xpk 0.15.0__py3-none-any.whl → 0.16.1__py3-none-any.whl

xpk/core/system_characteristics.py

@@ -15,11 +15,29 @@ limitations under the License.
 """
 
 from dataclasses import dataclass
+import dataclasses
+from typing import Callable, Literal, Optional
+
+from ..core.workload_decorators import rdma_decorator, tcpxo_decorator, tcpx_decorator
 from ..utils.topology import get_topology_product
 from enum import Enum
 
 SUB_SLICING_TOPOLOGIES = ['2x4', '4x4', '4x8', '8x8', '8x16', '16x16']
 
+INSTALLER_NCCL_TCPX = 'https://raw.githubusercontent.com/GoogleCloudPlatform/container-engine-accelerators/master/gpudirect-tcpx/nccl-tcpx-installer.yaml'
+INSTALLER_NCCL_TCPXO = 'https://raw.githubusercontent.com/GoogleCloudPlatform/container-engine-accelerators/master/gpudirect-tcpxo/nccl-tcpxo-installer.yaml'
+INSTALLER_NCCL_RDMA = 'https://raw.githubusercontent.com/GoogleCloudPlatform/container-engine-accelerators/master/gpudirect-rdma/nccl-rdma-installer.yaml'
+INSTALLER_NCCL_RDMA_A4X = 'https://raw.githubusercontent.com/GoogleCloudPlatform/container-engine-accelerators/master/gpudirect-rdma/nccl-rdma-installer-a4x.yaml'
+
+
+class DockerPlatform(str, Enum):
+  AMD = 'linux/amd64'
+  ARM = 'linux/arm64'
+
+
+AMD_PLATFORM = DockerPlatform.AMD
+ARM_PLATFORM = DockerPlatform.ARM
+
 
 class AcceleratorType(Enum):
   TPU = 1
@@ -56,6 +74,45 @@ AcceleratorTypeToAcceleratorCharacteristics = {
 }
 
 
+@dataclass
+class GpuConfig:
+  """Contains GPU-specific configuration and requirements."""
+
+  requires_topology: bool
+  gpu_direct_name: Literal['fastrak', 'rdma', 'tcpx', 'tcpxo'] = 'fastrak'
+  kjob_decorator_fn: Optional[Callable[[dict], dict]] = None
+  """A function to decorate the kjob template for GPU-specific configurations.
+
+  Args:
+    job_manifest (dict): The kjob manifest as a dictionary.
+
+  Returns:
+    dict: The modified kjob manifest as a dictionary.
+  """
+  nccl_installer: Optional[str] = None
+  jobset_decorator_fn: Optional[Callable[[str, list[str]], str]] = None
+  """A function to decorate the jobset for GPU-specific configurations.
+
+  Args:
+    jobset_manifest_str (str): The JobSet manifest as a YAML string.
+    sub_networks (list[str], optional): A list of sub-network names, used by some decorators.
+
+  Returns:
+    str: The modified JobSet manifest as a YAML string.
+  """
+
+  def __repr__(self) -> str:
+    """Returns a string representation of the GpuConfig, omitting memory addresses for functions."""
+    parts = []
+    for f in dataclasses.fields(self):
+      value = getattr(self, f.name)
+      if f.name in ('kjob_decorator_fn', 'jobset_decorator_fn') and value:
+        parts.append(f'{f.name}=<function {value.__name__}>')
+      else:
+        parts.append(f'{f.name}={repr(value)}')
+    return f"GpuConfig({', '.join(parts)})"
+
+
 @dataclass
 class SystemCharacteristics:
   """Contains the defining characteristics of a specific accelerator system.
@@ -92,12 +149,28 @@ class SystemCharacteristics:
   accelerator_type: AcceleratorType
   device_type: str
   supports_sub_slicing: bool
+  docker_platform: DockerPlatform
   requires_workload_policy: bool = False
+  gpu_config: Optional[GpuConfig] = None
 
   def __post_init__(self):
     if self.accelerator_type == AcceleratorType.GPU:
       self.requires_workload_policy = True
 
+      if self.gpu_config is None:
+        raise ValueError(
+            f"Validation Error: System '{self.device_type}' is a GPU, "
+            "but 'gpu_config' was not provided."
+        )
+
+  @property
+  def gpu_requires_topology(self) -> bool:
+    """
+    Safely returns whether the GPU config requires topology,
+    defaulting to False if no GPU config exists.
+    """
+    return self.gpu_config.requires_topology if self.gpu_config else False
+
 
 def get_system_characteristics(
     args,
@@ -167,6 +240,7 @@ def get_tpu_system_characteristics_map(
     machine_type: str,
     supported_topologies: list[str],
     supports_sub_slicing: bool,
+    docker_platform: DockerPlatform,
     tpu_type_requires_workload_policy: bool = False,
     default_topologies: set[str] | None = None,
 ) -> dict[str, SystemCharacteristics]:
@@ -189,6 +263,7 @@ def get_tpu_system_characteristics_map(
         requires_workload_policy=tpu_type_requires_workload_policy
         and vms_per_slice > 1,
         supports_sub_slicing=supports_sub_slicing,
+        docker_platform=docker_platform,
     )
     system_characteristics_map[f'{prefix}-{topology}'] = system
     if (
@@ -231,6 +306,8 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='l4-1',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(requires_topology=False),
+        docker_platform=AMD_PLATFORM,
     ),
     'l4-2': SystemCharacteristics(
         topology='N/A',
@@ -241,6 +318,8 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='l4-2',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(requires_topology=False),
+        docker_platform=AMD_PLATFORM,
     ),
     'l4-4': SystemCharacteristics(
         topology='N/A',
@@ -251,6 +330,8 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='l4-4',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(requires_topology=False),
+        docker_platform=AMD_PLATFORM,
     ),
     'l4-8': SystemCharacteristics(
         topology='N/A',
@@ -261,6 +342,8 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='l4-8',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(requires_topology=False),
+        docker_platform=AMD_PLATFORM,
     ),
     # A100-40gb-$CHIPSc
     'a100-40gb-1': SystemCharacteristics(
@@ -272,6 +355,8 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='a100-40gb-1',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(requires_topology=False),
+        docker_platform=AMD_PLATFORM,
     ),
     'a100-40gb-2': SystemCharacteristics(
         topology='N/A',
@@ -282,6 +367,8 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='a100-40gb-2',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(requires_topology=False),
+        docker_platform=AMD_PLATFORM,
     ),
     'a100-40gb-4': SystemCharacteristics(
         topology='N/A',
@@ -292,6 +379,8 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='a100-40gb-4',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(requires_topology=False),
+        docker_platform=AMD_PLATFORM,
     ),
     'a100-40gb-8': SystemCharacteristics(
         topology='N/A',
@@ -302,6 +391,8 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='a100-40gb-8',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(requires_topology=False),
+        docker_platform=AMD_PLATFORM,
     ),
     'gb200-4': SystemCharacteristics(
         topology='1x72',
@@ -312,6 +403,14 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='gb200-4',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(
+            requires_topology=True,
+            nccl_installer=INSTALLER_NCCL_RDMA_A4X,
+            kjob_decorator_fn=rdma_decorator.decorate_kjob_template,
+            jobset_decorator_fn=rdma_decorator.decorate_jobset,
+            gpu_direct_name='rdma',
+        ),
+        docker_platform=ARM_PLATFORM,
     ),
     'gb200-4-nolssd': SystemCharacteristics(
         topology='1x72',
@@ -322,6 +421,14 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='gb200-4',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(
+            requires_topology=True,
+            nccl_installer=INSTALLER_NCCL_RDMA_A4X,
+            kjob_decorator_fn=rdma_decorator.decorate_kjob_template,
+            jobset_decorator_fn=rdma_decorator.decorate_jobset,
+            gpu_direct_name='rdma',
+        ),
+        docker_platform=ARM_PLATFORM,
     ),
     'b200-8': SystemCharacteristics(
         topology='N/A',
@@ -332,6 +439,14 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='b200-8',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(
+            requires_topology=True,
+            nccl_installer=INSTALLER_NCCL_RDMA,
+            kjob_decorator_fn=rdma_decorator.decorate_kjob_template,
+            jobset_decorator_fn=rdma_decorator.decorate_jobset,
+            gpu_direct_name='rdma',
+        ),
+        docker_platform=AMD_PLATFORM,
     ),
     'h200-141gb-8': SystemCharacteristics(
         topology='N/A',
@@ -342,6 +457,14 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='h200-141gb-8',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(
+            requires_topology=True,
+            nccl_installer=INSTALLER_NCCL_RDMA,
+            kjob_decorator_fn=rdma_decorator.decorate_kjob_template,
+            jobset_decorator_fn=rdma_decorator.decorate_jobset,
+            gpu_direct_name='rdma',
+        ),
+        docker_platform=AMD_PLATFORM,
     ),
     # H100-80gb-$CHIPS
     'h100-80gb-8': SystemCharacteristics(
@@ -353,6 +476,14 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='h100-80gb-8',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(
+            requires_topology=True,
+            nccl_installer=INSTALLER_NCCL_TCPX,
+            kjob_decorator_fn=tcpx_decorator.decorate_kjob_template,
+            jobset_decorator_fn=tcpx_decorator.decorate_jobset,
+            gpu_direct_name='tcpx',
+        ),
+        docker_platform=AMD_PLATFORM,
     ),
     # H100-mega-80gb-$CHIPS
     'h100-mega-80gb-8': SystemCharacteristics(
@@ -364,6 +495,14 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.GPU,
         device_type='h100-mega-80gb-8',
         supports_sub_slicing=False,
+        gpu_config=GpuConfig(
+            requires_topology=True,
+            nccl_installer=INSTALLER_NCCL_TCPXO,
+            kjob_decorator_fn=tcpxo_decorator.decorate_kjob_template,
+            jobset_decorator_fn=tcpxo_decorator.decorate_jobset,
+            gpu_direct_name='tcpxo',
+        ),
+        docker_platform=AMD_PLATFORM,
     ),
     # TPU system characteristics
     **get_tpu_system_characteristics_map(
@@ -374,6 +513,7 @@ UserFacingNameToSystemCharacteristics = {
         supported_topologies=['1x1x1'],
         tpu_type_requires_workload_policy=True,
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     **get_tpu_system_characteristics_map(
         prefix='tpu7x',
@@ -382,6 +522,7 @@ UserFacingNameToSystemCharacteristics = {
         machine_type='tpu7x-standard-4t',
         tpu_type_requires_workload_policy=True,
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
         supported_topologies=generate_tpu_topologies(max_cubes=144),
         default_topologies=set([
             '12x12x12',
@@ -491,6 +632,7 @@ UserFacingNameToSystemCharacteristics = {
         machine_type='ct6e-standard-1t',
         supports_sub_slicing=False,
         supported_topologies=['1x1'],
+        docker_platform=AMD_PLATFORM,
     ),
     **get_tpu_system_characteristics_map(
         prefix='v6e',
@@ -501,6 +643,7 @@ UserFacingNameToSystemCharacteristics = {
         supported_topologies=[
             '2x2',
         ],
+        docker_platform=AMD_PLATFORM,
     ),
     **get_tpu_system_characteristics_map(
         prefix='v6e',
@@ -509,6 +652,7 @@ UserFacingNameToSystemCharacteristics = {
         machine_type='ct6e-standard-4t',
         supports_sub_slicing=True,
         supported_topologies=SUB_SLICING_TOPOLOGIES,
+        docker_platform=AMD_PLATFORM,
     ),
     **get_tpu_system_characteristics_map(
         prefix='v5p',
@@ -516,6 +660,7 @@ UserFacingNameToSystemCharacteristics = {
         gke_accelerator='tpu-v5p-slice',
         machine_type='ct5p-hightpu-4t',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
         supported_topologies=generate_tpu_topologies(max_cubes=140),
         default_topologies=set([
             '2x2x1',
@@ -621,6 +766,7 @@ UserFacingNameToSystemCharacteristics = {
         tensorcores_per_chip=1,
         gke_accelerator='tpu-v5-lite-podslice',
         machine_type='ct5lp-hightpu-4t',
+        docker_platform=AMD_PLATFORM,
         supports_sub_slicing=False,
         supported_topologies=['2x4', '4x4', '4x8', '8x8', '8x16', '16x16'],
     ),
@@ -629,6 +775,7 @@ UserFacingNameToSystemCharacteristics = {
         tensorcores_per_chip=2,
         gke_accelerator='tpu-v4-podslice',
         machine_type='ct4p-hightpu-4t',
+        docker_platform=AMD_PLATFORM,
         supports_sub_slicing=False,
         supported_topologies=generate_tpu_topologies(
             max_cubes=64, enforce_nondecreasing=False
@@ -660,6 +807,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='m1-megamem-96-1',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     # n2-standard-#vCPUs-#VMs
     'n2-standard-64-1': SystemCharacteristics(
@@ -671,6 +819,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-64-1',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-1': SystemCharacteristics(
         topology='N/A',
@@ -681,6 +830,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-1',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-2': SystemCharacteristics(
         topology='N/A',
@@ -691,6 +841,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-2',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-4': SystemCharacteristics(
         topology='N/A',
@@ -701,6 +852,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-4',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-8': SystemCharacteristics(
         topology='N/A',
@@ -711,6 +863,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-8',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-16': SystemCharacteristics(
         topology='N/A',
@@ -721,6 +874,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-16',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-32': SystemCharacteristics(
         topology='N/A',
@@ -731,6 +885,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-32',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-64': SystemCharacteristics(
         topology='N/A',
@@ -741,6 +896,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-64',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-128': SystemCharacteristics(
         topology='N/A',
@@ -751,6 +907,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-128',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-256': SystemCharacteristics(
         topology='N/A',
@@ -761,6 +918,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-256',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-512': SystemCharacteristics(
         topology='N/A',
@@ -771,6 +929,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-512',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-1024': SystemCharacteristics(
         topology='N/A',
@@ -781,6 +940,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-1024',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
     'n2-standard-32-2048': SystemCharacteristics(
         topology='N/A',
@@ -791,6 +951,7 @@ UserFacingNameToSystemCharacteristics = {
         accelerator_type=AcceleratorType.CPU,
         device_type='n2-standard-32-2048',
         supports_sub_slicing=False,
+        docker_platform=AMD_PLATFORM,
     ),
 }
 """ If you modify UserFacingNameToSystemCharacteristics you should also modify
@@ -808,3 +969,21 @@ def get_system_characteristics_keys_by_accelerator_type(
       for key, value in UserFacingNameToSystemCharacteristics.items()
       if value.accelerator_type in accelerators
   ]
+
+
+def create_accelerator_label(system: SystemCharacteristics) -> str:
+  if system.accelerator_type == AcceleratorType.CPU:
+    return ''
+  return (
+      f'{AcceleratorTypeToAcceleratorCharacteristics[system.accelerator_type].accelerator_label}:'
+      f' {system.gke_accelerator}'
+  )
+
+
+def create_machine_label(system: SystemCharacteristics) -> str:
+  if system.accelerator_type == AcceleratorType.TPU:
+    return (
+        f'{AcceleratorTypeToAcceleratorCharacteristics[system.accelerator_type].machine_label}:'
+        f' {system.topology}'
+    )
+  return ''

xpk/core/system_characteristics_test.py

@@ -14,7 +14,15 @@ See the License for the specific language governing permissions and
 limitations under the License.
 """
 
-from .system_characteristics import get_tpu_system_characteristics_map, generate_tpu_topologies, SystemCharacteristics, AcceleratorType
+import pytest
+from .system_characteristics import (
+    get_tpu_system_characteristics_map,
+    generate_tpu_topologies,
+    DockerPlatform,
+    SystemCharacteristics,
+    AcceleratorType,
+    GpuConfig,
+)
 
 
 def test_get_tpu_system_characteristics_map_returns_correct_values_for_1x1_topology():
@@ -25,6 +33,7 @@ def test_get_tpu_system_characteristics_map_returns_correct_values_for_1x1_topol
       machine_type="test",
       supported_topologies=["1x1"],
       supports_sub_slicing=False,
+      docker_platform=DockerPlatform.AMD,
       tpu_type_requires_workload_policy=False,
   )
 
@@ -37,6 +46,7 @@ def test_get_tpu_system_characteristics_map_returns_correct_values_for_1x1_topol
       accelerator_type=AcceleratorType.TPU,
       device_type="test-1",
       supports_sub_slicing=False,
+      docker_platform=DockerPlatform.AMD,
       requires_workload_policy=False,
   )
   assert result == {
@@ -53,6 +63,7 @@ def test_get_tpu_system_characteristics_map_returns_correct_values_for_2x2_topol
       machine_type="test",
       supported_topologies=["2x2"],
       supports_sub_slicing=False,
+      docker_platform=DockerPlatform.AMD,
       tpu_type_requires_workload_policy=True,
   )
 
@@ -65,6 +76,7 @@ def test_get_tpu_system_characteristics_map_returns_correct_values_for_2x2_topol
       accelerator_type=AcceleratorType.TPU,
       device_type="test-8",
       supports_sub_slicing=False,
+      docker_platform=DockerPlatform.AMD,
       requires_workload_policy=False,
   )
   assert result == {
@@ -81,6 +93,7 @@ def test_get_tpu_system_characteristics_map_returns_correct_values_for_2x2x2_top
       machine_type="test",
       supported_topologies=["2x2x2"],
       supports_sub_slicing=False,
+      docker_platform=DockerPlatform.AMD,
       tpu_type_requires_workload_policy=True,
   )
 
@@ -93,6 +106,7 @@ def test_get_tpu_system_characteristics_map_returns_correct_values_for_2x2x2_top
       accelerator_type=AcceleratorType.TPU,
       device_type="test-16",
       supports_sub_slicing=False,
+      docker_platform=DockerPlatform.AMD,
       requires_workload_policy=True,
   )
   assert result == {
@@ -109,6 +123,7 @@ def test_get_tpu_system_characteristics_map_prefers_default_topologies():
       machine_type="test",
       supported_topologies=["4x4x4", "4x4x32", "4x8x16", "8x8x8"],
       supports_sub_slicing=False,
+      docker_platform=DockerPlatform.AMD,
       default_topologies=set(["4x8x16"]),
   )
 
@@ -146,3 +161,36 @@ def test_generate_tpu_topologies_contains_sub_cube_slices():
   one_cube = generate_tpu_topologies(max_cubes=1)
 
   assert one_cube == ["2x2x1", "2x2x2", "2x2x4", "2x4x4", "4x4x4"]
+
+
+def test_system_characteristics_post_init_sets_workload_policy_for_gpu():
+  """Tests that __post_init__ correctly sets requires_workload_policy for GPUs."""
+  gpu_system = SystemCharacteristics(
+      topology="N/A",
+      vms_per_slice=1,
+      gke_accelerator="nvidia-l4",
+      gce_machine_type="g2-standard-12",
+      chips_per_vm=1,
+      accelerator_type=AcceleratorType.GPU,
+      device_type="l4-1",
+      supports_sub_slicing=False,
+      docker_platform=DockerPlatform.AMD,
+      gpu_config=GpuConfig(requires_topology=False),
+  )
+  assert gpu_system.requires_workload_policy is True
+
+
+def test_system_characteristics_post_init_throws_for_gpu_without_config():
+  """Tests that __post_init__ raises ValueError for GPU without gpu_config."""
+  with pytest.raises(ValueError, match="'gpu_config' was not provided"):
+    SystemCharacteristics(
+        topology="N/A",
+        vms_per_slice=1,
+        gke_accelerator="nvidia-l4",
+        gce_machine_type="g2-standard-12",
+        chips_per_vm=1,
+        accelerator_type=AcceleratorType.GPU,
+        device_type="l4-1",
+        supports_sub_slicing=False,
+        docker_platform=DockerPlatform.AMD,
+    )

xpk/core/telemetry.py

@@ -27,7 +27,7 @@ import requests
 from enum import Enum
 from typing import Any
 from dataclasses import dataclass
-from .config import xpk_config, CLIENT_ID_KEY, SEND_TELEMETRY_KEY, __version__ as xpk_version
+from .config import get_config, CLIENT_ID_KEY, SEND_TELEMETRY_KEY, __version__ as xpk_version
 from ..utils.execution_context import is_dry_run
 from ..utils.user_agent import get_user_agent
 from ..utils.feature_flags import FeatureFlags
@@ -36,7 +36,7 @@ from ..utils.feature_flags import FeatureFlags
 def should_send_telemetry():
   return (
       FeatureFlags.TELEMETRY_ENABLED
-      and xpk_config.get(SEND_TELEMETRY_KEY) != "false"
+      and get_config().get(SEND_TELEMETRY_KEY) != "false"
   )
 
 
@@ -254,10 +254,10 @@ def _get_session_id() -> str:
 
 def _ensure_client_id() -> str:
   """Generates Client ID and stores in configuration if not already present."""
-  current_client_id = xpk_config.get(CLIENT_ID_KEY)
+  current_client_id = get_config().get(CLIENT_ID_KEY)
   if current_client_id is not None:
     return current_client_id
 
   new_client_id = str(uuid.uuid4())
-  xpk_config.set(CLIENT_ID_KEY, new_client_id)
+  get_config().set(CLIENT_ID_KEY, new_client_id)
   return new_client_id

xpk/core/telemetry_test.py

@@ -16,7 +16,7 @@ limitations under the License.
 
 import pytest
 import json
-from .config import xpk_config, CLIENT_ID_KEY, SEND_TELEMETRY_KEY
+from .config import get_config, CLIENT_ID_KEY, SEND_TELEMETRY_KEY
 from .telemetry import MetricsCollector, MetricsEventMetadataKey, should_send_telemetry
 from ..utils.execution_context import set_dry_run
 from ..utils.feature_flags import FeatureFlags
@@ -31,9 +31,9 @@ def setup_mocks(mocker: MockerFixture):
   mocker.patch('os.path.basename', return_value='xpk.py')
   mocker.patch('os.path.abspath', return_value='/home/xpk_user')
   set_dry_run(False)
-  xpk_config.set(CLIENT_ID_KEY, 'client_id')
+  get_config().set(CLIENT_ID_KEY, 'client_id')
   yield
-  xpk_config.set(CLIENT_ID_KEY, None)
+  get_config().set(CLIENT_ID_KEY, None)
 
 
 @pytest.mark.parametrize(
@@ -48,13 +48,13 @@ def setup_mocks(mocker: MockerFixture):
 def test_should_send_telemetry_returns_correct_value(
     feature_flag: bool, config_value: str, expected: bool
 ):
-  xpk_config.set(SEND_TELEMETRY_KEY, config_value)
+  get_config().set(SEND_TELEMETRY_KEY, config_value)
   FeatureFlags.TELEMETRY_ENABLED = feature_flag
   assert should_send_telemetry() is expected
 
 
 def test_metrics_collector_generates_client_id_if_not_present():
-  xpk_config.set(CLIENT_ID_KEY, None)
+  get_config().set(CLIENT_ID_KEY, None)
   MetricsCollector.log_start(command='test')
   payload = json.loads(MetricsCollector.flush())
   extension_json = json.loads(payload['log_event'][0]['source_extension_json'])
@@ -79,12 +79,12 @@ def test_metrics_collector_logs_start_event_correctly():
       ],
       'event_name': 'start',
       'event_type': 'commands',
-      'release_version': 'v0.15.0',
+      'release_version': 'v0.0.0',
   }
 
 
 def test_metrics_collector_generates_client_id_when_not_present():
-  xpk_config.set(CLIENT_ID_KEY, None)
+  get_config().set(CLIENT_ID_KEY, None)
   MetricsCollector.log_start(command='test')
   payload = json.loads(MetricsCollector.flush())
   extension_json = json.loads(payload['log_event'][0]['source_extension_json'])
@@ -109,7 +109,7 @@ def test_metrics_collector_logs_complete_event_correctly():
       ],
       'event_name': 'complete',
       'event_type': 'commands',
-      'release_version': 'v0.15.0',
+      'release_version': 'v0.0.0',
   }
 
 
@@ -132,7 +132,7 @@ def test_metrics_collector_logs_custom_event_correctly():
       ],
       'event_name': 'test',
       'event_type': 'custom',
-      'release_version': 'v0.15.0',
+      'release_version': 'v0.0.0',
   }
 
 

xpk/core/vertex.py

@@ -15,7 +15,7 @@ limitations under the License.
 """
 
 from ..utils.console import xpk_print
-from .resources import CLUSTER_METADATA_CONFIGMAP, get_cluster_configmap
+from .resources import ConfigMapType, get_cluster_configmap
 
 DEFAULT_VERTEX_TENSORBOARD_NAME = 'tb-instance'
 
@@ -65,8 +65,9 @@ def create_vertex_experiment(args) -> dict | None:
       tensorboard,
   )
 
-  metadata_configmap_name = f'{args.cluster}-{CLUSTER_METADATA_CONFIGMAP}'
-  cluster_config_map = get_cluster_configmap(metadata_configmap_name)
+  cluster_config_map = get_cluster_configmap(
+      args.cluster, ConfigMapType.METADATA
+  )
 
   if cluster_config_map is None or 'tensorboard_name' not in cluster_config_map:
     xpk_print(