Refactoring - Split core.py into smaller files

.gitignore

@@ -1,5 +1,6 @@
 # editor and IDE paraphernalia
 .idea/
+.vscode/
 
 *__pycache__*
 tmp/

pytype-conf.cfg

@@ -11,12 +11,8 @@ exclude =
     src/xpk/commands
     src/xpk/core/tests
     src/xpk/core/__init__.py
-    src/xpk/core/app_profile.py
-    src/xpk/core/blueprint.py
     src/xpk/core/cluster_private.py
     src/xpk/core/commands.py
-    src/xpk/core/core.py
-    src/xpk/core/job_template.py
     src/xpk/core/kjob.py
     src/xpk/core/kueue.py
     src/xpk/core/nap.py

src/xpk/commands/batch.py

@@ -16,12 +16,12 @@
 
 from argparse import Namespace
 
+from ..core.commands import run_command_for_value
+from ..core.gcloud_context import add_zone_and_project
+from ..core.kjob import AppProfileDefaults
 from ..core.kueue import LOCAL_QUEUE_NAME
 from ..utils.console import xpk_exit, xpk_print
 from .common import set_cluster_command
-from ..core.core import add_zone_and_project
-from ..core.kjob import AppProfileDefaults
-from ..core.commands import run_command_for_value
 from .kind import set_local_cluster_command
 
 

src/xpk/commands/cluster.py

@@ -14,58 +14,56 @@
 limitations under the License.
 """
 
-from ..core.commands import run_command_for_value, run_command_with_updates
-from ..core.core import (
-    VERTEX_TENSORBOARD_FEATURE_FLAG,
-    add_zone_and_project,
-    create_cluster_configmaps,
-    create_cluster_network_config,
-    create_vertex_tensorboard,
-    delete_cluster_subnets,
+from tabulate import tabulate
+
+from ..core.capacity import H100_DEVICE_TYPE
+from ..core.cluster import (
     get_all_clusters_programmatic,
     get_cluster_credentials,
-    get_gke_control_plane_version,
-    get_gke_node_pool_version,
-    get_gke_server_config,
-    h100_device_type,
     install_nccl_on_cluster,
-    run_gke_node_pool_create_command,
     set_jobset_on_cluster,
-    set_up_cluster_network_for_gpu,
     setup_k8s_env,
     update_cluster_with_gcsfuse_driver_if_necessary,
     update_cluster_with_workload_identity_if_necessary,
-    zone_to_region,
-    get_user_input,
 )
 from ..core.cluster_private import authorize_private_cluster_access_if_necessary
-from ..core.kjob import (
-    verify_kjob_installed,
-    prepare_kjob,
-    apply_kjob_crds,
+from ..core.commands import run_command_for_value, run_command_with_updates
+from ..core.config import VERTEX_TENSORBOARD_FEATURE_FLAG
+from ..core.gcloud_context import (
+    add_zone_and_project,
+    get_gke_control_plane_version,
+    get_gke_server_config,
+    zone_to_region,
 )
+from ..core.kjob import apply_kjob_crds, prepare_kjob, verify_kjob_installed
 from ..core.kueue import (
     cluster_preheat_yml,
     install_kueue_crs,
     install_kueue_on_cluster,
     wait_for_kueue_available,
 )
 from ..core.nap import enable_autoprovisioning_on_cluster
+from ..core.network import (
+    create_cluster_network_config,
+    delete_cluster_subnets,
+    set_up_cluster_network_for_gpu,
+)
+from ..core.nodepool import get_gke_node_pool_version, run_gke_node_pool_create_command
 from ..core.ray import install_ray_cluster
+from ..core.resources import create_cluster_configmaps
 from ..core.storage import install_storage_crd
 from ..core.system_characteristics import (
     AcceleratorType,
     AcceleratorTypeToAcceleratorCharacteristics,
     SystemCharacteristics,
     get_system_characteristics,
 )
+from ..core.vertex import create_vertex_tensorboard
 from ..core.workload import get_workload_list
+from ..utils.console import get_user_input, xpk_exit, xpk_print
 from ..utils.file import write_tmp_file
-from ..utils.console import xpk_exit, xpk_print
 from . import cluster_gcluster
 
-from tabulate import tabulate
-
 
 def cluster_create(args) -> None:
   """Function around cluster creation.
@@ -148,7 +146,7 @@ def cluster_create(args) -> None:
     if set_up_cluster_network_code != 0:
       xpk_exit(set_up_cluster_network_code)
 
-  if system.device_type == h100_device_type:
+  if system.device_type == H100_DEVICE_TYPE:
     xpk_print('Creating Network Config for cluster')
     create_cluster_network_config_code = create_cluster_network_config(args)
     if create_cluster_network_config_code != 0:

src/xpk/commands/cluster_gcluster.py

@@ -14,16 +14,24 @@
 limitations under the License.
 """
 
-from ..core.blueprint.blueprint_generator import BlueprintGenerator, BlueprintGeneratorOutput, supported_device_types, a3mega_device_type, a3ultra_device_type
+import os
+
+from ..core.blueprint.blueprint_generator import (
+    BlueprintGenerator,
+    BlueprintGeneratorOutput,
+    a3mega_device_type,
+    a3ultra_device_type,
+    supported_device_types,
+)
+from ..core.capacity import get_capacity_type
 from ..core.docker_manager import DockerManager
+from ..core.gcloud_context import zone_to_region
 from ..core.gcluster_manager import GclusterManager
-from ..core.core import zone_to_region, get_capacity_type
 from ..utils.console import xpk_exit, xpk_print
-from ..utils.network import all_IPs_cidr
 from ..utils.file import ensure_directory_exists
+from ..utils.network import all_IPs_cidr
 from ..utils.objects import hash_string
 from .common import set_cluster_command
-import os
 
 blueprints_path = os.path.abspath('xpkclusters/blueprints')
 gcluster_working_dir = os.path.abspath('xpkclusters/gcluster-out')

src/xpk/commands/common.py

@@ -15,7 +15,7 @@
 """
 
 from ..core.commands import run_command_with_updates_retry
-from ..core.core import zone_to_region
+from ..core.gcloud_context import zone_to_region
 from ..utils.console import xpk_print
 
 

src/xpk/commands/info.py

@@ -14,19 +14,17 @@
 limitations under the License.
 """
 
-from ..utils.console import xpk_exit, xpk_print
-from ..core.kueue import verify_kueuectl
-from .common import set_cluster_command
-from ..core.commands import (
-    run_command_for_value,
-)
-from ..core.core import (
-    add_zone_and_project,
-)
 import json
-from tabulate import tabulate
 from argparse import Namespace
 
+from tabulate import tabulate
+
+from ..core.commands import run_command_for_value
+from ..core.gcloud_context import add_zone_and_project
+from ..core.kueue import verify_kueuectl
+from ..utils.console import xpk_exit, xpk_print
+from .common import set_cluster_command
+
 table_fmt = 'plain'
 
 

src/xpk/commands/inspector.py

@@ -14,17 +14,13 @@
 limitations under the License.
 """
 
+from ..core.cluster import get_cluster_credentials
 from ..core.commands import run_command_for_value
-from ..core.core import (
-    CLUSTER_METADATA_CONFIGMAP,
-    CLUSTER_RESOURCES_CONFIGMAP,
-    add_zone_and_project,
-    get_cluster_credentials,
-    zone_to_region,
-)
+from ..core.gcloud_context import add_zone_and_project, zone_to_region
 from ..core.kueue import CLUSTER_QUEUE_NAME, LOCAL_QUEUE_NAME
-from ..utils.file import append_tmp_file, write_tmp_file
+from ..core.resources import CLUSTER_METADATA_CONFIGMAP, CLUSTER_RESOURCES_CONFIGMAP
 from ..utils.console import xpk_exit, xpk_print
+from ..utils.file import append_tmp_file, write_tmp_file
 from .workload import get_workload_list
 
 

src/xpk/commands/job.py

@@ -14,16 +14,18 @@
 limitations under the License.
 """
 
-from .common import set_cluster_command
-from .kind import set_local_cluster_command
-from ..core.commands import run_command_for_value, run_command_with_updates
-from ..utils.console import xpk_exit, xpk_print
-from ..core.kjob import AppProfileDefaults
-from ..core.core import add_zone_and_project
-from ruamel.yaml import YAML
 import re
 import sys
 
+from ruamel.yaml import YAML
+
+from ..core.commands import run_command_for_value, run_command_with_updates
+from ..core.gcloud_context import add_zone_and_project
+from ..core.kjob import AppProfileDefaults
+from ..utils.console import xpk_exit, xpk_print
+from .common import set_cluster_command
+from .kind import set_local_cluster_command
+
 
 def job_info(args):
   """Run commands obtaining information about a job given by name.

src/xpk/commands/kind.py

@@ -18,7 +18,7 @@
     run_command_for_value,
     run_command_with_updates,
 )
-from ..core.core import (
+from ..core.cluster import (
     set_jobset_on_cluster,
 )
 from ..core.kjob import (

src/xpk/commands/run.py

@@ -16,12 +16,12 @@
 
 from argparse import Namespace
 
+from ..core.commands import run_command_with_full_controls
+from ..core.gcloud_context import add_zone_and_project
+from ..core.kjob import AppProfileDefaults
 from ..core.kueue import LOCAL_QUEUE_NAME
 from ..utils.console import xpk_exit, xpk_print
 from .common import set_cluster_command
-from ..core.core import add_zone_and_project
-from ..core.kjob import AppProfileDefaults
-from ..core.commands import run_command_with_full_controls
 from .kind import set_local_cluster_command
 
 

src/xpk/commands/storage.py

@@ -19,7 +19,7 @@
 from kubernetes import client as k8s_client
 from kubernetes.client.rest import ApiException
 
-from ..core.core import (
+from ..core.cluster import (
     setup_k8s_env,
     update_cluster_with_gcsfuse_driver_if_necessary,
     update_cluster_with_workload_identity_if_necessary,

src/xpk/commands/version.py

@@ -14,7 +14,7 @@
 limitations under the License.
 """
 
-from ..core.core import __version__
+from ..core.config import __version__
 from ..utils.console import xpk_print