[serve] Fix get endpoint when autoscaling config is set (ray-project#…

…34377) If autoscaling config is set for a deployment, we can't set the num replicas when returning the deployment details of that deployment. Otherwise, it breaks the entirety of the get metadata endpoint.
vitsai · Apr 17, 2023 · b8c847e · b8c847e
1 parent 825d9e3
commit b8c847e
Show file tree

Hide file tree

Showing 2 changed files with 29 additions and 11 deletions.
diff --git a/dashboard/modules/serve/tests/test_serve_agent.py b/dashboard/modules/serve/tests/test_serve_agent.py
@@ -460,7 +460,21 @@ def test_get_status(ray_start_stop):
 
 
 @pytest.mark.skipif(sys.platform == "darwin", reason="Flaky on OSX.")
-def test_get_serve_instance_details(ray_start_stop):
+@pytest.mark.parametrize(
+    "f_deployment_options",
+    [
+        {"name": "f", "ray_actor_options": {"num_cpus": 0.2}},
+        {
+            "name": "f",
+            "autoscaling_config": {
+                "min_replicas": 1,
+                "initial_replicas": 3,
+                "max_replicas": 10,
+            },
+        },
+    ],
+)
+def test_get_serve_instance_details(ray_start_stop, f_deployment_options):
     world_import_path = "ray.serve.tests.test_config_files.world.DagNode"
     fastapi_import_path = "ray.serve.tests.test_config_files.fastapi_deployment.node"
     config1 = {
@@ -474,12 +488,7 @@ def test_get_serve_instance_details(ray_start_stop):
                 "name": "app1",
                 "route_prefix": "/app1",
                 "import_path": world_import_path,
-                "deployments": [
-                    {
-                        "name": "f",
-                        "ray_actor_options": {"num_cpus": 0.2},
-                    },
-                ],
+                "deployments": [f_deployment_options],
             },
             {
                 "name": "app2",
@@ -553,7 +562,11 @@ def applications_running():
             assert "route_prefix" not in deployment.deployment_config.dict(
                 exclude_unset=True
             )
-            assert len(deployment.replicas) == deployment.deployment_config.num_replicas
+            if isinstance(deployment.deployment_config.num_replicas, int):
+                assert (
+                    len(deployment.replicas)
+                    == deployment.deployment_config.num_replicas
+                )
 
             for replica in deployment.replicas:
                 assert replica.state == ReplicaState.RUNNING

diff --git a/python/ray/serve/schema.py b/python/ray/serve/schema.py
@@ -271,12 +271,10 @@ def _deployment_info_to_schema(name: str, info: DeploymentInfo) -> DeploymentSch
     codepath)
     """
 
-    return DeploymentSchema(
+    schema = DeploymentSchema(
         name=name,
-        num_replicas=info.deployment_config.num_replicas,
         max_concurrent_queries=info.deployment_config.max_concurrent_queries,
         user_config=info.deployment_config.user_config,
-        autoscaling_config=info.deployment_config.autoscaling_config,
         graceful_shutdown_wait_loop_s=(
             info.deployment_config.graceful_shutdown_wait_loop_s
         ),
@@ -287,6 +285,13 @@ def _deployment_info_to_schema(name: str, info: DeploymentInfo) -> DeploymentSch
         is_driver_deployment=info.is_driver_deployment,
     )
 
+    if info.deployment_config.autoscaling_config is not None:
+        schema.autoscaling_config = info.deployment_config.autoscaling_config
+    else:
+        schema.num_replicas = info.deployment_config.num_replicas
+
+    return schema
+
 
 @PublicAPI(stability="beta")
 class ServeApplicationSchema(BaseModel, extra=Extra.forbid):