implement aks gpu instance profile (Azure#3895)

* implement aks gpu instance profile

* exclude test from live run due to feature registration

* wrap enum opts with variable

* fix bad merge

* fix ordering

src/aks-preview/azcli_aks_live_test/configs/ext_matrix_default.json

@@ -38,7 +38,8 @@
             "test_aks_enable_monitoring_with_aad_auth_uai",
             "test_aks_create_and_update_with_managed_nat_gateway_outbound",
             "test_aks_create_with_http_proxy_config",
-            "test_aks_nodepool_add_with_workload_runtime"
+            "test_aks_nodepool_add_with_workload_runtime",
+            "test_aks_nodepool_add_with_gpu_instance_profile"
         ]
     }
 }

src/aks-preview/azext_aks_preview/_consts.py

@@ -105,3 +105,9 @@
 
 CONST_MANAGED_IDENTITY_OPERATOR_ROLE = 'Managed Identity Operator'
 CONST_MANAGED_IDENTITY_OPERATOR_ROLE_ID = 'f1a07417-d97a-45cb-824c-7a7467783830'
+
+CONST_GPU_INSTANCE_PROFILE_MIG1_G = "MIG1g"
+CONST_GPU_INSTANCE_PROFILE_MIG2_G = "MIG2g"
+CONST_GPU_INSTANCE_PROFILE_MIG3_G = "MIG3g"
+CONST_GPU_INSTANCE_PROFILE_MIG4_G = "MIG4g"
+CONST_GPU_INSTANCE_PROFILE_MIG7_G = "MIG7g"

src/aks-preview/azext_aks_preview/_help.py

@@ -355,6 +355,9 @@
         - name: --workload-runtime
           type: string
           short-summary: Determines the type of workload a node can run. Defaults to OCIContainer.
+        - name: --gpu-instance-profile
+          type: string
+          short-summary: GPU instance profile to partition multi-gpu Nvidia GPUs.
     examples:
         - name: Create a Kubernetes cluster with an existing SSH public key.
           text: az aks create -g MyResourceGroup -n MyManagedCluster --ssh-key-value /path/to/publickey
@@ -942,6 +945,9 @@
         - name: --workload-runtime
           type: string
           short-summary: Determines the type of workload a node can run. Defaults to OCIContainer.
+        - name: --gpu-instance-profile
+          type: string
+          short-summary: GPU instance profile to partition multi-gpu Nvidia GPUs.
     examples:
         - name: Create a nodepool in an existing AKS cluster with ephemeral os enabled.
           text: az aks nodepool add -g MyResourceGroup -n nodepool1 --cluster-name MyManagedCluster --node-osdisk-type Ephemeral --node-osdisk-size 48

src/aks-preview/azext_aks_preview/_params.py

@@ -32,8 +32,10 @@
     CONST_OS_DISK_TYPE_MANAGED, CONST_OS_DISK_TYPE_EPHEMERAL, \
     CONST_RAPID_UPGRADE_CHANNEL, CONST_STABLE_UPGRADE_CHANNEL, CONST_PATCH_UPGRADE_CHANNEL, CONST_NODE_IMAGE_UPGRADE_CHANNEL, CONST_NONE_UPGRADE_CHANNEL, \
     CONST_WORKLOAD_RUNTIME_OCI_CONTAINER, CONST_WORKLOAD_RUNTIME_WASM_WASI
+from ._consts import CONST_GPU_INSTANCE_PROFILE_MIG1_G, CONST_GPU_INSTANCE_PROFILE_MIG2_G, CONST_GPU_INSTANCE_PROFILE_MIG3_G, CONST_GPU_INSTANCE_PROFILE_MIG4_G, CONST_GPU_INSTANCE_PROFILE_MIG7_G
 
 workload_runtimes = [CONST_WORKLOAD_RUNTIME_OCI_CONTAINER, CONST_WORKLOAD_RUNTIME_WASM_WASI]
+gpu_instance_profiles = [CONST_GPU_INSTANCE_PROFILE_MIG1_G, CONST_GPU_INSTANCE_PROFILE_MIG2_G, CONST_GPU_INSTANCE_PROFILE_MIG3_G, CONST_GPU_INSTANCE_PROFILE_MIG4_G, CONST_GPU_INSTANCE_PROFILE_MIG7_G]
 
 
 def load_arguments(self, _):
@@ -143,6 +145,7 @@ def load_arguments(self, _):
         c.argument('enable_secret_rotation', action='store_true')
         c.argument('assign_kubelet_identity', type=str, validator=validate_assign_kubelet_identity)
         c.argument('disable_local_accounts', action='store_true')
+        c.argument('gpu_instance_profile', arg_type=get_enum_type(gpu_instance_profiles))
         c.argument('yes', options_list=['--yes', '-y'], help='Do not prompt for confirmation.', action='store_true')
         c.argument('workload_runtime', arg_type=get_enum_type(workload_runtimes), default=CONST_WORKLOAD_RUNTIME_OCI_CONTAINER)
 
@@ -238,6 +241,7 @@ def load_arguments(self, _):
             c.argument('enable_encryption_at_host', options_list=['--enable-encryption-at-host'], action='store_true')
             c.argument('enable_ultra_ssd', action='store_true')
             c.argument('workload_runtime', arg_type=get_enum_type(workload_runtimes), default=CONST_WORKLOAD_RUNTIME_OCI_CONTAINER)
+            c.argument('gpu_instance_profile', arg_type=get_enum_type(gpu_instance_profiles))
 
     for scope in ['aks nodepool show', 'aks nodepool delete', 'aks nodepool scale', 'aks nodepool upgrade', 'aks nodepool update']:
         with self.argument_context(scope) as c:

src/aks-preview/azext_aks_preview/custom.py

@@ -810,6 +810,7 @@ def aks_create(cmd,     # pylint: disable=too-many-locals,too-many-statements,to
                no_wait=False,
                assign_kubelet_identity=None,
                workload_runtime=None,
+               gpu_instance_profile=None,
                yes=False):
     if not no_ssh_key:
         try:
@@ -870,7 +871,8 @@ def aks_create(cmd,     # pylint: disable=too-many-locals,too-many-statements,to
         enable_ultra_ssd=enable_ultra_ssd,
         max_pods=int(max_pods) if max_pods else None,
         type=vm_set_type,
-        workload_runtime=workload_runtime
+        workload_runtime=workload_runtime,
+        gpu_instance_profile=gpu_instance_profile
     )
 
     if node_osdisk_size:
@@ -2378,6 +2380,7 @@ def aks_agentpool_add(cmd,      # pylint: disable=unused-argument,too-many-local
                       enable_encryption_at_host=False,
                       enable_ultra_ssd=False,
                       workload_runtime=None,
+                      gpu_instance_profile=None,
                       no_wait=False):
     instances = client.list(resource_group_name, cluster_name)
     for agentpool_profile in instances:
@@ -2432,7 +2435,8 @@ def aks_agentpool_add(cmd,      # pylint: disable=unused-argument,too-many-local
         enable_encryption_at_host=enable_encryption_at_host,
         enable_ultra_ssd=enable_ultra_ssd,
         mode=mode,
-        workload_runtime=workload_runtime
+        workload_runtime=workload_runtime,
+        gpu_instance_profile=gpu_instance_profile
     )
 
     if priority == CONST_SCALE_SET_PRIORITY_SPOT: