xpk 0.17.3__py3-none-any.whl → 1.1.0__py3-none-any.whl

xpk/core/cluster.py

@@ -158,7 +158,7 @@ def install_nri_on_cluster() -> int:
 
 
 def get_cluster_nodes_info() -> list[dict]:
-  """Get list of cluster's nodes descrition in yaml format
+  """Get list of cluster's nodes description in yaml format
 
   Returns:
     List of nodes info yaml objects.
@@ -391,14 +391,15 @@ def project_id_to_project_number(project_id: str) -> str:
 
 
 def setup_k8s_env(args) -> k8s_client.ApiClient:
-  if not getattr(args, 'kind_cluster', False):
-    add_zone_and_project(args)
-    get_cluster_credentials(args)
-    args.project_number = (
-        project_id_to_project_number(args.project)
-        if not args.dry_run
-        else abs(hash(args.project) % (10**12))  # 12 digit hash
-    )
+  add_zone_and_project(args)
+  get_cluster_credentials(args)
+  # Use provided project number if available, otherwise fetch via API
+  if getattr(args, 'project_number', None):
+    xpk_print(f'Using provided project number: {args.project_number}')
+  elif args.dry_run:
+    args.project_number = abs(hash(args.project) % (10**12))  # 12 digit hash
+  else:
+    args.project_number = project_id_to_project_number(args.project)
 
   config.load_kube_config()
   return k8s_client.ApiClient()
@@ -750,6 +751,6 @@ def _get_credentials(
 def _are_credentials_valid() -> bool:
   kubectl_command = 'kubectl get pods'
   kubectl_return_code = run_command_with_updates(
-      kubectl_command, 'Test kubectl credentials'
+      kubectl_command, 'Test kubectl credentials', verbose=False
   )
   return kubectl_return_code == 0

xpk/core/cluster_private.py

@@ -61,7 +61,7 @@ def authorize_private_cluster_access_if_necessary(args) -> int:
   if new_authorized_networks_needed or not is_current_machine_in_network:
     return update_cluster_new_authorized_networks(args, authorized_networks)
 
-  xpk_print("Current machine's IP adrress is already authorized.")
+  xpk_print("Current machine's IP address is already authorized.")
   return 0
 
 
@@ -84,7 +84,7 @@ def add_current_machine_to_networks_if_needed(
       is_current_machine_in_any_network(authorized_networks)
   )
   if is_current_machine_in_network_return_code != 0:
-    xpk_print("Error on checking current machine's IP adrress.")
+    xpk_print("Error on checking current machine's IP address.")
     return is_current_machine_in_network_return_code, False, authorized_networks
 
   if not is_current_machine_in_network:
@@ -148,7 +148,7 @@ def is_cluster_private(args) -> bool:
 
 
 def get_cluster_authorized_networks(args) -> list[str]:
-  """Retreives the networks list that are authorized to have access to Control Plane.
+  """Retrieves the networks list that are authorized to have access to Control Plane.
   Args:
     args: user provided arguments for running the command.
 

xpk/core/cluster_test.py

@@ -41,11 +41,15 @@ def command_args(mocker: MockerFixture):
   return mocker.Mock(cluster="cluster", project="project", zone="zone")
 
 
-def test_get_cluster_credentials_returns_1_when_retrieval_command_fails(
+def test_get_cluster_credentials_returns_1_when_retrieval_commands_fail(
     commands_tester: CommandsTester, command_args
 ):
   commands_tester.set_result_for_command(
-      (1, ""), "gcloud container clusters get-credentials"
+      (1, ""), "gcloud container clusters get-credentials", " --dns-endpoint"
+  )
+  commands_tester.set_result_for_command(
+      (1, ""),
+      "gcloud container clusters get-credentials",
   )
   assert get_cluster_credentials(command_args) == 1
 
@@ -95,6 +99,29 @@ def test_get_cluster_credentials_retries_without_dns_when_dns_retrieval_fails(
   assert len(non_dns_endpoint_commands) == 1
 
 
+def test_get_cluster_credentials_retries_without_dns_when_dns_retrieval_returns_error(
+    commands_tester: CommandsTester, command_args
+):
+  commands_tester.set_result_for_command(
+      (1, ""), "gcloud container clusters get-credentials", "--dns-endpoint"
+  )
+  commands_tester.set_result_for_command(
+      (0, ""),
+      "gcloud container clusters get-credentials",
+  )
+
+  assert get_cluster_credentials(command_args) == 0
+
+  non_dns_endpoint_commands = [
+      c
+      for c in commands_tester.get_matching_commands(
+          "gcloud container clusters get-credentials"
+      )
+      if "dns-endpoint" not in c
+  ]
+  assert len(non_dns_endpoint_commands) == 1
+
+
 def test_update_cluster_with_lustre_driver_if_necessary_with_default_port_runs_correct_checks(
     commands_tester: CommandsTester, command_args
 ):

xpk/core/config.py

@@ -19,6 +19,7 @@ import os
 import ruamel.yaml
 from abc import ABC, abstractmethod
 from ..utils import file
+from ..utils.execution_context import is_dry_run
 from ..utils.console import xpk_print
 from setuptools_scm import get_version as setuptools_get_version
 from importlib.metadata import version, PackageNotFoundError
@@ -96,8 +97,7 @@ class FileSystemConfig(Config):
     self._allowed_keys = DEFAULT_KEYS
 
   def _open_configs(self) -> dict | None:
-    dir_path = '/'.join(self._config.split('/')[:-1])
-    file.ensure_directory_exists(dir_path)
+    file.ensure_directory_exists(os.path.dirname(self._config))
 
     if not os.path.exists(self._config):
       return None
@@ -107,6 +107,9 @@ class FileSystemConfig(Config):
       return config_yaml
 
   def _save_configs(self, config_yaml: dict) -> None:
+    if is_dry_run():
+      return None
+
     with open(self._config, encoding='utf-8', mode='w') as stream:
       yaml.dump(config_yaml, stream)
 

xpk/core/docker_container.py

@@ -17,9 +17,7 @@ limitations under the License.
 from ..utils.console import xpk_exit, xpk_print
 from .docker_image import setup_docker_image
 from .docker_resources import (
-    add_container_ports,
     add_image_pull_policy_for_pw_or_gpu,
-    add_jax_coordinator_port,
     get_env_container,
     get_main_container_resources,
     get_volume_mounts,
@@ -112,13 +110,12 @@ def get_main_container(args, system, docker_image, resource_type) -> str:
         'touch /shared-volume/stacktrace_signal; '
     )
 
-  yaml = """- name: {docker_name}
+  containers = []
+  container_yaml = """
+              - name: {docker_name}
                 image: {docker_image}
                 {image_pull_policy}
                 env: {env}
-                ports:
-                {container_ports}
-                {jax_coordinator_port}
                 securityContext:
                   privileged: true
                 command:
@@ -145,29 +142,39 @@ def get_main_container(args, system, docker_image, resource_type) -> str:
                   limits:
                     {resources}
 """
+  docker_name = get_main_container_docker_image(args, system)
   volume_mounts = get_volume_mounts(args, system)
   if volume_mounts != '':
-    yaml += """
+    container_yaml += """
                 volumeMounts:
                 {volume_mounts}
 """
-  return yaml.format(
-      args=args,
-      system=system,
-      image_pull_policy=add_image_pull_policy_for_pw_or_gpu(args, system),
-      env=get_env_container(args, system),
-      container_ports=add_container_ports(args, system),
-      jax_coordinator_port=add_jax_coordinator_port(system),
-      docker_name=get_main_container_docker_image(args, system),
-      docker_image=docker_image,
-      gsutil_test_command=gsutil_test_command,
-      command=command,
-      tpu_stacktrace_terminate_command=tpu_stacktrace_terminate_command,
-      gpu_workload_terminate_command=gpu_workload_terminate_command,
-      xpk_internal_commands=xpk_internal_commands,
-      resources=get_main_container_resources(args, system, resource_type),
-      volume_mounts=volume_mounts,
-  )
+  # pathways job running on 2 parallel containers is not verified yet
+  if args.use_pathways:
+    system.parallel_containers = 1
+
+  env = get_env_container(args, system)
+  image_pull_policy = add_image_pull_policy_for_pw_or_gpu(args, system)
+  for i in range(system.parallel_containers):
+    docker_name_sufix = f'-{i + 1}' if system.parallel_containers > 1 else ''
+    containers.append(
+        container_yaml.format(
+            args=args,
+            system=system,
+            image_pull_policy=image_pull_policy,
+            env=env,
+            docker_name=f'{docker_name}{docker_name_sufix}',
+            docker_image=docker_image,
+            gsutil_test_command=gsutil_test_command,
+            command=command,
+            tpu_stacktrace_terminate_command=tpu_stacktrace_terminate_command,
+            gpu_workload_terminate_command=gpu_workload_terminate_command,
+            xpk_internal_commands=xpk_internal_commands,
+            resources=get_main_container_resources(args, system, resource_type),
+            volume_mounts=volume_mounts,
+        )
+    )
+  return ''.join(containers)
 
 
 def get_user_workload_container(args, system: SystemCharacteristics):

xpk/core/docker_manager.py

@@ -44,7 +44,7 @@ class CommandRunner(ABC):
 
   @abstractmethod
   def initialize(self) -> None:
-    """initialize is a method that should implement all steps neccessary to run command.
+    """initialize is a method that should implement all steps necessary to run command.
 
     Returns:
         None
@@ -95,7 +95,7 @@ class DockerManager(CommandRunner):
     - gcloud_cfg_path (str) : path to directory containing gcloud configuration
     - working_dir (str) : path to directory in which gcluster deployment directory will be saved
     - client (DockerClient) : docker client
-    - nocache (bool) : wheter to use docker cache when building image
+    - nocache (bool) : whether to use docker cache when building image
     - img_name (str) : name of docker image to create
     - container_name (str) : name of the container that will be created from img_name
     - rm_container_after (bool) : if set to True, docker container in which command is executed will be removed after each execution.
@@ -294,12 +294,12 @@ class DockerManager(CommandRunner):
       xpk_print(f"error while building image {self.img_name}: {e.msg}")
       xpk_exit(dockerBuildErrorCode)
     except APIError as e:
-      xpk_print(f"erro while building image {self.img_name}: {e.explanation}")
+      xpk_print(f"error while building image {self.img_name}: {e.explanation}")
       xpk_exit(dockerBuildErrorCode)
     except TypeError as e:
       xpk_print(f"TypeError while building image {self.img_name}: {e.args}")
       xpk_exit(dockerBuildErrorCode)
-    xpk_print("Docker image build succesfully.")
+    xpk_print("Docker image build successfully.")
     os.remove(self.dockerfile_path)
     tmp_dockerfile_dir = "/".join(self.dockerfile_path.split("/")[:-1])
     os.rmdir(tmp_dockerfile_dir)

xpk/core/docker_resources.py

@@ -53,7 +53,10 @@ def get_main_container_resources(
     offset_vCPUs = int(system.chips_per_vm) * 0.95
     return f'{resource_type}: {offset_vCPUs}'
 
-  return f'{resource_type}: {system.chips_per_vm}'
+  return (
+      f'{resource_type}:'
+      f' {int(system.chips_per_vm / system.parallel_containers)}'
+  )
 
 
 def get_env_container(args, system: SystemCharacteristics) -> str:

xpk/core/kueue_manager.py

@@ -41,8 +41,8 @@ from ..utils.console import xpk_print, xpk_exit, ask_for_user_consent
 from ..utils.templates import TEMPLATE_PATH, get_templates_absolute_path
 from packaging.version import Version
 
-KUEUE_VERSION = Version("v0.14.3")
-LATEST_BREAKING_VERSION = Version("v0.14.0")
+KUEUE_VERSION = Version("v0.15.2")
+LATEST_BREAKING_VERSION = Version("v0.15.0")
 WAIT_FOR_KUEUE_TIMEOUT = "10m"
 CLUSTER_QUEUE_NAME = "cluster-queue"
 LOCAL_QUEUE_NAME = "multislice-queue"
@@ -290,6 +290,7 @@ class KueueManager:
         cpu_limit=cpu_limit,
         memory_limit=memory_limit,
         topology_name=topology_name,
+        configure_super_slicing=kueue_config.configure_super_slicing,
     )
 
     config_yaml = template.render(context)
@@ -316,6 +317,7 @@ class KueueManager:
       cpu_limit: int,
       memory_limit: str,
       topology_name: str | None,
+      configure_super_slicing: bool,
   ) -> Dict[str, Any]:
     """Prepares the context for the Jinja2 template."""
     # Main accelerator flavor
@@ -328,11 +330,7 @@ class KueueManager:
       key, value = accelerator_label.split(":", 1)
       node_labels_dict[key] = value.strip()
 
-    if system.supports_super_slicing:
-      node_labels_dict["cloud.google.com/gke-tpu-partition-4x4x4-state"] = (
-          "HEALTHY"
-      )
-    elif not autoprovisioning:
+    if not autoprovisioning and not configure_super_slicing:
       machine_label = create_machine_label(system)
       if machine_label:
         key, value = machine_label.split(":", 1)
@@ -383,7 +381,7 @@ class KueueManager:
       })
 
     admission_checks = []
-    if system.supports_super_slicing:
+    if configure_super_slicing:
       admission_checks.append("ss-kueue-operator")
     if flex and is_queued_cluster(num_slices, system.accelerator_type):
       admission_checks.append("dws-prov")

xpk/core/kueue_manager_test.py

@@ -36,6 +36,7 @@ TPU_SYSTEM: SystemCharacteristics = SystemCharacteristics(
     device_type="v5p-8",
     supports_sub_slicing=False,
     supports_super_slicing=False,
+    supports_accelerator_network_profile=False,
     docker_platform=DockerPlatform.ARM,
 )
 
@@ -112,7 +113,7 @@ def test_install_or_upgrade_when_outdated(
   result = kueue_manager.install_or_upgrade(KUEUE_CONFIG)
 
   assert result == 0
-  mock_commands.assert_command_run("kubectl apply", "v0.14.3/manifests.yaml")
+  mock_commands.assert_command_run("kubectl apply", "v0.15.2/manifests.yaml")
   mock_commands.assert_command_run("kubectl apply -f", "/tmp/")
 
 
@@ -125,7 +126,7 @@ def test_install_or_upgrade_when_not_installed(
   result = kueue_manager.install_or_upgrade(KUEUE_CONFIG)
 
   assert result == 0
-  mock_commands.assert_command_run("kubectl apply", "v0.14.3/manifests.yaml")
+  mock_commands.assert_command_run("kubectl apply", "v0.15.2/manifests.yaml")
   mock_commands.assert_command_run("kubectl apply -f", "/tmp/")
 
 
@@ -134,7 +135,7 @@ def test_upgrade_when_no_breaking_changes_between_versions_no_preparation_needed
     kueue_manager: KueueManager,
     mock_ask_for_user_consent: MagicMock,
 ):
-  set_installed_kueue_version(mock_commands, Version("0.14.0"))
+  set_installed_kueue_version(mock_commands, Version("0.15.0"))
 
   kueue_manager.install_or_upgrade(KUEUE_CONFIG)
 
@@ -161,7 +162,7 @@ def test_upgrade_with_breaking_changes_between_versions_runs_preparation(
   assert result == 0
   mock_ask_for_user_consent.assert_called_once()
   assert (
-      "CHANGELOG/CHANGELOG-0.14.md"
+      "CHANGELOG/CHANGELOG-0.15.md"
       in mock_ask_for_user_consent.mock_calls[0].args[0]
   )
   mock_commands.assert_command_run(
@@ -411,6 +412,7 @@ def test_configure_generates_correct_manifest_with_gke_default_topology(
           supports_sub_slicing=False,
           supports_super_slicing=False,
           docker_platform=DockerPlatform.ARM,
+          supports_accelerator_network_profile=True,
           gpu_config=GpuConfig(requires_topology=True),
       ),
   )
@@ -490,7 +492,6 @@ def test_configure_generates_correct_manifest_with_super_slicing(
   assert resource_flavor["spec"]["topologyName"] == "super-slice-topology"
   assert resource_flavor["spec"]["nodeLabels"] == {
       "cloud.google.com/gke-tpu-accelerator": "tpu7x",
-      "cloud.google.com/gke-tpu-partition-4x4x4-state": "HEALTHY",
   }
   topology = _first(doc for doc in manifest_docs if doc["kind"] == "Topology")
   assert topology["metadata"]["name"] == "super-slice-topology"

xpk/core/nap.py

@@ -24,7 +24,8 @@ from .capacity import (
     CapacityType,
     get_capacity_node_selectors_from_capacity_type,
     get_capacity_type,
-    verify_reservation_exists,
+    get_reservations_list,
+    verify_reservations_exist,
 )
 from .commands import run_command_with_updates, run_commands
 from .gcloud_context import get_cluster_location
@@ -345,14 +346,24 @@ def get_autoprovisioning_node_selector_args(args) -> tuple[str, int]:
       )
       if return_code != 0:
         return node_selector_args, return_code
-      return_code = verify_reservation_exists(args)
+      return_code = verify_reservations_exist(args)
       if return_code > 0:
         xpk_print('Unable to verify reservation name saved in config map.')
         return node_selector_args, return_code
 
   # Check if reservation id is valid. Shared function with cluster creation.
+  reservation_name = None
+  if capacity_type_str == CapacityType.RESERVATION.name:
+    reservations = get_reservations_list(args)
+    if len(reservations) > 1:
+      xpk_print('Error: NAP based clusters only support a single reservation.')
+      return node_selector_args, 1
+    reservation_name = reservations[0] if len(reservations) > 0 else None
+
   node_selector_args, return_code = (
-      get_capacity_node_selectors_from_capacity_type(args, capacity_type_str)
+      get_capacity_node_selectors_from_capacity_type(
+          capacity_type_str, reservation_name
+      )
   )
   if return_code != 0:
     xpk_print('Unable to get node selectors from capacity type.')

xpk/core/nodepool.py

@@ -14,7 +14,8 @@ See the License for the specific language governing permissions and
 limitations under the License.
 """
 
-from typing import List
+from typing import Iterator, List
+from itertools import cycle
 
 from ..utils.feature_flags import FeatureFlags
 from ..utils.console import ask_for_user_consent, xpk_print
@@ -25,6 +26,7 @@ from .capacity import (
     CapacityType,
     get_capacity_arguments_from_capacity_type,
     get_capacity_type,
+    get_reservations_list,
     print_reservations,
 )
 from .commands import run_command_for_value, run_commands, FailedCommand
@@ -85,12 +87,6 @@ def run_gke_node_pool_create_command(
     max_nodes = system.vms_per_slice
   else:
     max_nodes = 1000
-  capacity_args, return_code = get_capacity_arguments_from_capacity_type(
-      args, capacity_type, max_nodes, system.accelerator_type
-  )
-  if return_code > 0:
-    xpk_print('Parsing capacity arguments failed!')
-    return return_code
 
   desired_node_pool_count = (
       1 if system.accelerator_type == AcceleratorType.GPU else args.num_slices
@@ -274,9 +270,34 @@ def run_gke_node_pool_create_command(
 
   create_commands = []
   create_task_names = []
-  for node_pool_name in desired_node_pool_names:
-    if node_pool_name in node_pools_to_remain:
-      continue
+  node_pools_to_create = [
+      np for np in desired_node_pool_names if np not in node_pools_to_remain
+  ]
+
+  reservations_iter: Iterator[str] | None = None
+  if capacity_type == CapacityType.RESERVATION:
+    reservations = get_reservations_list(args)
+    if (
+        _validate_reservation_count(reservations, len(node_pools_to_create))
+        != 0
+    ):
+      return 1
+    reservations_iter = (
+        cycle(reservations) if len(reservations) == 1 else iter(reservations)
+    )
+
+  for node_pool_name in node_pools_to_create:
+    capacity_args, return_code = get_capacity_arguments_from_capacity_type(
+        args,
+        capacity_type,
+        max_nodes,
+        system.accelerator_type,
+        reservation_name=next(reservations_iter) if reservations_iter else None,
+    )
+    if return_code > 0:
+      xpk_print('Parsing capacity arguments failed!')
+      return return_code
+
     command = (
         'gcloud beta container node-pools create'
         f' {node_pool_name}'
@@ -289,6 +310,12 @@ def run_gke_node_pool_create_command(
         f'{placement_args}'
         ' --enable-gvnic'
     )
+
+    if system.supports_accelerator_network_profile:
+      command += (
+          ' --accelerator-network-profile=auto'
+          ' --node-labels=cloud.google.com/gke-networking-dra-driver=true'
+      )
     if system.accelerator_type == AcceleratorType.TPU:
       command += f' --node-version={gke_node_pool_version}'
       if capacity_type == CapacityType.FLEX_START:
@@ -626,7 +653,7 @@ def ensure_resource_policy_exists(
 ) -> None:
   return_code, _ = run_command_for_value(
       (
-          'gcloud compute resource-policies describe'
+          'gcloud beta compute resource-policies describe'
           f' {resource_policy_name}'
           f' --project={project}'
           f' --region={zone_to_region(zone)}'
@@ -637,13 +664,12 @@ def ensure_resource_policy_exists(
   if return_code == 0:
     return
 
-  # TODO: b/465696970 - Verify the flag below before launching SUPER_SLICING:
   accelerator_topology_mode = (
       ' --accelerator-topology-mode=PROVISION_ONLY' if super_slicing else ''
   )
   return_code, _ = run_command_for_value(
       (
-          'gcloud compute resource-policies create workload-policy'
+          'gcloud beta compute resource-policies create workload-policy'
           f' {resource_policy_name} --project={project} --region={zone_to_region(zone)} --type=HIGH_THROUGHPUT'
           f' --accelerator-topology={topology}{accelerator_topology_mode}'
       ),
@@ -652,3 +678,16 @@ def ensure_resource_policy_exists(
 
   if return_code != 0:
     raise RuntimeError('Unable to create resource policy')
+
+
+def _validate_reservation_count(
+    reservations: List[str], num_node_pools_to_create: int
+) -> int:
+  """Validate that reservation count matches new nodepool count or is 1."""
+  if len(reservations) > 1 and len(reservations) != num_node_pools_to_create:
+    xpk_print(
+        f'Error: Number of reservations ({len(reservations)}) must match'
+        f' the number of NEW nodepools ({num_node_pools_to_create}) or be 1.'
+    )
+    return 1
+  return 0