Add options to install FSx Lustre and Nvidia software

Signed-off-by: Hanwen <[email protected]>

CHANGELOG.md

@@ -3,6 +3,9 @@ CHANGELOG
 3.12.0
 ------
 
+**ENHANCEMENTS**
+- Add new build image configuration section `Build/Installation` to turn on/off Nvidia software and Lustre client installations. By default, Nvidia software, although included in official ParallelCluster AMIs, is not installed by `build-image`. By default, Lustre client is installed.
+
 **CHANGES**
 - The CLI commands `export-cluster-logs` and `export-image-logs` can now by default export the logs to the default ParallelCluster bucket or to the CustomS3Bucket if specified in the config.
 

cli/src/pcluster/config/imagebuilder_config.py

@@ -33,6 +33,7 @@
 from pcluster.validators.imagebuilder_validators import (
     AMIVolumeSizeValidator,
     ComponentsValidator,
+    InstanceTypeSoftwareValidator,
     SecurityGroupsAndSubnetValidator,
 )
 from pcluster.validators.kms_validators import KmsKeyIdEncryptedValidator, KmsKeyValidator
@@ -143,6 +144,41 @@ def __init__(
         self.enabled = enabled
 
 
+class LustreClient(Resource):
+    """Represent the LustreClient configuration for the ImageBuilder."""
+
+    def __init__(
+        self,
+        enabled: bool = None,
+    ):
+        super().__init__()
+        self.enabled = Resource.init_param(enabled, default=True)
+
+
+class NvidiaSoftware(Resource):
+    """Represent the NvidiaSoftware configuration for the ImageBuilder."""
+
+    def __init__(
+        self,
+        enabled: bool = None,
+    ):
+        super().__init__()
+        self.enabled = Resource.init_param(enabled, default=False)
+
+
+class Installation(Resource):
+    """Represent the installation configuration for the ImageBuilder."""
+
+    def __init__(
+        self,
+        lustre_client: LustreClient = None,
+        nvidia_software: NvidiaSoftware = None,
+    ):
+        super().__init__()
+        self.lustre_client = lustre_client or LustreClient()
+        self.nvidia_software = nvidia_software or NvidiaSoftware()
+
+
 class Build(Resource):
     """Represent the build configuration for the ImageBuilder."""
 
@@ -157,6 +193,7 @@ def __init__(
         components: List[Component] = None,
         update_os_packages: UpdateOsPackages = None,
         imds: Imds = None,
+        installation: Installation = None,
     ):
         super().__init__()
         self.instance_type = Resource.init_param(instance_type)
@@ -168,13 +205,19 @@ def __init__(
         self.components = components
         self.update_os_packages = update_os_packages
         self.imds = imds or Imds(implied="v2.0")
+        self.installation = installation or Installation()
 
     def _register_validators(self, context: ValidatorContext = None):  # noqa: D102 #pylint: disable=unused-argument
         self._register_validator(
             InstanceTypeBaseAMICompatibleValidator,
             instance_type=self.instance_type,
             image=self.parent_image,
         )
+        self._register_validator(
+            InstanceTypeSoftwareValidator,
+            instance_type=self.instance_type,
+            nvidia=self.installation.nvidia_software.enabled,
+        )
         self._register_validator(
             ComponentsValidator,
             components=self.components,
@@ -282,21 +325,24 @@ def lambda_functions_vpc_config(self):
 class ImageBuilderExtraChefAttributes(ExtraChefAttributes):
     """Extra Attributes for ImageBuilder Chef Client."""
 
-    def __init__(self, dev_settings: ImagebuilderDevSettings):
-        super().__init__(dev_settings)
+    def __init__(self, config: ImageBuilderConfig):
+        super().__init__(config.dev_settings)
         self.region = None
         self.nvidia = None
+        self.lustre = None
         self.is_official_ami_build = None
         self.custom_node_package = None
         self.custom_awsbatchcli_package = None
         self.base_os = None
         self.disable_kernel_update = None
         self.slurm_patches_s3_archive = None
-        self._set_default(dev_settings)
+        self._set_default(config)
 
-    def _set_default(self, dev_settings: ImagebuilderDevSettings):
+    def _set_default(self, config: ImageBuilderConfig):
+        dev_settings = config.dev_settings
         self.region = "{{ build.AWSRegion.outputs.stdout }}"
-        self.nvidia = {"enabled": "no"}
+        self.nvidia = {"enabled": "yes"} if config.build.installation.nvidia_software.enabled else {"enabled": "no"}
+        self.lustre = {"enabled": "yes"} if config.build.installation.lustre_client.enabled else {"enabled": "no"}
         self.is_official_ami_build = "false"
         self.custom_node_package = dev_settings.node_package if dev_settings and dev_settings.node_package else ""
         self.custom_awsbatchcli_package = (

cli/src/pcluster/schemas/imagebuilder_schema.py

@@ -26,6 +26,9 @@
     ImageBuilderConfig,
     ImagebuilderDeploymentSettings,
     ImagebuilderDevSettings,
+    Installation,
+    LustreClient,
+    NvidiaSoftware,
     UpdateOsPackages,
     Volume,
 )
@@ -167,6 +170,40 @@ def make_resource(self, data, **kwargs):
         return UpdateOsPackages(**data)
 
 
+class LustreClientSchema(BaseSchema):
+    """Represent the schema of the ImageBuilder NvidiaSoftware."""
+
+    enabled = fields.Bool()
+
+    @post_load
+    def make_resource(self, data, **kwargs):
+        """Generate resource."""
+        return LustreClient(**data)
+
+
+class NvidiaSoftwareSchema(BaseSchema):
+    """Represent the schema of the ImageBuilder NvidiaSoftware."""
+
+    enabled = fields.Bool()
+
+    @post_load
+    def make_resource(self, data, **kwargs):
+        """Generate resource."""
+        return NvidiaSoftware(**data)
+
+
+class InstallationSchema(BaseSchema):
+    """Represent the schema of the ImageBuilder Installation."""
+
+    lustre_client = fields.Nested(LustreClientSchema)
+    nvidia_software = fields.Nested(NvidiaSoftwareSchema)
+
+    @post_load
+    def make_resource(self, data, **kwargs):
+        """Generate resource."""
+        return Installation(**data)
+
+
 class BuildSchema(BaseSchema):
     """Represent the schema of the ImageBuilder Build."""
 
@@ -179,6 +216,7 @@ class BuildSchema(BaseSchema):
     subnet_id = fields.Str(validate=get_field_validator("subnet_id"))
     update_os_packages = fields.Nested(UpdateOsPackagesSchema)
     imds = fields.Nested(ImdsSchema)
+    installation = fields.Nested(InstallationSchema)
 
     @post_load
     def make_resource(self, data, **kwargs):

cli/src/pcluster/templates/imagebuilder_stack.py

@@ -197,7 +197,7 @@ def _add_cfn_parameters(self):
             self,
             "CfnParamChefDnaJson",
             type="String",
-            default=ImageBuilderExtraChefAttributes(self.config.dev_settings).dump_json(),
+            default=ImageBuilderExtraChefAttributes(self.config).dump_json(),
             description="ChefAttributes",
         )
         CfnParameter(

cli/src/pcluster/validators/imagebuilder_validators.py

@@ -54,3 +54,17 @@ def _validate(self, security_group_ids: list, subnet_id: str):
                     "Subnet id {0} is specified, security groups is required.".format(subnet_id),
                     FailureLevel.ERROR,
                 )
+
+
+class InstanceTypeSoftwareValidator(Validator):
+    """Validate software compatibility with instance type."""
+
+    def _validate(self, instance_type: str, nvidia: bool):
+        if nvidia:
+            instance_type_info = AWSApi.instance().ec2.get_instance_type_info(instance_type)
+            if instance_type_info.gpu_count() == 0:
+                self._add_failure(
+                    f"Instance type {instance_type} does not have GPU. "
+                    "NVIDIA software can only be installed on GPU instances.",
+                    FailureLevel.ERROR,
+                )

cli/tests/pcluster/config/dummy_imagebuilder_config.py

@@ -19,6 +19,9 @@
     ImageBuilderConfig,
     ImagebuilderDeploymentSettings,
     ImagebuilderDevSettings,
+    Installation,
+    LustreClient,
+    NvidiaSoftware,
     UpdateOsPackages,
     Volume,
 )
@@ -39,6 +42,9 @@
     "additional_iam_policies": AdditionalIamPolicy,
     "update_os_packages": UpdateOsPackages,
     "imds": Imds,
+    "installation": Installation,
+    "lustre_client": LustreClient,
+    "nvidia_software": NvidiaSoftware,
 }