skypilot-nightly 1.0.0.dev20250107__py3-none-any.whl → 1.0.0.dev20250109__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- sky/__init__.py +2 -2
- sky/authentication.py +20 -8
- sky/clouds/aws.py +11 -1
- sky/provision/aws/config.py +11 -3
- sky/provision/docker_utils.py +8 -3
- sky/provision/runpod/instance.py +10 -1
- sky/provision/runpod/utils.py +170 -13
- sky/setup_files/dependencies.py +3 -1
- sky/skylet/providers/command_runner.py +5 -7
- sky/templates/runpod-ray.yml.j2 +13 -0
- {skypilot_nightly-1.0.0.dev20250107.dist-info → skypilot_nightly-1.0.0.dev20250109.dist-info}/METADATA +13 -4
- {skypilot_nightly-1.0.0.dev20250107.dist-info → skypilot_nightly-1.0.0.dev20250109.dist-info}/RECORD +16 -16
- {skypilot_nightly-1.0.0.dev20250107.dist-info → skypilot_nightly-1.0.0.dev20250109.dist-info}/WHEEL +1 -1
- {skypilot_nightly-1.0.0.dev20250107.dist-info → skypilot_nightly-1.0.0.dev20250109.dist-info}/LICENSE +0 -0
- {skypilot_nightly-1.0.0.dev20250107.dist-info → skypilot_nightly-1.0.0.dev20250109.dist-info}/entry_points.txt +0 -0
- {skypilot_nightly-1.0.0.dev20250107.dist-info → skypilot_nightly-1.0.0.dev20250109.dist-info}/top_level.txt +0 -0
sky/__init__.py
CHANGED
@@ -5,7 +5,7 @@ from typing import Optional
|
|
5
5
|
import urllib.request
|
6
6
|
|
7
7
|
# Replaced with the current commit when building the wheels.
|
8
|
-
_SKYPILOT_COMMIT_SHA = '
|
8
|
+
_SKYPILOT_COMMIT_SHA = '544c34e2e7eb995530c05fe10148aad87a4d0501'
|
9
9
|
|
10
10
|
|
11
11
|
def _get_git_commit():
|
@@ -35,7 +35,7 @@ def _get_git_commit():
|
|
35
35
|
|
36
36
|
|
37
37
|
__commit__ = _get_git_commit()
|
38
|
-
__version__ = '1.0.0.
|
38
|
+
__version__ = '1.0.0.dev20250109'
|
39
39
|
__root_dir__ = os.path.dirname(os.path.abspath(__file__))
|
40
40
|
|
41
41
|
|
sky/authentication.py
CHANGED
@@ -408,14 +408,26 @@ def setup_kubernetes_authentication(config: Dict[str, Any]) -> Dict[str, Any]:
|
|
408
408
|
secret = k8s.client.V1Secret(
|
409
409
|
metadata=k8s.client.V1ObjectMeta(**secret_metadata),
|
410
410
|
string_data={secret_field_name: public_key})
|
411
|
-
|
412
|
-
|
413
|
-
|
414
|
-
secret_name
|
415
|
-
|
416
|
-
|
417
|
-
|
418
|
-
|
411
|
+
try:
|
412
|
+
if kubernetes_utils.check_secret_exists(secret_name, namespace,
|
413
|
+
context):
|
414
|
+
logger.debug(f'Key {secret_name} exists in the cluster, '
|
415
|
+
'patching it...')
|
416
|
+
kubernetes.core_api(context).patch_namespaced_secret(
|
417
|
+
secret_name, namespace, secret)
|
418
|
+
else:
|
419
|
+
logger.debug(f'Key {secret_name} does not exist in the cluster, '
|
420
|
+
'creating it...')
|
421
|
+
kubernetes.core_api(context).create_namespaced_secret(
|
422
|
+
namespace, secret)
|
423
|
+
except kubernetes.api_exception() as e:
|
424
|
+
if e.status == 409 and e.reason == 'AlreadyExists':
|
425
|
+
logger.debug(f'Key {secret_name} was created concurrently, '
|
426
|
+
'patching it...')
|
427
|
+
kubernetes.core_api(context).patch_namespaced_secret(
|
428
|
+
secret_name, namespace, secret)
|
429
|
+
else:
|
430
|
+
raise e
|
419
431
|
|
420
432
|
private_key_path, _ = get_or_generate_keys()
|
421
433
|
if network_mode == nodeport_mode:
|
sky/clouds/aws.py
CHANGED
@@ -95,6 +95,8 @@ class AWSIdentityType(enum.Enum):
|
|
95
95
|
|
96
96
|
CONTAINER_ROLE = 'container-role'
|
97
97
|
|
98
|
+
CUSTOM_PROCESS = 'custom-process'
|
99
|
+
|
98
100
|
# Name Value Type Location
|
99
101
|
# ---- ----- ---- --------
|
100
102
|
# profile <not set> None None
|
@@ -614,10 +616,16 @@ class AWS(clouds.Cloud):
|
|
614
616
|
hints = f'AWS IAM role is set.{single_cloud_hint}'
|
615
617
|
elif identity_type == AWSIdentityType.CONTAINER_ROLE:
|
616
618
|
# Similar to the IAM ROLE, an ECS container may not store credentials
|
617
|
-
# in the~/.aws/credentials file. So we don't check for the existence of
|
619
|
+
# in the ~/.aws/credentials file. So we don't check for the existence of
|
618
620
|
# the file. i.e. the container will be assigned the IAM role of the
|
619
621
|
# task: skypilot-v1.
|
620
622
|
hints = f'AWS container-role is set.{single_cloud_hint}'
|
623
|
+
elif identity_type == AWSIdentityType.CUSTOM_PROCESS:
|
624
|
+
# Similar to the IAM ROLE, a custom process may not store credentials
|
625
|
+
# in the ~/.aws/credentials file. So we don't check for the existence of
|
626
|
+
# the file. i.e. the custom process will be assigned the IAM role of the
|
627
|
+
# task: skypilot-v1.
|
628
|
+
hints = f'AWS custom-process is set.{single_cloud_hint}'
|
621
629
|
else:
|
622
630
|
# This file is required because it is required by the VMs launched on
|
623
631
|
# other clouds to access private s3 buckets and resources like EC2.
|
@@ -677,6 +685,8 @@ class AWS(clouds.Cloud):
|
|
677
685
|
return AWSIdentityType.CONTAINER_ROLE
|
678
686
|
elif _is_access_key_of_type(AWSIdentityType.ENV.value):
|
679
687
|
return AWSIdentityType.ENV
|
688
|
+
elif _is_access_key_of_type(AWSIdentityType.CUSTOM_PROCESS.value):
|
689
|
+
return AWSIdentityType.CUSTOM_PROCESS
|
680
690
|
else:
|
681
691
|
return AWSIdentityType.SHARED_CREDENTIALS_FILE
|
682
692
|
|
sky/provision/aws/config.py
CHANGED
@@ -383,10 +383,13 @@ def _usable_subnets(
|
|
383
383
|
raise exc
|
384
384
|
|
385
385
|
if not subnets:
|
386
|
+
vpc_msg = (f'Does a default VPC exist in region '
|
387
|
+
f'{ec2.meta.client.meta.region_name}? ') if (
|
388
|
+
vpc_id_of_sg is None) else ''
|
386
389
|
_skypilot_log_error_and_exit_for_failover(
|
387
|
-
'No usable subnets found
|
388
|
-
'manually creating an instance in your specified region to '
|
389
|
-
'populate the list of subnets and
|
390
|
+
f'No usable subnets found. {vpc_msg}'
|
391
|
+
'Try manually creating an instance in your specified region to '
|
392
|
+
'populate the list of subnets and try again. '
|
390
393
|
'Note that the subnet must map public IPs '
|
391
394
|
'on instance launch unless you set `use_internal_ips: true` in '
|
392
395
|
'the `provider` config.')
|
@@ -495,6 +498,11 @@ def _get_subnet_and_vpc_id(ec2, security_group_ids: Optional[List[str]],
|
|
495
498
|
vpc_id_of_sg = None
|
496
499
|
|
497
500
|
all_subnets = list(ec2.subnets.all())
|
501
|
+
# If no VPC is specified, use the default VPC.
|
502
|
+
# We filter only for default VPCs to avoid using subnets that users may
|
503
|
+
# not want SkyPilot to use.
|
504
|
+
if vpc_id_of_sg is None:
|
505
|
+
all_subnets = [s for s in all_subnets if s.vpc.is_default]
|
498
506
|
subnets, vpc_id = _usable_subnets(
|
499
507
|
ec2,
|
500
508
|
user_specified_subnets=None,
|
sky/provision/docker_utils.py
CHANGED
@@ -38,6 +38,13 @@ class DockerLoginConfig:
|
|
38
38
|
password: str
|
39
39
|
server: str
|
40
40
|
|
41
|
+
def format_image(self, image: str) -> str:
|
42
|
+
"""Format the image name with the server prefix."""
|
43
|
+
server_prefix = f'{self.server}/'
|
44
|
+
if not image.startswith(server_prefix):
|
45
|
+
return f'{server_prefix}{image}'
|
46
|
+
return image
|
47
|
+
|
41
48
|
@classmethod
|
42
49
|
def from_env_vars(cls, d: Dict[str, str]) -> 'DockerLoginConfig':
|
43
50
|
return cls(
|
@@ -220,9 +227,7 @@ class DockerInitializer:
|
|
220
227
|
wait_for_docker_daemon=True)
|
221
228
|
# We automatically add the server prefix to the image name if
|
222
229
|
# the user did not add it.
|
223
|
-
|
224
|
-
if not specific_image.startswith(server_prefix):
|
225
|
-
specific_image = f'{server_prefix}{specific_image}'
|
230
|
+
specific_image = docker_login_config.format_image(specific_image)
|
226
231
|
|
227
232
|
if self.docker_config.get('pull_before_run', True):
|
228
233
|
assert specific_image, ('Image must be included in config if ' +
|
sky/provision/runpod/instance.py
CHANGED
@@ -83,7 +83,8 @@ def run_instances(region: str, cluster_name_on_cloud: str,
|
|
83
83
|
node_type = 'head' if head_instance_id is None else 'worker'
|
84
84
|
try:
|
85
85
|
instance_id = utils.launch(
|
86
|
-
|
86
|
+
cluster_name=cluster_name_on_cloud,
|
87
|
+
node_type=node_type,
|
87
88
|
instance_type=config.node_config['InstanceType'],
|
88
89
|
region=region,
|
89
90
|
disk_size=config.node_config['DiskSize'],
|
@@ -92,6 +93,8 @@ def run_instances(region: str, cluster_name_on_cloud: str,
|
|
92
93
|
public_key=config.node_config['PublicKey'],
|
93
94
|
preemptible=config.node_config['Preemptible'],
|
94
95
|
bid_per_gpu=config.node_config['BidPerGPU'],
|
96
|
+
docker_login_config=config.provider_config.get(
|
97
|
+
'docker_login_config'),
|
95
98
|
)
|
96
99
|
except Exception as e: # pylint: disable=broad-except
|
97
100
|
logger.warning(f'run_instances error: {e}')
|
@@ -145,6 +148,8 @@ def terminate_instances(
|
|
145
148
|
"""See sky/provision/__init__.py"""
|
146
149
|
del provider_config # unused
|
147
150
|
instances = _filter_instances(cluster_name_on_cloud, None)
|
151
|
+
template_name, registry_auth_id = utils.get_registry_auth_resources(
|
152
|
+
cluster_name_on_cloud)
|
148
153
|
for inst_id, inst in instances.items():
|
149
154
|
logger.debug(f'Terminating instance {inst_id}: {inst}')
|
150
155
|
if worker_only and inst['name'].endswith('-head'):
|
@@ -157,6 +162,10 @@ def terminate_instances(
|
|
157
162
|
f'Failed to terminate instance {inst_id}: '
|
158
163
|
f'{common_utils.format_exception(e, use_bracket=False)}'
|
159
164
|
) from e
|
165
|
+
if template_name is not None:
|
166
|
+
utils.delete_pod_template(template_name)
|
167
|
+
if registry_auth_id is not None:
|
168
|
+
utils.delete_register_auth(registry_auth_id)
|
160
169
|
|
161
170
|
|
162
171
|
def get_cluster_info(
|
sky/provision/runpod/utils.py
CHANGED
@@ -2,10 +2,11 @@
|
|
2
2
|
|
3
3
|
import base64
|
4
4
|
import time
|
5
|
-
from typing import Any, Dict, List, Optional
|
5
|
+
from typing import Any, Dict, List, Optional, Tuple
|
6
6
|
|
7
7
|
from sky import sky_logging
|
8
8
|
from sky.adaptors import runpod
|
9
|
+
from sky.provision import docker_utils
|
9
10
|
import sky.provision.runpod.api.commands as runpod_commands
|
10
11
|
from sky.skylet import constants
|
11
12
|
from sky.utils import common_utils
|
@@ -47,6 +48,11 @@ GPU_NAME_MAP = {
|
|
47
48
|
}
|
48
49
|
|
49
50
|
|
51
|
+
def _construct_docker_login_template_name(cluster_name: str) -> str:
|
52
|
+
"""Constructs the registry auth template name."""
|
53
|
+
return f'{cluster_name}-docker-login-template'
|
54
|
+
|
55
|
+
|
50
56
|
def retry(func):
|
51
57
|
"""Decorator to retry a function."""
|
52
58
|
|
@@ -66,9 +72,83 @@ def retry(func):
|
|
66
72
|
return wrapper
|
67
73
|
|
68
74
|
|
75
|
+
# Adapted from runpod.api.queries.pods.py::QUERY_POD.
|
76
|
+
# Adding containerRegistryAuthId to the query.
|
77
|
+
_QUERY_POD = """
|
78
|
+
query myPods {
|
79
|
+
myself {
|
80
|
+
pods {
|
81
|
+
id
|
82
|
+
containerDiskInGb
|
83
|
+
containerRegistryAuthId
|
84
|
+
costPerHr
|
85
|
+
desiredStatus
|
86
|
+
dockerArgs
|
87
|
+
dockerId
|
88
|
+
env
|
89
|
+
gpuCount
|
90
|
+
imageName
|
91
|
+
lastStatusChange
|
92
|
+
machineId
|
93
|
+
memoryInGb
|
94
|
+
name
|
95
|
+
podType
|
96
|
+
port
|
97
|
+
ports
|
98
|
+
uptimeSeconds
|
99
|
+
vcpuCount
|
100
|
+
volumeInGb
|
101
|
+
volumeMountPath
|
102
|
+
runtime {
|
103
|
+
ports{
|
104
|
+
ip
|
105
|
+
isIpPublic
|
106
|
+
privatePort
|
107
|
+
publicPort
|
108
|
+
type
|
109
|
+
}
|
110
|
+
}
|
111
|
+
machine {
|
112
|
+
gpuDisplayName
|
113
|
+
}
|
114
|
+
}
|
115
|
+
}
|
116
|
+
}
|
117
|
+
"""
|
118
|
+
|
119
|
+
|
120
|
+
def _sky_get_pods() -> dict:
|
121
|
+
"""List all pods with extra registry auth information.
|
122
|
+
|
123
|
+
Adapted from runpod.get_pods() to include containerRegistryAuthId.
|
124
|
+
"""
|
125
|
+
raw_return = runpod.runpod.api.graphql.run_graphql_query(_QUERY_POD)
|
126
|
+
cleaned_return = raw_return['data']['myself']['pods']
|
127
|
+
return cleaned_return
|
128
|
+
|
129
|
+
|
130
|
+
_QUERY_POD_TEMPLATE_WITH_REGISTRY_AUTH = """
|
131
|
+
query myself {
|
132
|
+
myself {
|
133
|
+
podTemplates {
|
134
|
+
name
|
135
|
+
containerRegistryAuthId
|
136
|
+
}
|
137
|
+
}
|
138
|
+
}
|
139
|
+
"""
|
140
|
+
|
141
|
+
|
142
|
+
def _list_pod_templates_with_container_registry() -> dict:
|
143
|
+
"""List all pod templates."""
|
144
|
+
raw_return = runpod.runpod.api.graphql.run_graphql_query(
|
145
|
+
_QUERY_POD_TEMPLATE_WITH_REGISTRY_AUTH)
|
146
|
+
return raw_return['data']['myself']['podTemplates']
|
147
|
+
|
148
|
+
|
69
149
|
def list_instances() -> Dict[str, Dict[str, Any]]:
|
70
150
|
"""Lists instances associated with API key."""
|
71
|
-
instances =
|
151
|
+
instances = _sky_get_pods()
|
72
152
|
|
73
153
|
instance_dict: Dict[str, Dict[str, Any]] = {}
|
74
154
|
for instance in instances:
|
@@ -100,14 +180,75 @@ def list_instances() -> Dict[str, Dict[str, Any]]:
|
|
100
180
|
return instance_dict
|
101
181
|
|
102
182
|
|
103
|
-
def
|
104
|
-
|
105
|
-
|
183
|
+
def delete_pod_template(template_name: str) -> None:
|
184
|
+
"""Deletes a pod template."""
|
185
|
+
try:
|
186
|
+
runpod.runpod.api.graphql.run_graphql_query(
|
187
|
+
f'mutation {{deleteTemplate(templateName: "{template_name}")}}')
|
188
|
+
except runpod.runpod.error.QueryError as e:
|
189
|
+
logger.warning(f'Failed to delete template {template_name}: {e}'
|
190
|
+
'Please delete it manually.')
|
191
|
+
|
192
|
+
|
193
|
+
def delete_register_auth(registry_auth_id: str) -> None:
|
194
|
+
"""Deletes a registry auth."""
|
195
|
+
try:
|
196
|
+
runpod.runpod.delete_container_registry_auth(registry_auth_id)
|
197
|
+
except runpod.runpod.error.QueryError as e:
|
198
|
+
logger.warning(f'Failed to delete registry auth {registry_auth_id}: {e}'
|
199
|
+
'Please delete it manually.')
|
200
|
+
|
201
|
+
|
202
|
+
def _create_template_for_docker_login(
|
203
|
+
cluster_name: str,
|
204
|
+
image_name: str,
|
205
|
+
docker_login_config: Optional[Dict[str, str]],
|
206
|
+
) -> Tuple[str, Optional[str]]:
|
207
|
+
"""Creates a template for the given image with the docker login config.
|
208
|
+
|
209
|
+
Returns:
|
210
|
+
formatted_image_name: The formatted image name.
|
211
|
+
template_id: The template ID. None for no docker login config.
|
212
|
+
"""
|
213
|
+
if docker_login_config is None:
|
214
|
+
return image_name, None
|
215
|
+
login_config = docker_utils.DockerLoginConfig(**docker_login_config)
|
216
|
+
container_registry_auth_name = f'{cluster_name}-registry-auth'
|
217
|
+
container_template_name = _construct_docker_login_template_name(
|
218
|
+
cluster_name)
|
219
|
+
# The `name` argument is only for display purpose and the registry server
|
220
|
+
# will be splitted from the docker image name (Tested with AWS ECR).
|
221
|
+
# Here we only need the username and password to create the registry auth.
|
222
|
+
# TODO(tian): Now we create a template and a registry auth for each cluster.
|
223
|
+
# Consider create one for each server and reuse them. Challenges including
|
224
|
+
# calculate the reference count and delete them when no longer needed.
|
225
|
+
create_auth_resp = runpod.runpod.create_container_registry_auth(
|
226
|
+
name=container_registry_auth_name,
|
227
|
+
username=login_config.username,
|
228
|
+
password=login_config.password,
|
229
|
+
)
|
230
|
+
registry_auth_id = create_auth_resp['id']
|
231
|
+
create_template_resp = runpod.runpod.create_template(
|
232
|
+
name=container_template_name,
|
233
|
+
image_name=None,
|
234
|
+
registry_auth_id=registry_auth_id,
|
235
|
+
)
|
236
|
+
return login_config.format_image(image_name), create_template_resp['id']
|
237
|
+
|
238
|
+
|
239
|
+
def launch(cluster_name: str, node_type: str, instance_type: str, region: str,
|
240
|
+
disk_size: int, image_name: str, ports: Optional[List[int]],
|
241
|
+
public_key: str, preemptible: Optional[bool], bid_per_gpu: float,
|
242
|
+
docker_login_config: Optional[Dict[str, str]]) -> str:
|
106
243
|
"""Launches an instance with the given parameters.
|
107
244
|
|
108
245
|
Converts the instance_type to the RunPod GPU name, finds the specs for the
|
109
246
|
GPU, and launches the instance.
|
247
|
+
|
248
|
+
Returns:
|
249
|
+
instance_id: The instance ID.
|
110
250
|
"""
|
251
|
+
name = f'{cluster_name}-{node_type}'
|
111
252
|
gpu_type = GPU_NAME_MAP[instance_type.split('_')[1]]
|
112
253
|
gpu_quantity = int(instance_type.split('_')[0].replace('x', ''))
|
113
254
|
cloud_type = instance_type.split('_')[2]
|
@@ -139,21 +280,24 @@ def launch(name: str, instance_type: str, region: str, disk_size: int,
|
|
139
280
|
# Use base64 to deal with the tricky quoting issues caused by runpod API.
|
140
281
|
encoded = base64.b64encode(setup_cmd.encode('utf-8')).decode('utf-8')
|
141
282
|
|
283
|
+
docker_args = (f'bash -c \'echo {encoded} | base64 --decode > init.sh; '
|
284
|
+
f'bash init.sh\'')
|
285
|
+
|
142
286
|
# Port 8081 is occupied for nginx in the base image.
|
143
287
|
custom_ports_str = ''
|
144
288
|
if ports is not None:
|
145
289
|
custom_ports_str = ''.join([f'{p}/tcp,' for p in ports])
|
290
|
+
ports_str = (f'22/tcp,'
|
291
|
+
f'{custom_ports_str}'
|
292
|
+
f'{constants.SKY_REMOTE_RAY_DASHBOARD_PORT}/http,'
|
293
|
+
f'{constants.SKY_REMOTE_RAY_PORT}/http')
|
146
294
|
|
147
|
-
|
148
|
-
|
149
|
-
ports = (f'22/tcp,'
|
150
|
-
f'{custom_ports_str}'
|
151
|
-
f'{constants.SKY_REMOTE_RAY_DASHBOARD_PORT}/http,'
|
152
|
-
f'{constants.SKY_REMOTE_RAY_PORT}/http')
|
295
|
+
image_name_formatted, template_id = _create_template_for_docker_login(
|
296
|
+
cluster_name, image_name, docker_login_config)
|
153
297
|
|
154
298
|
params = {
|
155
299
|
'name': name,
|
156
|
-
'image_name':
|
300
|
+
'image_name': image_name_formatted,
|
157
301
|
'gpu_type_id': gpu_type,
|
158
302
|
'cloud_type': cloud_type,
|
159
303
|
'container_disk_in_gb': disk_size,
|
@@ -161,9 +305,10 @@ def launch(name: str, instance_type: str, region: str, disk_size: int,
|
|
161
305
|
'min_memory_in_gb': gpu_specs['memoryInGb'] * gpu_quantity,
|
162
306
|
'gpu_count': gpu_quantity,
|
163
307
|
'country_code': region,
|
164
|
-
'ports':
|
308
|
+
'ports': ports_str,
|
165
309
|
'support_public_ip': True,
|
166
310
|
'docker_args': docker_args,
|
311
|
+
'template_id': template_id,
|
167
312
|
}
|
168
313
|
|
169
314
|
if preemptible is None or not preemptible:
|
@@ -177,6 +322,18 @@ def launch(name: str, instance_type: str, region: str, disk_size: int,
|
|
177
322
|
return new_instance['id']
|
178
323
|
|
179
324
|
|
325
|
+
def get_registry_auth_resources(
|
326
|
+
cluster_name: str) -> Tuple[Optional[str], Optional[str]]:
|
327
|
+
"""Gets the registry auth resources."""
|
328
|
+
container_registry_auth_name = _construct_docker_login_template_name(
|
329
|
+
cluster_name)
|
330
|
+
for template in _list_pod_templates_with_container_registry():
|
331
|
+
if template['name'] == container_registry_auth_name:
|
332
|
+
return container_registry_auth_name, template[
|
333
|
+
'containerRegistryAuthId']
|
334
|
+
return None, None
|
335
|
+
|
336
|
+
|
180
337
|
def remove(instance_id: str) -> None:
|
181
338
|
"""Terminates the given instance."""
|
182
339
|
runpod.runpod.terminate_pod(instance_id)
|
sky/setup_files/dependencies.py
CHANGED
@@ -123,7 +123,9 @@ extras_require: Dict[str, List[str]] = {
|
|
123
123
|
'oci': ['oci'] + local_ray,
|
124
124
|
'kubernetes': ['kubernetes>=20.0.0'],
|
125
125
|
'remote': remote,
|
126
|
-
|
126
|
+
# For the container registry auth api. Reference:
|
127
|
+
# https://github.com/runpod/runpod-python/releases/tag/1.6.1
|
128
|
+
'runpod': ['runpod>=1.6.1'],
|
127
129
|
'fluidstack': [], # No dependencies needed for fluidstack
|
128
130
|
'cudo': ['cudo-compute>=0.1.10'],
|
129
131
|
'paperspace': [], # No dependencies needed for paperspace
|
@@ -25,7 +25,7 @@ def docker_start_cmds(
|
|
25
25
|
docker_cmd,
|
26
26
|
):
|
27
27
|
"""Generating docker start command without --rm.
|
28
|
-
|
28
|
+
|
29
29
|
The code is borrowed from `ray.autoscaler._private.docker`.
|
30
30
|
|
31
31
|
Changes we made:
|
@@ -159,19 +159,17 @@ class SkyDockerCommandRunner(DockerCommandRunner):
|
|
159
159
|
return True
|
160
160
|
|
161
161
|
# SkyPilot: Docker login if user specified a private docker registry.
|
162
|
-
if
|
162
|
+
if 'docker_login_config' in self.docker_config:
|
163
163
|
# TODO(tian): Maybe support a command to get the login password?
|
164
|
-
docker_login_config: docker_utils.DockerLoginConfig =
|
165
|
-
|
164
|
+
docker_login_config: docker_utils.DockerLoginConfig = (
|
165
|
+
self.docker_config['docker_login_config'])
|
166
166
|
self._run_with_retry(
|
167
167
|
f'{self.docker_cmd} login --username '
|
168
168
|
f'{docker_login_config.username} --password '
|
169
169
|
f'{docker_login_config.password} {docker_login_config.server}')
|
170
170
|
# We automatically add the server prefix to the image name if
|
171
171
|
# the user did not add it.
|
172
|
-
|
173
|
-
if not specific_image.startswith(server_prefix):
|
174
|
-
specific_image = f'{server_prefix}{specific_image}'
|
172
|
+
specific_image = docker_login_config.format_image(specific_image)
|
175
173
|
|
176
174
|
if self.docker_config.get('pull_before_run', True):
|
177
175
|
assert specific_image, ('Image must be included in config if '
|
sky/templates/runpod-ray.yml.j2
CHANGED
@@ -10,6 +10,19 @@ provider:
|
|
10
10
|
module: sky.provision.runpod
|
11
11
|
region: "{{region}}"
|
12
12
|
disable_launch_config_check: true
|
13
|
+
# For RunPod, we directly set the image id for the docker as runtime environment
|
14
|
+
# support, thus we need to avoid the DockerInitializer detects the docker field
|
15
|
+
# and performs the initialization. Therefore we put the docker login config in
|
16
|
+
# the provider config here.
|
17
|
+
{%- if docker_login_config is not none %}
|
18
|
+
docker_login_config:
|
19
|
+
username: |-
|
20
|
+
{{docker_login_config.username}}
|
21
|
+
password: |-
|
22
|
+
{{docker_login_config.password}}
|
23
|
+
server: |-
|
24
|
+
{{docker_login_config.server}}
|
25
|
+
{%- endif %}
|
13
26
|
|
14
27
|
auth:
|
15
28
|
ssh_user: root
|
@@ -1,6 +1,6 @@
|
|
1
|
-
Metadata-Version: 2.
|
1
|
+
Metadata-Version: 2.2
|
2
2
|
Name: skypilot-nightly
|
3
|
-
Version: 1.0.0.
|
3
|
+
Version: 1.0.0.dev20250109
|
4
4
|
Summary: SkyPilot: An intercloud broker for the clouds
|
5
5
|
Author: SkyPilot Team
|
6
6
|
License: Apache 2.0
|
@@ -90,7 +90,7 @@ Requires-Dist: grpcio!=1.48.0,<=1.51.3,>=1.42.0; (python_version >= "3.10" and s
|
|
90
90
|
Requires-Dist: protobuf!=3.19.5,>=3.15.3; extra == "remote"
|
91
91
|
Requires-Dist: pydantic!=2.0.*,!=2.1.*,!=2.2.*,!=2.3.*,!=2.4.*,<3; extra == "remote"
|
92
92
|
Provides-Extra: runpod
|
93
|
-
Requires-Dist: runpod>=1.
|
93
|
+
Requires-Dist: runpod>=1.6.1; extra == "runpod"
|
94
94
|
Provides-Extra: fluidstack
|
95
95
|
Provides-Extra: cudo
|
96
96
|
Requires-Dist: cudo-compute>=0.1.10; extra == "cudo"
|
@@ -140,12 +140,21 @@ Requires-Dist: grpcio!=1.48.0,<=1.51.3,>=1.32.0; (python_version < "3.10" and sy
|
|
140
140
|
Requires-Dist: grpcio!=1.48.0,<=1.51.3,>=1.42.0; (python_version >= "3.10" and sys_platform != "darwin") and extra == "all"
|
141
141
|
Requires-Dist: protobuf!=3.19.5,>=3.15.3; extra == "all"
|
142
142
|
Requires-Dist: pydantic!=2.0.*,!=2.1.*,!=2.2.*,!=2.3.*,!=2.4.*,<3; extra == "all"
|
143
|
-
Requires-Dist: runpod>=1.
|
143
|
+
Requires-Dist: runpod>=1.6.1; extra == "all"
|
144
144
|
Requires-Dist: cudo-compute>=0.1.10; extra == "all"
|
145
145
|
Requires-Dist: pydo>=0.3.0; extra == "all"
|
146
146
|
Requires-Dist: azure-core>=1.24.0; extra == "all"
|
147
147
|
Requires-Dist: azure-common; extra == "all"
|
148
148
|
Requires-Dist: pyvmomi==8.0.1.0.2; extra == "all"
|
149
|
+
Dynamic: author
|
150
|
+
Dynamic: classifier
|
151
|
+
Dynamic: description
|
152
|
+
Dynamic: description-content-type
|
153
|
+
Dynamic: license
|
154
|
+
Dynamic: project-url
|
155
|
+
Dynamic: provides-extra
|
156
|
+
Dynamic: requires-dist
|
157
|
+
Dynamic: summary
|
149
158
|
|
150
159
|
<p align="center">
|
151
160
|
<img alt="SkyPilot" src="https://raw.githubusercontent.com/skypilot-org/skypilot/master/docs/source/images/skypilot-wide-light-1k.png" width=55%>
|
{skypilot_nightly-1.0.0.dev20250107.dist-info → skypilot_nightly-1.0.0.dev20250109.dist-info}/RECORD
RENAMED
@@ -1,6 +1,6 @@
|
|
1
|
-
sky/__init__.py,sha256=
|
1
|
+
sky/__init__.py,sha256=hTD_EweXO9PykuI558pLbxRiUikldWFXfOtq9aQeZ7c,5944
|
2
2
|
sky/admin_policy.py,sha256=hPo02f_A32gCqhUueF0QYy1fMSSKqRwYEg_9FxScN_s,3248
|
3
|
-
sky/authentication.py,sha256=
|
3
|
+
sky/authentication.py,sha256=LXUDABKP1FJCS256xTTDJa40WXwHKF5x49S-4hZbD1M,21501
|
4
4
|
sky/check.py,sha256=s8deMVL-k9y8gd519K7NWZc3DqWsEySwiAr0uH3Vvcc,9459
|
5
5
|
sky/cli.py,sha256=KbvwSMgQJyW-XuX-N_-JFvBRVAEvy295j4UcxypwbyU,214398
|
6
6
|
sky/cloud_stores.py,sha256=PcLT57_8SZy7o6paAluElfBynaLkbaOq3l-8dNg1AVM,23672
|
@@ -41,7 +41,7 @@ sky/benchmark/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
41
41
|
sky/benchmark/benchmark_state.py,sha256=X8CXmuU9KgsDRhKedhFgjeRMUFWtQsjFs1qECvPG2yg,8723
|
42
42
|
sky/benchmark/benchmark_utils.py,sha256=mP8Ox2WiKfthq6LcUAZnHknFQ0n8v9o_rCh1LXLgkqc,26192
|
43
43
|
sky/clouds/__init__.py,sha256=iORaV6auqMxa6-6FKgt1C9f3UqRk1GASUtakua3tb9A,1395
|
44
|
-
sky/clouds/aws.py,sha256=
|
44
|
+
sky/clouds/aws.py,sha256=bP7ZuHNE47pbCKXob3rxMhby92lbQVrpJk6dAbPIK9k,53474
|
45
45
|
sky/clouds/azure.py,sha256=KtnnNZn4ZEr7xndBHxX91v0YXSI1QWPgIefuM1zDUBA,30784
|
46
46
|
sky/clouds/cloud.py,sha256=5_ZduUcyCEY1JnX_h0PrJ5xwtPP4oor4jf6cICgSArc,35370
|
47
47
|
sky/clouds/cloud_registry.py,sha256=oLoYFjm_SDTgdHokY7b6A5Utq80HXRQNxV0fLjDdVsQ,2361
|
@@ -109,13 +109,13 @@ sky/jobs/dashboard/templates/index.html,sha256=su1tqgcsXNl1lGl9hfIR6ig1f531OO57x
|
|
109
109
|
sky/provision/__init__.py,sha256=rnuwL9x3qQGhX3EYW6KyZWfw_yBqGSiFBDz86T5xus4,6339
|
110
110
|
sky/provision/common.py,sha256=E8AlSUFcn0FYQq1erNmoVfMAdsF9tP2yxfyk-9PLvQU,10286
|
111
111
|
sky/provision/constants.py,sha256=oc_XDUkcoLQ_lwDy5yMeMSWviKS0j0s1c0pjlvpNeWY,800
|
112
|
-
sky/provision/docker_utils.py,sha256=
|
112
|
+
sky/provision/docker_utils.py,sha256=ENm0LkyrYWic3Ikyacho8X5uDMvGsbkZQsb6kNH1DuI,19629
|
113
113
|
sky/provision/instance_setup.py,sha256=8Pudbpke6ah0xufr2UwtsDnNZ64-aAYkz8M44ZA0huI,23218
|
114
114
|
sky/provision/logging.py,sha256=yZWgejrFBhhRjAtvFu5N5bRXIMK5TuwNjp1vKQqz2pw,2103
|
115
115
|
sky/provision/metadata_utils.py,sha256=LrxeV4wD2QPzNdXV_npj8q-pr35FatxBBjF_jSbpOT0,4013
|
116
116
|
sky/provision/provisioner.py,sha256=ZOgFOO0NB4QZVPwd4qikRqi615Bq67n0Vcl3cTDVxNE,29153
|
117
117
|
sky/provision/aws/__init__.py,sha256=mxq8PeWJqUtalDozTNpbtENErRZ1ktEs8uf2aG9UUgU,731
|
118
|
-
sky/provision/aws/config.py,sha256=
|
118
|
+
sky/provision/aws/config.py,sha256=_8jvi8UVMtIVChDDnv5uHV2CoPyKvKqvxJ4xIEBYdDc,24629
|
119
119
|
sky/provision/aws/instance.py,sha256=eCslJ2XfJo_pkQMnKFQqhGnUIRvwKiT12oxBY5-klss,40750
|
120
120
|
sky/provision/aws/utils.py,sha256=m49pS-SHGW7Au3bhDeTPsL8N5iRzbwOXzyEWRCc1Vho,3238
|
121
121
|
sky/provision/azure/__init__.py,sha256=87cgk1_Ws7n9rqaDDPv-HpfrkVeSQMdFQnhnXwyx9g4,548
|
@@ -166,8 +166,8 @@ sky/provision/paperspace/instance.py,sha256=YAIzHEYdO8MRK8iLdweMpY2U28OMCoUvPEwf
|
|
166
166
|
sky/provision/paperspace/utils.py,sha256=uOmxbDKjV6skFizC4gYXSxRuEqso5ck2kF7MbtNmhEs,9580
|
167
167
|
sky/provision/runpod/__init__.py,sha256=6HYvHI27EaLrX1SS0vWVhdLu5HDBeZCdvAeDJuwM5pk,556
|
168
168
|
sky/provision/runpod/config.py,sha256=9ulZJVL7nHuxhTdoj8D7lNn7SdicJ5zc6FIcHIG9tcg,321
|
169
|
-
sky/provision/runpod/instance.py,sha256=
|
170
|
-
sky/provision/runpod/utils.py,sha256=
|
169
|
+
sky/provision/runpod/instance.py,sha256=FNalpTQMvnmACdtFsGvVPVhEkzdqrlmd_pExn33bIQ8,10358
|
170
|
+
sky/provision/runpod/utils.py,sha256=2L4MUrFi4rQOP26IeAkPNQ9oWGPJhH5fvz7NAbL3nNQ,12421
|
171
171
|
sky/provision/runpod/api/__init__.py,sha256=eJwjPeQZ5B7chf4-Bl4YeI2Uo9aLX4M1rr2NmPk89_E,112
|
172
172
|
sky/provision/runpod/api/commands.py,sha256=oh77PS0H0wZudHV8II9ceRuaFQ8FN4NJ4S3-6_PeqPM,4238
|
173
173
|
sky/provision/runpod/api/pods.py,sha256=GMwxgNr9NnHPfyh2Y9b8S_vLhrLY4h7LybFBBQNAyfw,4948
|
@@ -199,7 +199,7 @@ sky/serve/serve_utils.py,sha256=WgPcqEw3WyMOdgRTFg8DSsWyIG1xnRbRkI1-f09tNKg,3974
|
|
199
199
|
sky/serve/service.py,sha256=7bvK9R9D48PZSYcOKSievXQ2mHUMk1d3AAIxtra7WOI,12083
|
200
200
|
sky/serve/service_spec.py,sha256=mph-IRhamiFls4RX_8yhC0lTLzbIxGYbnvosy7VZTi4,14999
|
201
201
|
sky/setup_files/MANIFEST.in,sha256=WF0T89NLichHxZDDSQzvSpiONtAEFyur2MPmGczgTIo,555
|
202
|
-
sky/setup_files/dependencies.py,sha256=
|
202
|
+
sky/setup_files/dependencies.py,sha256=ujGG0rFG8TWGgJv21N-xD0fetYRlaf_xV8ILdsCQDiU,6087
|
203
203
|
sky/setup_files/setup.py,sha256=HMqAIxHrhtQUOlm6_Iz5E_bL4dUvsYgXc9YVQIFayPs,7417
|
204
204
|
sky/skylet/LICENSE,sha256=BnFrJSvUFpMUoH5mOpWnEvaC5R6Uux8W6WXgrte8iYg,12381
|
205
205
|
sky/skylet/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
@@ -214,7 +214,7 @@ sky/skylet/log_lib.pyi,sha256=VpA_VoL970Noj-YrBkKqLxFi34JVMY7KLrOQ3o4AqEI,4336
|
|
214
214
|
sky/skylet/skylet.py,sha256=U9plr5hmhD9-Nyy0LMCymlE8DWtRXTFXQvfbFsS746Y,1153
|
215
215
|
sky/skylet/subprocess_daemon.py,sha256=gcL-_Hea7-SrBUyZfAbo40RBFbaeuBmPCW0dm4YYkPo,3537
|
216
216
|
sky/skylet/providers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
217
|
-
sky/skylet/providers/command_runner.py,sha256=
|
217
|
+
sky/skylet/providers/command_runner.py,sha256=DdBKP0QX325_N3zAVYwnmXmfbfXNqkzWQZpF9DSR7Go,16259
|
218
218
|
sky/skylet/providers/ibm/__init__.py,sha256=GXo5F9ztvs0qMDI_G9wM5KvzySfYugslJMHHAnBENRc,94
|
219
219
|
sky/skylet/providers/ibm/node_provider.py,sha256=PoKp-lyIT2eE6BLqljufhZtGlTrJLe8ebk6mMj2KMdU,38265
|
220
220
|
sky/skylet/providers/ibm/utils.py,sha256=63vhKqLLOhAZdibSp8VWWONeyCER9F6U2VLrSpzlizk,1292
|
@@ -247,7 +247,7 @@ sky/templates/lambda-ray.yml.j2,sha256=HyvO_tX2vxwSsc4IFVSqGuIbjLMk0bevP9bcxb8ZQ
|
|
247
247
|
sky/templates/local-ray.yml.j2,sha256=FNHeyHF6nW9nU9QLIZceUWfvrFTTcO51KqhTnYCEFaA,1185
|
248
248
|
sky/templates/oci-ray.yml.j2,sha256=92dvXGaUd2Kwep9fgTjOsAPJiBLr8GQTjy7pFvuPAyE,4562
|
249
249
|
sky/templates/paperspace-ray.yml.j2,sha256=HQjZNamrB_a4fOMCxQXSVdV5JIHtbGtAE0JzEO8uuVQ,4021
|
250
|
-
sky/templates/runpod-ray.yml.j2,sha256=
|
250
|
+
sky/templates/runpod-ray.yml.j2,sha256=bUiF4Y_EkCA_GKLtTzPXbajdL-NOUiJ38Pe4dZd2dys,4284
|
251
251
|
sky/templates/scp-ray.yml.j2,sha256=I9u8Ax-lit-d6UrCC9BVU8avst8w1cwK6TrzZBcz_JM,5608
|
252
252
|
sky/templates/sky-serve-controller.yaml.j2,sha256=V1IiYhArv_D_7JzC3sVN4nKlSCCCL1AYtIdxFSa-f4c,1628
|
253
253
|
sky/templates/vsphere-ray.yml.j2,sha256=cOQ-qdpxGA2FHajMMhTJI-SmlYzdPterX4Gsiq-nkb0,3587
|
@@ -288,9 +288,9 @@ sky/utils/kubernetes/k8s_gpu_labeler_job.yaml,sha256=k0TBoQ4zgf79-sVkixKSGYFHQ7Z
|
|
288
288
|
sky/utils/kubernetes/k8s_gpu_labeler_setup.yaml,sha256=VLKT2KKimZu1GDg_4AIlIt488oMQvhRZWwsj9vBbPUg,3812
|
289
289
|
sky/utils/kubernetes/rsync_helper.sh,sha256=h4YwrPFf9727CACnMJvF3EyK_0OeOYKKt4su_daKekw,1256
|
290
290
|
sky/utils/kubernetes/ssh_jump_lifecycle_manager.py,sha256=Kq1MDygF2IxFmu9FXpCxqucXLmeUrvs6OtRij6XTQbo,6554
|
291
|
-
skypilot_nightly-1.0.0.
|
292
|
-
skypilot_nightly-1.0.0.
|
293
|
-
skypilot_nightly-1.0.0.
|
294
|
-
skypilot_nightly-1.0.0.
|
295
|
-
skypilot_nightly-1.0.0.
|
296
|
-
skypilot_nightly-1.0.0.
|
291
|
+
skypilot_nightly-1.0.0.dev20250109.dist-info/LICENSE,sha256=emRJAvE7ngL6x0RhQvlns5wJzGI3NEQ_WMjNmd9TZc4,12170
|
292
|
+
skypilot_nightly-1.0.0.dev20250109.dist-info/METADATA,sha256=3esWriJs7mn5M_6alzm4FmYsK8_6E-X60jkRwSdODtY,20632
|
293
|
+
skypilot_nightly-1.0.0.dev20250109.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
|
294
|
+
skypilot_nightly-1.0.0.dev20250109.dist-info/entry_points.txt,sha256=StA6HYpuHj-Y61L2Ze-hK2IcLWgLZcML5gJu8cs6nU4,36
|
295
|
+
skypilot_nightly-1.0.0.dev20250109.dist-info/top_level.txt,sha256=qA8QuiNNb6Y1OF-pCUtPEr6sLEwy2xJX06Bd_CrtrHY,4
|
296
|
+
skypilot_nightly-1.0.0.dev20250109.dist-info/RECORD,,
|
File without changes
|
File without changes
|
File without changes
|