [ServeSystemActorStatus.STARTING]: 1,
   [ServeSystemActorStatus.HEALTHY]: 2,
+  [ServeSystemActorStatus.DRAINING]: 3,
 };
 
 export const useServeApplications = () => {

dashboard/client/src/type/serve.ts

@@ -91,6 +91,7 @@ export enum ServeSystemActorStatus {
   STARTING = "STARTING",
   HEALTHY = "HEALTHY",
   UNHEALTHY = "UNHEALTHY",
+  DRAINING = "DRAINING",
 }
 
 export type ServeSystemActor = {

python/ray/serve/_private/common.py

@@ -376,6 +376,7 @@ class HTTPProxyStatus(str, Enum):
     STARTING = "STARTING"
     HEALTHY = "HEALTHY"
     UNHEALTHY = "UNHEALTHY"
+    DRAINING = "DRAINING"
 
 
 class ServeComponentType(str, Enum):

python/ray/serve/_private/deployment_state.py

@@ -10,7 +10,7 @@
 from collections import defaultdict, OrderedDict
 from copy import copy
 from enum import Enum
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+from typing import Any, Callable, Dict, List, Optional, Set, Tuple, Union
 
 import ray
 from ray import ObjectRef, cloudpickle
@@ -1227,6 +1227,20 @@ def get_running_replica_infos(self) -> List[RunningReplicaInfo]:
             for replica in self._replicas.get([ReplicaState.RUNNING])
         ]
 
+    def get_active_node_ids(self) -> Set[str]:
+        """Get the node ids of all running replicas in this deployment.
+
+        This is used to determine which node has replicas. Only nodes with replicas and
+        head node should have active proxies.
+        """
+        active_states = [
+            ReplicaState.STARTING,
+            ReplicaState.UPDATING,
+            ReplicaState.RECOVERING,
+            ReplicaState.RUNNING,
+        ]
+        return {replica.actor_node_id for replica in self._replicas.get(active_states)}
+
     def list_replica_details(self) -> List[ReplicaDetails]:
         return [replica.actor_details for replica in self._replicas.get()]
 
@@ -2496,3 +2510,14 @@ def record_multiplexed_replica_info(self, info: MultiplexedReplicaInfo):
         self._deployment_states[info.deployment_name].record_multiplexed_model_ids(
             info.replica_tag, info.model_ids
         )
+
+    def get_active_node_ids(self) -> Set[str]:
+        """Return set of node ids with running replicas of any deployment.
+
+        This is used to determine which node has replicas. Only nodes with replicas and
+        head node should have active proxies.
+        """
+        node_ids = set()
+        for deployment_state in self._deployment_states.values():
+            node_ids.update(deployment_state.get_active_node_ids())
+        return node_ids