Check spec version then fall back to env var

Co-authored-by: Jun-Hao Wan <ken89@kimo.com>
Signed-off-by: Alima Azamat <92766804+alimaazamat@users.noreply.github.com>
Signed-off-by: alimaazamat <alima.azamat2003@gmail.com>

Author: alimaazamat

ray-operator/controllers/ray/utils/validation.go

@@ -609,10 +609,14 @@ func validateWorkerGroupIdleTimeout(workerGroup rayv1.WorkerGroupSpec, spec *ray
 		}
 
 		// idleTimeoutSeconds only allowed on autoscaler v2
+		if IsAutoscalingV2Enabled(spec) {
+			return nil
+		}
 		envVar, exists := EnvVarByName(RAY_ENABLE_AUTOSCALER_V2, spec.HeadGroupSpec.Template.Spec.Containers[RayContainerIndex].Env)
-		if !exists || (envVar.Value != "1" && envVar.Value != "true") {
-			return fmt.Errorf("worker group %s has idleTimeoutSeconds set, but %s environment variable is not set to 'true' in the head pod", workerGroup.GroupName, RAY_ENABLE_AUTOSCALER_V2)
+		if exists && (envVar.Value == "1" || envVar.Value == "true") {
+			return nil
 		}
+		return fmt.Errorf("worker group %s has idleTimeoutSeconds set, but autoscaler v2 is not enabled. Please set .spec.autoscalerOptions.version to 'v2' (or set %s environment variable to 'true' in the head pod if using KubeRay < 1.4.0)", workerGroup.GroupName, RAY_ENABLE_AUTOSCALER_V2)
 	}
 
 	return nil

ray-operator/controllers/ray/utils/validation_test.go

@@ -1964,7 +1964,7 @@ func TestValidateWorkerGroupIdleTimeout(t *testing.T) {
 					},
 				},
 			},
-			expectedErr: "worker group worker-group-1 has idleTimeoutSeconds set, but RAY_enable_autoscaler_v2 environment variable is not set to 'true' in the head pod",
+			expectedErr: fmt.Sprintf("worker group worker-group-1 has idleTimeoutSeconds set, but autoscaler v2 is not enabled. Please set .spec.autoscalerOptions.version to 'v2' (or set %s environment variable to 'true' in the head pod if using KubeRay < 1.4.0)", RAY_ENABLE_AUTOSCALER_V2),
 		},
 		"should reject idleTimeoutSeconds when autoscaler version is not set": {
 			spec: rayv1.RayClusterSpec{
@@ -1982,7 +1982,7 @@ func TestValidateWorkerGroupIdleTimeout(t *testing.T) {
 					},
 				},
 			},
-			expectedErr: "worker group worker-group-1 has idleTimeoutSeconds set, but RAY_enable_autoscaler_v2 environment variable is not set to 'true' in the head pod",
+			expectedErr: fmt.Sprintf("worker group worker-group-1 has idleTimeoutSeconds set, but autoscaler v2 is not enabled. Please set .spec.autoscalerOptions.version to 'v2' (or set %s environment variable to 'true' in the head pod if using KubeRay < 1.4.0)", RAY_ENABLE_AUTOSCALER_V2),
 		},
 		"should reject idleTimeoutSeconds when AutoscalerOptions is nil": {
 			spec: rayv1.RayClusterSpec{
@@ -2000,7 +2000,7 @@ func TestValidateWorkerGroupIdleTimeout(t *testing.T) {
 					},
 				},
 			},
-			expectedErr: "worker group worker-group-1 has idleTimeoutSeconds set, but RAY_enable_autoscaler_v2 environment variable is not set to 'true' in the head pod",
+			expectedErr: fmt.Sprintf("worker group worker-group-1 has idleTimeoutSeconds set, but autoscaler v2 is not enabled. Please set .spec.autoscalerOptions.version to 'v2' (or set %s environment variable to 'true' in the head pod if using KubeRay < 1.4.0)", RAY_ENABLE_AUTOSCALER_V2),
 		},
 		"should reject idleTimeoutSeconds when env var is set to invalid value": {
 			spec: rayv1.RayClusterSpec{
@@ -2020,7 +2020,7 @@ func TestValidateWorkerGroupIdleTimeout(t *testing.T) {
 					},
 				},
 			},
-			expectedErr: "worker group worker-group-1 has idleTimeoutSeconds set, but RAY_enable_autoscaler_v2 environment variable is not set to 'true' in the head pod",
+			expectedErr: fmt.Sprintf("worker group worker-group-1 has idleTimeoutSeconds set, but autoscaler v2 is not enabled. Please set .spec.autoscalerOptions.version to 'v2' (or set %s environment variable to 'true' in the head pod if using KubeRay < 1.4.0)", RAY_ENABLE_AUTOSCALER_V2),
 		},
 		"should accept worker group with idleTimeoutSeconds when env var is set to true": {
 			spec: rayv1.RayClusterSpec{