Enable GPU field to be set in CLI when creating new container apps

danielofficewala · danielofficewala · commit 1cdd6a326b28 · 2025-01-16T13:44:00.000-08:00
diff --git a/src/containerapp/HISTORY.rst b/src/containerapp/HISTORY.rst
@@ -12,6 +12,7 @@ upcoming
 * 'az containerapp env http-route-config': Add commands for the http-route-config feature area.
 * 'az containerapp env java-component': Support more flexible configuration updates with new parameters `--set-configurations`, `--replace-configurations`, `--remove-configurations` and `--remove-all-configurations`.
 * 'az containerapp env java-component gateway-for-spring create/update': Support `--bind` and `--unbind`
+* 'az containerapp create/update': Add an option to specify GPUs per container using the --gpu feature.
 
 1.1.0b1
 ++++++
diff --git a/src/containerapp/azext_containerapp/_compose_utils.py b/src/containerapp/azext_containerapp/_compose_utils.py
@@ -6,6 +6,8 @@
 
 from knack.log import get_logger
 
+from azure.cli.command_modules.containerapp._compose_utils import service_deploy_resources_exists
+
 logger = get_logger(__name__)
 
 
@@ -47,3 +49,14 @@ def validate_memory_and_cpu_setting(cpu, memory, managed_environment):
         logger.warning(  # pylint: disable=W1203
             f"Invalid CPU reservation request of {cpu}. The default resource values will be used.")
     return (None, None)
+
+
+def resolve_gpu_configuration_from_service(service):
+    gpu = None
+    if service_deploy_resources_exists(service):
+        resources = service.deploy.resources
+        if resources.reservations is not None and resources.reservations.gpu is not None:
+            gpu = str(resources.reservations.gpu)
+    elif service.gpu is not None:
+        gpu = str(service.gpu)
+    return gpu
diff --git a/src/containerapp/azext_containerapp/_help.py b/src/containerapp/azext_containerapp/_help.py
@@ -908,6 +908,12 @@
           az containerapp create -n my-containerapp -g MyResourceGroup \\
               --image my-app:v1.0 --environment MyContainerappEnv \\
               --enable-java-agent
+    - name: Create a container app with resource requirements and GPU defined.
+      text: |
+          az containerapp create -n my-containerapp -g MyResourceGroup \\
+              --image my-gpu-app:v1.0 --environment MyContainerappEnv \\
+              --cpu 2 --memory 4.0Gi \\
+              --gpu 1 --workload-profile-name my-gpu-wlp
 """
 
 # containerapp update for preview
diff --git a/src/containerapp/azext_containerapp/_models.py b/src/containerapp/azext_containerapp/_models.py
@@ -64,7 +64,8 @@
 
 ContainerResources = {
     "cpu": None,
-    "memory": None
+    "memory": None,
+    "gpu": None,
 }
 
 VolumeMount = {
diff --git a/src/containerapp/azext_containerapp/_sdk_models.py b/src/containerapp/azext_containerapp/_sdk_models.py
@@ -3900,6 +3900,8 @@ class ContainerResources(_serialization.Model):
     :vartype memory: str
     :ivar ephemeral_storage: Ephemeral Storage, e.g. "1Gi".
     :vartype ephemeral_storage: str
+    :ivar gpu: Required GPUs, e.g. 1.
+    :vartype gpu: float
     """
 
     _validation = {
@@ -3910,6 +3912,7 @@ class ContainerResources(_serialization.Model):
         "cpu": {"key": "cpu", "type": "float"},
         "memory": {"key": "memory", "type": "str"},
         "ephemeral_storage": {"key": "ephemeralStorage", "type": "str"},
+        "gpu": {"key": "gpu", "type": "float"},
     }
 
     def __init__(
@@ -3927,6 +3930,7 @@ def __init__(
         self.cpu = cpu
         self.memory = memory
         self.ephemeral_storage = None
+        self.gpu = None
 
 
 class CookieExpiration(_serialization.Model):
diff --git a/src/containerapp/azext_containerapp/custom.py b/src/containerapp/azext_containerapp/custom.py
@@ -455,6 +455,7 @@ def create_containerapp(cmd,
                         env_vars=None,
                         cpu=None,
                         memory=None,
+                        gpu=None,
                         registry_server=None,
                         registry_user=None,
                         registry_pass=None,
@@ -537,6 +538,7 @@ def update_containerapp_logic(cmd,
                               remove_all_env_vars=False,
                               cpu=None,
                               memory=None,
+                              gpu=None,
                               revision_suffix=None,
                               startup_command=None,
                               args=None,
@@ -603,6 +605,7 @@ def update_containerapp(cmd,
                         remove_all_env_vars=False,
                         cpu=None,
                         memory=None,
+                        gpu=None,
                         revision_suffix=None,
                         startup_command=None,
                         args=None,
@@ -643,6 +646,7 @@ def update_containerapp(cmd,
                                      remove_all_env_vars=remove_all_env_vars,
                                      cpu=cpu,
                                      memory=memory,
+                                     gpu=gpu,
                                      revision_suffix=revision_suffix,
                                      startup_command=startup_command,
                                      args=args,
@@ -902,6 +906,7 @@ def create_containerappsjob(cmd,
                             env_vars=None,
                             cpu=None,
                             memory=None,
+                            gpu=None,
                             registry_server=None,
                             registry_user=None,
                             registry_pass=None,
@@ -958,6 +963,7 @@ def update_containerappsjob(cmd,
                             remove_all_env_vars=False,
                             cpu=None,
                             memory=None,
+                            gpu=None,
                             startup_command=None,
                             args=None,
                             scale_rule_metadata=None,
@@ -1438,7 +1444,7 @@ def create_containerapps_from_compose(cmd,  # pylint: disable=R0914
                                                                        resolve_replicas_from_service,
                                                                        resolve_environment_from_service,
                                                                        resolve_secret_from_service)
-    from ._compose_utils import validate_memory_and_cpu_setting
+    from ._compose_utils import resolve_gpu_configuration_from_service, validate_memory_and_cpu_setting
 
     # Validate managed environment
     parsed_managed_env = parse_resource_id(managed_env)
@@ -1483,6 +1489,7 @@ def create_containerapps_from_compose(cmd,  # pylint: disable=R0914
             resolve_memory_configuration_from_service(service),
             managed_environment
         )
+        gpu = resolve_gpu_configuration_from_service(service)
         replicas = resolve_replicas_from_service(service)
         environment = resolve_environment_from_service(service)
         secret_vars, secret_env_ref = resolve_secret_from_service(service, parsed_compose_file.secrets)
@@ -1529,6 +1536,7 @@ def create_containerapps_from_compose(cmd,  # pylint: disable=R0914
                                 args=startup_args,
                                 cpu=cpu,
                                 memory=memory,
+                                gpu=gpu,
                                 env_vars=environment,
                                 secrets=secret_vars,
                                 min_replicas=replicas,
diff --git a/src/containerapp/azext_containerapp/tests/latest/test_containerapp_gpu.py b/src/containerapp/azext_containerapp/tests/latest/test_containerapp_gpu.py
@@ -0,0 +1,86 @@
+# --------------------------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License. See License.txt in the project root for license information.
+# --------------------------------------------------------------------------------------------
+
+import os
+import time
+
+from azure.cli.testsdk.scenario_tests import AllowLargeResponse
+from azure.cli.testsdk import (ScenarioTest, ResourceGroupPreparer, JMESPathCheck, live_only)
+
+from azext_containerapp.tests.latest.common import (write_test_file, clean_up_test_file)
+from .common import TEST_LOCATION
+
+TEST_DIR = os.path.abspath(os.path.join(os.path.abspath(__file__), '..'))
+
+class ContainerAppWorkloadProfilesGPUTest(ScenarioTest):
+    def __init__(self, *arg, **kwargs):
+        super().__init__(*arg, random_config_dir=True, **kwargs)
+
+    @AllowLargeResponse(8192)
+    @ResourceGroupPreparer(location="northeurope")
+    def test_containerapp_create_enable_dedicated_gpu(self, resource_group):
+        self.cmd('configure --defaults location={}'.format("northeurope"))
+        env = self.create_random_name(prefix='gpu-env', length=24)
+        gpu_default_name = "gpu"
+        gpu_default_type = "NC24-A100"
+        self.cmd('containerapp env create -g {} -n {} --logs-destination none --enable-dedicated-gpu'.format(
+            resource_group, env), expect_failure=False, checks=[
+            JMESPathCheck("name", env),
+            JMESPathCheck("properties.provisioningState", "Succeeded"),
+            JMESPathCheck("length(properties.workloadProfiles)", 2),
+            JMESPathCheck('properties.workloadProfiles[0].name', "Consumption", case_sensitive=False),
+            JMESPathCheck('properties.workloadProfiles[0].workloadProfileType', "Consumption", case_sensitive=False),
+            JMESPathCheck('properties.workloadProfiles[1].name', gpu_default_name, case_sensitive=False),
+            JMESPathCheck('properties.workloadProfiles[1].workloadProfileType', gpu_default_type, case_sensitive=False),
+            JMESPathCheck('properties.workloadProfiles[1].maximumCount', 1),
+            JMESPathCheck('properties.workloadProfiles[1].minimumCount', 0),
+        ])
+        containerapp_env = self.cmd('containerapp env show -g {} -n {}'.format(resource_group, env)).get_output_in_json()
+
+        while containerapp_env["properties"]["provisioningState"].lower() == "waiting":
+            time.sleep(5)
+            containerapp_env = self.cmd('containerapp env show -g {} -n {}'.format(resource_group, env)).get_output_in_json()
+        app1 = self.create_random_name(prefix='app1', length=24)
+        self.cmd(f'containerapp create -n {app1} -g {resource_group} --image mcr.microsoft.com/azuredocs/samples-tf-mnist-demo:gpu --environment {env} -w {gpu_default_name} --min-replicas 1 --cpu 0.1 --memory 0.1', checks=[
+            JMESPathCheck("properties.provisioningState", "Succeeded"),
+            JMESPathCheck("properties.workloadProfileName", gpu_default_name),
+            JMESPathCheck('properties.template.containers[0].resources.cpu', '0.1'),
+            JMESPathCheck('properties.template.containers[0].resources.memory', '0.1Gi'),
+            JMESPathCheck('properties.template.containers[0].resources.gpu', '1'),
+            JMESPathCheck('properties.template.scale.minReplicas', '1'),
+            JMESPathCheck('properties.template.scale.maxReplicas', '10')
+        ])
+
+    @AllowLargeResponse(8192)
+    @ResourceGroupPreparer(location="eastus2")
+    def test_containerapp_create_enable_consumption_gpu(self, resource_group):
+        self.cmd('configure --defaults location={}'.format("northeurope"))
+        env = self.create_random_name(prefix='consumption-gpu-env', length=24)
+        self.cmd('containerapp env create -g {} -n {} --logs-destination none --enable-workload-profiles'.format(
+            resource_group, env), expect_failure=False, checks=[
+            JMESPathCheck("name", env),
+            JMESPathCheck("properties.provisioningState", "Succeeded"),
+            JMESPathCheck("length(properties.workloadProfiles)", 1),
+            JMESPathCheck('properties.workloadProfiles[0].name', "Consumption", case_sensitive=False),
+            JMESPathCheck('properties.workloadProfiles[0].workloadProfileType', "Consumption", case_sensitive=False),
+        ])
+        consumption_gpu_wp_name = "Consumption-T4"
+
+        self.cmd("az containerapp env workload-profile set -g {} -n {} --workload-profile-name {consumption_gpu_wp_name} --workload-profile-type Consumption_GPU_NC8as_T4".format(
+            resource_group, env), expect_failure=False)
+
+        containerapp_env = self.cmd('containerapp env show -g {} -n {}'.format(resource_group, env)).get_output_in_json()
+
+        while containerapp_env["properties"]["provisioningState"].lower() == "waiting":
+            time.sleep(5)
+            containerapp_env = self.cmd('containerapp env show -g {} -n {}'.format(resource_group, env)).get_output_in_json()
+        app1 = self.create_random_name(prefix='app1', length=24)
+        self.cmd(f'containerapp create -n {app1} -g {resource_group} --image mcr.microsoft.com/azuredocs/samples-tf-mnist-demo:gpu --environment {env} -w {consumption_gpu_wp_name} --cpu 0.1 --memory 0.1 --gpu 1', checks=[
+            JMESPathCheck("properties.provisioningState", "Succeeded"),
+            JMESPathCheck("properties.workloadProfileName", consumption_gpu_wp_name),
+            JMESPathCheck('properties.template.containers[0].resources.cpu', '0.1'),
+            JMESPathCheck('properties.template.containers[0].resources.memory', '0.1Gi'),
+            JMESPathCheck('properties.template.containers[0].resources.gpu', '1'),
+        ])
diff --git a/src/containerapp/azext_containerapp/tests/latest/test_containerapp_workload_profile_commands.py b/src/containerapp/azext_containerapp/tests/latest/test_containerapp_workload_profile_commands.py
@@ -437,42 +437,3 @@ def assertContainerappProperties(self, containerapp_env, rg, app, workload_profi
             JMESPathCheck("properties.template.containers[0].resources.cpu", cpu),
             JMESPathCheck("properties.template.containers[0].resources.memory", mem)
         ])
-
-
-class ContainerAppWorkloadProfilesGPUTest(ScenarioTest):
-    def __init__(self, *arg, **kwargs):
-        super().__init__(*arg, random_config_dir=True, **kwargs)
-
-    @AllowLargeResponse(8192)
-    @ResourceGroupPreparer(location="northeurope")
-    def test_containerapp_create_enable_dedicated_gpu(self, resource_group):
-        self.cmd('configure --defaults location={}'.format("northeurope"))
-        env = self.create_random_name(prefix='gpu-env', length=24)
-        gpu_default_name = "gpu"
-        gpu_default_type = "NC24-A100"
-        self.cmd('containerapp env create -g {} -n {} --logs-destination none --enable-dedicated-gpu'.format(
-            resource_group, env), expect_failure=False, checks=[
-            JMESPathCheck("name", env),
-            JMESPathCheck("properties.provisioningState", "Succeeded"),
-            JMESPathCheck("length(properties.workloadProfiles)", 2),
-            JMESPathCheck('properties.workloadProfiles[0].name', "Consumption", case_sensitive=False),
-            JMESPathCheck('properties.workloadProfiles[0].workloadProfileType', "Consumption", case_sensitive=False),
-            JMESPathCheck('properties.workloadProfiles[1].name', gpu_default_name, case_sensitive=False),
-            JMESPathCheck('properties.workloadProfiles[1].workloadProfileType', gpu_default_type, case_sensitive=False),
-            JMESPathCheck('properties.workloadProfiles[1].maximumCount', 1),
-            JMESPathCheck('properties.workloadProfiles[1].minimumCount', 0),
-        ])
-        containerapp_env = self.cmd('containerapp env show -g {} -n {}'.format(resource_group, env)).get_output_in_json()
-
-        while containerapp_env["properties"]["provisioningState"].lower() == "waiting":
-            time.sleep(5)
-            containerapp_env = self.cmd('containerapp env show -g {} -n {}'.format(resource_group, env)).get_output_in_json()
-        app1 = self.create_random_name(prefix='app1', length=24)
-        self.cmd(f'containerapp create -n {app1} -g {resource_group} --image mcr.microsoft.com/azuredocs/samples-tf-mnist-demo:gpu --environment {env} -w {gpu_default_name} --min-replicas 1 --cpu 0.1 --memory 0.1', checks=[
-            JMESPathCheck("properties.provisioningState", "Succeeded"),
-            JMESPathCheck("properties.workloadProfileName", gpu_default_name),
-            JMESPathCheck('properties.template.containers[0].resources.cpu', '0.1'),
-            JMESPathCheck('properties.template.containers[0].resources.memory', '0.1Gi'),
-            JMESPathCheck('properties.template.scale.minReplicas', '1'),
-            JMESPathCheck('properties.template.scale.maxReplicas', '10')
-        ])

Original file line number	Diff line number	Diff line change
`@@ -64,7 +64,8 @@`
`64`	`64`
`65`	`65`	`ContainerResources = {`
`66`	`66`	`"cpu": None,`
`67`		`- "memory": None`
	`67`	`+ "memory": None,`
	`68`	`+ "gpu": None,`
`68`	`69`	`}`
`69`	`70`
`70`	`71`	`VolumeMount = {`