feat(RHOAIEG-57444): Add autoscaling gates for Kueue

Author: kryanbeane

src/codeflare_sdk/ray/cluster/build_ray_cluster.py

@@ -118,6 +118,17 @@ def build_ray_cluster(cluster: "codeflare_sdk.ray.cluster.Cluster"):
     # Determine autoscaling vs fixed-size worker replica settings
     autoscaling_enabled = cluster.config.enable_autoscaling
     if autoscaling_enabled:
+        from codeflare_sdk.common.kueue.kueue import get_default_kueue_name
+
+        lq_name = cluster.config.local_queue or get_default_kueue_name(
+            cluster.config.namespace
+        )
+        if lq_name is not None:
+            raise ValueError(
+                "Autoscaling is not supported when Kueue is enabled. "
+                "Please remove the autoscaler configuration from your "
+                "ClusterConfiguration."
+            )
         worker_replicas = cluster.config.min_workers
         worker_min_replicas = cluster.config.min_workers
         worker_max_replicas = cluster.config.max_workers

src/codeflare_sdk/ray/cluster/test_config.py

@@ -296,6 +296,10 @@ def test_autoscaling_disabled_ignores_workers():
 def test_autoscaling_spec_generation(mocker):
     mocker.patch("kubernetes.client.ApisApi.get_api_versions")
     mocker.patch("kubernetes.client.CustomObjectsApi.list_namespaced_custom_object")
+    mocker.patch(
+        "codeflare_sdk.common.kueue.kueue.get_default_kueue_name",
+        return_value=None,
+    )
 
     cluster = Cluster(
         ClusterConfiguration(
@@ -315,6 +319,71 @@ def test_autoscaling_spec_generation(mocker):
     assert worker_group["maxReplicas"] == 10
 
 
+def test_autoscaling_blocked_when_local_queue_set(mocker):
+    mocker.patch("kubernetes.client.ApisApi.get_api_versions")
+    mocker.patch("kubernetes.client.CustomObjectsApi.list_namespaced_custom_object")
+
+    with pytest.raises(
+        ValueError,
+        match="Autoscaling is not supported when Kueue is enabled",
+    ):
+        Cluster(
+            ClusterConfiguration(
+                name="autoscale-kueue-explicit",
+                namespace="ns",
+                enable_autoscaling=True,
+                min_workers=1,
+                max_workers=8,
+                local_queue="my-queue",
+            )
+        )
+
+
+def test_autoscaling_blocked_when_default_queue_exists(mocker):
+    mocker.patch("kubernetes.client.ApisApi.get_api_versions")
+    mocker.patch("kubernetes.client.CustomObjectsApi.list_namespaced_custom_object")
+    mocker.patch(
+        "codeflare_sdk.common.kueue.kueue.get_default_kueue_name",
+        return_value="default-queue",
+    )
+
+    with pytest.raises(
+        ValueError,
+        match="Autoscaling is not supported when Kueue is enabled",
+    ):
+        Cluster(
+            ClusterConfiguration(
+                name="autoscale-kueue-default",
+                namespace="ns",
+                enable_autoscaling=True,
+                min_workers=1,
+                max_workers=8,
+            )
+        )
+
+
+def test_autoscaling_allowed_when_no_queue(mocker):
+    mocker.patch("kubernetes.client.ApisApi.get_api_versions")
+    mocker.patch("kubernetes.client.CustomObjectsApi.list_namespaced_custom_object")
+    mocker.patch(
+        "codeflare_sdk.common.kueue.kueue.get_default_kueue_name",
+        return_value=None,
+    )
+
+    cluster = Cluster(
+        ClusterConfiguration(
+            name="autoscale-no-kueue",
+            namespace="ns",
+            enable_autoscaling=True,
+            min_workers=1,
+            max_workers=8,
+        )
+    )
+
+    spec = cluster.resource_yaml["spec"]
+    assert spec["enableInTreeAutoscaling"] is True
+
+
 def test_autoscaling_disabled_spec_unchanged(mocker):
     mocker.patch("kubernetes.client.ApisApi.get_api_versions")
     mocker.patch("kubernetes.client.CustomObjectsApi.list_namespaced_custom_object")