add cli parameter

Author: pintaoz

doc/cli/cluster_management/cli_cluster_management.md

@@ -358,7 +358,7 @@ The `config.yaml` file supports the following parameters:
 | `create_s3_endpoint_stack` | BOOLEAN | Create S3 Endpoint stack | true |
 | `enable_hp_inference_feature` | BOOLEAN | Enable inference operator | false |
 | `stage` | TEXT | Deployment stage ("gamma" or "prod") | "prod" |
-| `custom_bucket_name` | TEXT | S3 bucket name for templates | "aws-sagemaker-hyperpod-cluster" |
+| `custom_bucket_name` | TEXT | S3 bucket name for templates | "aws-sagemaker-hyperpod-cluster-setup" |
 | `create_life_cycle_script_stack` | BOOLEAN | Create Life Cycle Script Stack | true |
 | `create_s3_bucket_stack` | BOOLEAN | Create S3 Bucket Stack | true |
 | `s3_bucket_name` | TEXT | S3 bucket for cluster lifecycle scripts | "s3-bucket" |

hyperpod-cluster-stack-template/hyperpod_cluster_stack_template/v1_0/model.py

@@ -35,7 +35,7 @@ class ClusterStackBase(BaseModel):
     create_s3_endpoint_stack: Optional[bool] = Field(True, description="Boolean to Create S3 Endpoint stack")
     enable_hp_inference_feature: Optional[bool] = Field(False, description="Boolean to enable inference operator in Hyperpod cluster")
     stage: Optional[str] = Field("prod", description="Deployment stage used in S3 bucket naming for inference operator. Valid values: \"gamma\", \"prod\"")
-    custom_bucket_name: str = Field("aws-sagemaker-hyperpod-cluster", description="S3 bucket name for templates")
+    custom_bucket_name: str = Field("aws-sagemaker-hyperpod-cluster-setup", description="S3 bucket name for templates")
     create_life_cycle_script_stack: Optional[bool] = Field(True, description="Boolean to Create Life Cycle Script Stack")
     create_s3_bucket_stack: Optional[bool] = Field(True, description="Boolean to Create S3 Bucket Stack")
     s3_bucket_name: Optional[str] = Field("s3-bucket", description="The name of the S3 bucket used to store the cluster lifecycle scripts")
@@ -51,7 +51,6 @@ class ClusterStackBase(BaseModel):
     file_system_type_version: Optional[float] = Field(2.15, description="File system type version for the FSx file system")
     storage_capacity: Optional[int] = Field(1200, description="Storage capacity for the FSx file system in GiB")
     fsx_file_system_id: Optional[str] = Field("", description="Existing FSx file system ID")
-    template_version: Optional[str] = Field("1", description="Version number of cluster creation template")
 
     @field_validator('kubernetes_version', mode='before')
     @classmethod
@@ -121,7 +120,7 @@ def to_config(self, region: str = None):
 
         # Set fixed defaults
         defaults = {
-            'custom_bucket_name': 'aws-sagemaker-hyperpod-cluster',
+            'custom_bucket_name': 'aws-sagemaker-hyperpod-cluster-setup',
             'github_raw_url': 'https://raw.githubusercontent.com/aws-samples/awsome-distributed-training/refs/heads/main/1.architectures/7.sagemaker-hyperpod-eks/LifecycleScripts/base-config/on_create.sh',
             'helm_repo_url': 'https://github.com/aws/sagemaker-hyperpod-cli.git',
             'helm_repo_path': 'helm_chart/HyperPodHelmChart'

src/sagemaker/hyperpod/cli/commands/cluster_stack.py

@@ -53,8 +53,9 @@ def parse_status_list(ctx, param, value):
 @click.argument("config-file", required=True)
 @click.argument("stack-name", required=True)
 @click.option("--region", help="AWS region")
+@click.option("--template-version", help="Version number of cluster creation template")
 @click.option("--debug", is_flag=True, help="Enable debug logging")
-def create_cluster_stack(config_file, region, debug):
+def create_cluster_stack(config_file, region, template_version, debug):
     """Create a new HyperPod cluster stack using the provided configuration.
 
     Creates a CloudFormation stack for a HyperPod cluster using settings from a YAML configuration file.
@@ -66,7 +67,7 @@ def create_cluster_stack(config_file, region, debug):
        .. code-block:: bash
 
           # Create cluster stack with config file
-          hyp create hyp-cluster cluster-config.yaml my-stack-name --region us-west-2
+          hyp create hyp-cluster cluster-config.yaml my-stack-name --region us-west-2 --template-version 1
 
           # Create with debug logging
           hyp create hyp-cluster cluster-config.yaml my-stack-name --debug
@@ -95,7 +96,7 @@ def create_cluster_stack(config_file, region, debug):
             config = model_instance.to_config(region=region)
 
             # Create the cluster stack
-            stack_id = HpClusterStack(**config).create(region)
+            stack_id = HpClusterStack(**config).create(region, template_version)
 
             logger.info(f"Stack creation initiated successfully with ID: {stack_id}")
             logger.info("You can monitor the stack creation in the AWS CloudFormation console.")

src/sagemaker/hyperpod/cluster_management/hp_cluster_stack.py

@@ -66,7 +66,8 @@ def get_template() -> str:
 
     @_hyperpod_telemetry_emitter(Feature.HYPERPOD, "create_cluster_stack")
     def create(self,
-               region: Optional[str] = None) -> str:
+               region: Optional[str] = None,
+               template_version: Optional[int] = 1) -> str:
         """Creates a new HyperPod cluster CloudFormation stack.
 
         **Parameters:**
@@ -112,7 +113,7 @@ def create(self,
         stack_name = f"HyperpodClusterStack-{str(uuid.uuid4())[:5]}"
         # Use the fixed bucket name from the model
         bucket_name = self.custom_bucket_name
-        template_key = f"{self.template_version}/main-stack-eks-based-template.yaml"
+        template_key = f"{template_version}/templates/main-stack-eks-based-template.yaml"
 
         try:
             # Use TemplateURL for large templates (>51KB)