|
14 | 14 | limitations under the License.
|
15 | 15 | """
|
16 | 16 |
|
17 |
| -from ..core.commands import run_command_for_value, run_command_with_updates |
18 |
| -from ..core.core import ( |
19 |
| - VERTEX_TENSORBOARD_FEATURE_FLAG, |
20 |
| - add_zone_and_project, |
21 |
| - create_cluster_configmaps, |
22 |
| - create_cluster_network_config, |
23 |
| - create_vertex_tensorboard, |
24 |
| - delete_cluster_subnets, |
| 17 | +from tabulate import tabulate |
| 18 | + |
| 19 | +from ..core.capacity import H100_DEVICE_TYPE |
| 20 | +from ..core.cluster import ( |
25 | 21 | get_all_clusters_programmatic,
|
26 | 22 | get_cluster_credentials,
|
27 |
| - get_gke_control_plane_version, |
28 |
| - get_gke_node_pool_version, |
29 |
| - get_gke_server_config, |
30 |
| - h100_device_type, |
31 | 23 | install_nccl_on_cluster,
|
32 |
| - run_gke_node_pool_create_command, |
33 | 24 | set_jobset_on_cluster,
|
34 |
| - set_up_cluster_network_for_gpu, |
35 | 25 | setup_k8s_env,
|
36 | 26 | update_cluster_with_gcsfuse_driver_if_necessary,
|
37 | 27 | update_cluster_with_workload_identity_if_necessary,
|
38 |
| - zone_to_region, |
39 |
| - get_user_input, |
40 | 28 | )
|
41 | 29 | from ..core.cluster_private import authorize_private_cluster_access_if_necessary
|
42 |
| -from ..core.kjob import ( |
43 |
| - verify_kjob_installed, |
44 |
| - prepare_kjob, |
45 |
| - apply_kjob_crds, |
| 30 | +from ..core.commands import run_command_for_value, run_command_with_updates |
| 31 | +from ..core.config import VERTEX_TENSORBOARD_FEATURE_FLAG |
| 32 | +from ..core.gcloud_context import ( |
| 33 | + add_zone_and_project, |
| 34 | + get_gke_control_plane_version, |
| 35 | + get_gke_server_config, |
| 36 | + zone_to_region, |
46 | 37 | )
|
| 38 | +from ..core.kjob import apply_kjob_crds, prepare_kjob, verify_kjob_installed |
47 | 39 | from ..core.kueue import (
|
48 | 40 | cluster_preheat_yml,
|
49 | 41 | install_kueue_crs,
|
50 | 42 | install_kueue_on_cluster,
|
51 | 43 | wait_for_kueue_available,
|
52 | 44 | )
|
53 | 45 | from ..core.nap import enable_autoprovisioning_on_cluster
|
| 46 | +from ..core.network import ( |
| 47 | + create_cluster_network_config, |
| 48 | + delete_cluster_subnets, |
| 49 | + set_up_cluster_network_for_gpu, |
| 50 | +) |
| 51 | +from ..core.nodepool import get_gke_node_pool_version, run_gke_node_pool_create_command |
54 | 52 | from ..core.ray import install_ray_cluster
|
| 53 | +from ..core.resources import create_cluster_configmaps |
55 | 54 | from ..core.storage import install_storage_crd
|
56 | 55 | from ..core.system_characteristics import (
|
57 | 56 | AcceleratorType,
|
58 | 57 | AcceleratorTypeToAcceleratorCharacteristics,
|
59 | 58 | SystemCharacteristics,
|
60 | 59 | get_system_characteristics,
|
61 | 60 | )
|
| 61 | +from ..core.vertex import create_vertex_tensorboard |
62 | 62 | from ..core.workload import get_workload_list
|
| 63 | +from ..utils.console import get_user_input, xpk_exit, xpk_print |
63 | 64 | from ..utils.file import write_tmp_file
|
64 |
| -from ..utils.console import xpk_exit, xpk_print |
65 | 65 | from . import cluster_gcluster
|
66 | 66 |
|
67 |
| -from tabulate import tabulate |
68 |
| - |
69 | 67 |
|
70 | 68 | def cluster_create(args) -> None:
|
71 | 69 | """Function around cluster creation.
|
@@ -148,7 +146,7 @@ def cluster_create(args) -> None:
|
148 | 146 | if set_up_cluster_network_code != 0:
|
149 | 147 | xpk_exit(set_up_cluster_network_code)
|
150 | 148 |
|
151 |
| - if system.device_type == h100_device_type: |
| 149 | + if system.device_type == H100_DEVICE_TYPE: |
152 | 150 | xpk_print('Creating Network Config for cluster')
|
153 | 151 | create_cluster_network_config_code = create_cluster_network_config(args)
|
154 | 152 | if create_cluster_network_config_code != 0:
|
|
0 commit comments