Add validation for CLI input

Author: benieric

src/hyperpod_cli/commands/job.py

@@ -662,7 +662,7 @@ def start_job(
                 config["cluster"]["cluster_config"]["volumes"] = volume_mount
 
             if label_selector is not None:
-                config["cluster"]["cluster_config"]["label_selector"] = label_selector
+                config["cluster"]["cluster_config"]["label_selector"] = json.loads(label_selector)
             elif deep_health_check_passed_nodes_only:
                 config["cluster"]["cluster_config"]["label_selector"] = (
                     DEEP_HEALTH_CHECK_PASSED_ONLY_NODE_AFFINITY_DICT
@@ -672,15 +672,14 @@ def start_job(
                     NODE_AFFINITY_DICT
                 )
 
-            label_selector = config["cluster"].setdefault("label_selector",{})
+            label_selector = config["cluster"]["cluster_config"].setdefault("label_selector",{})
             required_labels = label_selector.setdefault("required", {})
 
             if not required_labels.get(KUBERNETES_INSTANCE_TYPE_LABEL_KEY):
                 required_labels[KUBERNETES_INSTANCE_TYPE_LABEL_KEY] = (
                     [str(instance_type)]
                 )
 
-
             if auto_resume:
                 # Set max_retry default to 1
                 if max_retry is None:

src/hyperpod_cli/validators/job_validator.py

@@ -192,11 +192,6 @@ def validate_yaml_content(data):
         required_labels[KUBERNETES_INSTANCE_TYPE_LABEL_KEY] = (
             [str(instance_type)]
         )
-    if instance_type not in required_labels[KUBERNETES_INSTANCE_TYPE_LABEL_KEY]:
-        logger.error(
-            f"Please ensure 'instance-type' in 'cluster' matches with 'instance-type' in 'label_selector.required.beta.kubernetes.io/instance-type' in config file"
-        )
-        return False
 
     auto_resume = False
     max_retry = None

test/unit_tests/test_job.py

@@ -576,7 +576,6 @@ def capture_yaml_dump(config, *args, **kwargs):
         if result.exception:
             print(f"Exception: {result.exception}")
 
-
     @mock.patch('subprocess.run')
     @mock.patch("yaml.dump")
     @mock.patch("os.path.exists", return_value=True)

test/unit_tests/validators/test_job_validator.py

@@ -1087,8 +1087,7 @@ def test_validate_yaml_content_valid(self):
         result = validate_yaml_content(mock_data)
         self.assertTrue(result)
 
-    def test_validate_yaml_content_valid_required_instance_type_label(self):
-        # Test auto-populate label selector - "beta.kubernetes.io/instance_type"
+    def test_validate_yaml_content_required_instance_type_label(self):
         expected_label_selector = {
             "required": {
                 "beta.kubernetes.io/instance-type": [
@@ -1097,6 +1096,7 @@ def test_validate_yaml_content_valid_required_instance_type_label(self):
             }
         }
 
+        # User does not provide label_selector
         mock_data = {
             "cluster": {
                 "cluster_type": "k8s",
@@ -1124,6 +1124,7 @@ def test_validate_yaml_content_valid_required_instance_type_label(self):
             }
         }
 
+        # User provides label_selector without instance_type
         mock_data = {
             "cluster": {
                 "cluster_type": "k8s",
@@ -1146,59 +1147,7 @@ def test_validate_yaml_content_valid_required_instance_type_label(self):
         self.assertEqual(
             mock_data["cluster"]["cluster_config"]["label_selector"], expected_label_selector
         )
-        
-        expected_label_selector = {
-            "required": {
-                "beta.kubernetes.io/instance-type": [
-                    "ml.g5.xlarge"
-                ]
-            }
-        }
-
-        mock_data = {
-            "cluster": {
-                "cluster_type": "k8s",
-                "instance_type": "ml.g5.xlarge",
-                "cluster_config": {
-                    "scheduler": "SageMaker",
-                    "label_selector": {
-                        "required": {
-                            "beta.kubernetes.io/instance-type": [
-                                "ml.g5.xlarge"
-                            ]
-                        }
-                    }
-                }
-            }
-        }
 
-        result = validate_yaml_content(mock_data)
-        self.assertTrue(result)
-        self.assertEqual(
-            mock_data["cluster"]["cluster_config"]["label_selector"], expected_label_selector
-        )
-        
-    def test_validate_yaml_content_invalid_required_instance_type_label(self):
-        # Test resepect user selection label selector
-        mock_data = {
-            "cluster": {
-                "cluster_type": "k8s",
-                "instance_type": "ml.g5.xlarge",
-                "cluster_config": {
-                    "scheduler": "SageMaker",
-                    "label_selector": {
-                        "required": {
-                            "beta.kubernetes.io/instance-type": [
-                                "ml.g5.2xlarge"
-                            ]
-                        }
-                    }
-                }
-            }
-        }
-
-        result = validate_yaml_content(mock_data)
-        self.assertFalse(result)
 
     def test_validate_yaml_content_error_no_cluster(
         self,