skypilot-nightly 1.0.0.dev20250408__py3-none-any.whl → 1.0.0.dev20250411__py3-none-any.whl

sky/__init__.py

@@ -5,7 +5,7 @@ from typing import Optional
 import urllib.request
 
 # Replaced with the current commit when building the wheels.
-_SKYPILOT_COMMIT_SHA = 'e0674be528e87191ade88961c44c6449d01232fa'
+_SKYPILOT_COMMIT_SHA = 'b4202948016cdf48a5939ea9bf1769a2d31f73bf'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20250408'
+__version__ = '1.0.0.dev20250411'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

sky/adaptors/azure.py

@@ -232,7 +232,7 @@ def get_client(name: str,
                                 'Must provide resource_group_name keyword '
                                 'arguments for container client.')
                             sky_logger.info(
-                                'Failed to check the existance of the '
+                                'Failed to check the existence of the '
                                 f'container {container_url!r} due to '
                                 'insufficient IAM role for storage '
                                 f'account {storage_account_name!r}.')

sky/adaptors/nebius.py

@@ -1,7 +1,6 @@
 """Nebius cloud adaptor."""
 import os
 import threading
-from typing import Optional
 
 from sky.adaptors import common
 from sky.utils import annotations
@@ -168,7 +167,7 @@ def session():
 
 
 @annotations.lru_cache(scope='global')
-def resource(resource_name: str, region: str = DEFAULT_REGION, **kwargs):
+def resource(resource_name: str, **kwargs):
     """Create a Nebius resource.
 
     Args:
@@ -181,21 +180,13 @@ def resource(resource_name: str, region: str = DEFAULT_REGION, **kwargs):
     # Reference: https://stackoverflow.com/a/59635814
 
     session_ = session()
-    nebius_credentials = get_nebius_credentials(session_)
-    endpoint = create_endpoint(region)
 
-    return session_.resource(
-        resource_name,
-        endpoint_url=endpoint,
-        aws_access_key_id=nebius_credentials.access_key,
-        aws_secret_access_key=nebius_credentials.secret_key,
-        region_name=region,
-        **kwargs)
+    return session_.resource(resource_name, **kwargs)
 
 
 @annotations.lru_cache(scope='global')
-def client(service_name: str, region):
-    """Create an Nebius client of a certain service.
+def client(service_name: str):
+    """Create Nebius client of a certain service.
 
     Args:
         service_name: Nebius service name (e.g., 's3').
@@ -207,14 +198,8 @@ def client(service_name: str, region):
     # Reference: https://stackoverflow.com/a/59635814
 
     session_ = session()
-    nebius_credentials = get_nebius_credentials(session_)
-    endpoint = create_endpoint(region)
 
-    return session_.client(service_name,
-                           endpoint_url=endpoint,
-                           aws_access_key_id=nebius_credentials.access_key,
-                           aws_secret_access_key=nebius_credentials.secret_key,
-                           region_name=region)
+    return session_.client(service_name)
 
 
 @common.load_lazy_modules(_LAZY_MODULES)
@@ -223,10 +208,3 @@ def botocore_exceptions():
     # pylint: disable=import-outside-toplevel
     from botocore import exceptions
     return exceptions
-
-
-def create_endpoint(region: Optional[str] = DEFAULT_REGION) -> str:
-    """Reads accountid necessary to interact with Nebius Object Storage"""
-    if region is None:
-        region = DEFAULT_REGION
-    return f'https://storage.{region}.nebius.cloud:443'

sky/backends/backend.py

@@ -1,6 +1,6 @@
 """Sky backend interface."""
 import typing
-from typing import Dict, Generic, Optional
+from typing import Dict, Generic, Optional, Tuple
 
 from sky.usage import usage_lib
 from sky.utils import cluster_utils
@@ -53,7 +53,7 @@ class Backend(Generic[_ResourceHandleType]):
         cluster_name: Optional[str] = None,
         retry_until_up: bool = False,
         skip_unnecessary_provisioning: bool = False,
-    ) -> Optional[_ResourceHandleType]:
+    ) -> Tuple[Optional[_ResourceHandleType], bool]:
         """Provisions resources for the given task.
 
         Args:
@@ -68,13 +68,15 @@ class Backend(Generic[_ResourceHandleType]):
                 the existing cluster will be reused and re-provisioned.
             retry_until_up: If True, retry provisioning until resources are
                 successfully launched.
-            skip_if_no_cluster_updates: If True, compare the cluster config to
-                the existing cluster_name's config. Skip provisioning if no
+            skip_unnecessary_provisioning: If True, compare the cluster config
+                to the existing cluster_name's config. Skip provisioning if no
                 updates are needed for the existing cluster.
 
         Returns:
-            A ResourceHandle object for the provisioned resources, or None if
-            dryrun is True.
+            - A ResourceHandle object for the provisioned resources, or None if
+              dryrun is True.
+            - A boolean that is True if the provisioning was skipped, and False
+              if provisioning actually happened. Dryrun always gives False.
         """
         if cluster_name is None:
             cluster_name = cluster_utils.generate_cluster_name()
@@ -159,7 +161,7 @@ class Backend(Generic[_ResourceHandleType]):
         cluster_name: str,
         retry_until_up: bool = False,
         skip_unnecessary_provisioning: bool = False,
-    ) -> Optional[_ResourceHandleType]:
+    ) -> Tuple[Optional[_ResourceHandleType], bool]:
         raise NotImplementedError
 
     def _sync_workdir(self, handle: _ResourceHandleType, workdir: Path) -> None:

sky/backends/cloud_vm_ray_backend.py

@@ -2829,7 +2829,7 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
         cluster_name: str,
         retry_until_up: bool = False,
         skip_unnecessary_provisioning: bool = False,
-    ) -> Optional[CloudVmRayResourceHandle]:
+    ) -> Tuple[Optional[CloudVmRayResourceHandle], bool]:
         """Provisions the cluster, or re-provisions an existing cluster.
 
         Use the SKYPILOT provisioner if it's supported by the cloud, otherwise
@@ -2969,7 +2969,7 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
                             failover_history=e.failover_history) from None
             if dryrun:
                 record = global_user_state.get_cluster_from_name(cluster_name)
-                return record['handle'] if record is not None else None
+                return record['handle'] if record is not None else None, False
 
             if config_dict['provisioning_skipped']:
                 # Skip further provisioning.
@@ -2980,7 +2980,7 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
                 record = global_user_state.get_cluster_from_name(cluster_name)
                 assert record is not None and record['handle'] is not None, (
                     cluster_name, record)
-                return record['handle']
+                return record['handle'], True
 
             if 'provision_record' in config_dict:
                 # New provisioner is used here.
@@ -3022,7 +3022,7 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
                 self._update_after_cluster_provisioned(
                     handle, to_provision_config.prev_handle, task,
                     prev_cluster_status, lock_path, config_hash)
-                return handle
+                return handle, False
 
             cluster_config_file = config_dict['ray']
             handle = config_dict['handle']
@@ -3094,7 +3094,7 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
             self._update_after_cluster_provisioned(
                 handle, to_provision_config.prev_handle, task,
                 prev_cluster_status, lock_path, config_hash)
-            return handle
+            return handle, False
 
     def _open_ports(self, handle: CloudVmRayResourceHandle) -> None:
         cloud = handle.launched_resources.cloud
@@ -3435,7 +3435,7 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
         mkdir_code = (f'{cd} && mkdir -p {remote_log_dir} && '
                       f'touch {remote_log_path}')
         encoded_script = shlex.quote(codegen)
-        create_script_code = (f'{{ echo {encoded_script} > {script_path}; }}')
+        create_script_code = f'{{ echo {encoded_script} > {script_path}; }}'
         job_submit_cmd = (
             # JOB_CMD_IDENTIFIER is used for identifying the process retrieved
             # with pid is the same driver process.
@@ -4331,7 +4331,7 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
         cluster_name_on_cloud = handle.cluster_name_on_cloud
         cloud = handle.launched_resources.cloud
 
-        if (terminate and handle.launched_resources.is_image_managed is True):
+        if terminate and handle.launched_resources.is_image_managed is True:
             # Delete the image when terminating a "cloned" cluster, i.e.,
             # whose image is created by SkyPilot (--clone-disk-from)
             logger.debug(f'Deleting image {handle.launched_resources.image_id}')

sky/backends/local_docker_backend.py

@@ -139,7 +139,7 @@ class LocalDockerBackend(backends.Backend['LocalDockerResourceHandle']):
         cluster_name: str,
         retry_until_up: bool = False,
         skip_unnecessary_provisioning: bool = False,
-    ) -> Optional[LocalDockerResourceHandle]:
+    ) -> Tuple[Optional[LocalDockerResourceHandle], bool]:
         """Builds docker image for the task and returns cluster name as handle.
 
         Since resource demands are ignored, There's no provisioning in local
@@ -149,7 +149,7 @@ class LocalDockerBackend(backends.Backend['LocalDockerResourceHandle']):
         assert task.name is not None, ('Task name cannot be None - have you '
                                        'specified a task name?')
         if dryrun:
-            return None
+            return None, False
         if retry_until_up:
             logger.warning(
                 f'Retrying until up is not supported in backend: {self.NAME}. '
@@ -175,7 +175,7 @@ class LocalDockerBackend(backends.Backend['LocalDockerResourceHandle']):
                                                 requested_resources=set(
                                                     task.resources),
                                                 ready=False)
-        return handle
+        return handle, False
 
     def _sync_workdir(self, handle: LocalDockerResourceHandle,
                       workdir: Path) -> None:

sky/client/common.py

@@ -75,7 +75,8 @@ def download_logs_from_api_server(
     body = payloads.DownloadBody(folder_paths=list(paths_on_api_server),)
     response = requests.post(f'{server_common.get_server_url()}/download',
                              json=json.loads(body.model_dump_json()),
-                             stream=True)
+                             stream=True,
+                             cookies=server_common.get_api_cookie_jar())
     if response.status_code == 200:
         remote_home_path = response.headers.get('X-Home-Path')
         assert remote_home_path is not None, response.headers
@@ -176,7 +177,8 @@ def _upload_chunk_with_retry(params: UploadChunkParams) -> None:
                 },
                 content=FileChunkIterator(f, _UPLOAD_CHUNK_BYTES,
                                           params.chunk_index),
-                headers={'Content-Type': 'application/octet-stream'})
+                headers={'Content-Type': 'application/octet-stream'},
+                cookies=server_common.get_api_cookie_jar())
             if response.status_code == 200:
                 data = response.json()
                 status = data.get('status')

sky/client/sdk.py

@@ -102,7 +102,8 @@ def check(clouds: Optional[Tuple[str]],
     """
     body = payloads.CheckBody(clouds=clouds, verbose=verbose)
     response = requests.post(f'{server_common.get_server_url()}/check',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -118,7 +119,8 @@ def enabled_clouds() -> server_common.RequestId:
     Request Returns:
         A list of enabled clouds in string format.
     """
-    response = requests.get(f'{server_common.get_server_url()}/enabled_clouds')
+    response = requests.get(f'{server_common.get_server_url()}/enabled_clouds',
+                            cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -168,7 +170,8 @@ def list_accelerators(gpus_only: bool = True,
     )
     response = requests.post(
         f'{server_common.get_server_url()}/list_accelerators',
-        json=json.loads(body.model_dump_json()))
+        json=json.loads(body.model_dump_json()),
+        cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -208,7 +211,8 @@ def list_accelerator_counts(
     )
     response = requests.post(
         f'{server_common.get_server_url()}/list_accelerator_counts',
-        json=json.loads(body.model_dump_json()))
+        json=json.loads(body.model_dump_json()),
+        cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -246,7 +250,8 @@ def optimize(
                                  minimize=minimize,
                                  request_options=admin_policy_request_options)
     response = requests.post(f'{server_common.get_server_url()}/optimize',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -281,7 +286,8 @@ def validate(
     body = payloads.ValidateBody(dag=dag_str,
                                  request_options=admin_policy_request_options)
     response = requests.post(f'{server_common.get_server_url()}/validate',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     if response.status_code == 400:
         with ux_utils.print_exception_no_traceback():
             raise exceptions.deserialize_exception(
@@ -493,6 +499,7 @@ def launch(
         f'{server_common.get_server_url()}/launch',
         json=json.loads(body.model_dump_json()),
         timeout=5,
+        cookies=server_common.get_api_cookie_jar(),
     )
     return server_common.get_request_id(response)
 
@@ -576,6 +583,7 @@ def exec(  # pylint: disable=redefined-builtin
         f'{server_common.get_server_url()}/exec',
         json=json.loads(body.model_dump_json()),
         timeout=5,
+        cookies=server_common.get_api_cookie_jar(),
     )
     return server_common.get_request_id(response)
 
@@ -626,7 +634,8 @@ def tail_logs(cluster_name: str,
         json=json.loads(body.model_dump_json()),
         stream=True,
         timeout=(client_common.API_SERVER_REQUEST_CONNECTION_TIMEOUT_SECONDS,
-                 None))
+                 None),
+        cookies=server_common.get_api_cookie_jar())
     request_id = server_common.get_request_id(response)
     return stream_response(request_id, response, output_stream)
 
@@ -663,7 +672,8 @@ def download_logs(cluster_name: str,
         job_ids=job_ids,
     )
     response = requests.post(f'{server_common.get_server_url()}/download_logs',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     job_id_remote_path_dict = stream_and_get(
         server_common.get_request_id(response))
     remote2local_path_dict = client_common.download_logs_from_api_server(
@@ -745,6 +755,7 @@ def start(
         f'{server_common.get_server_url()}/start',
         json=json.loads(body.model_dump_json()),
         timeout=5,
+        cookies=server_common.get_api_cookie_jar(),
     )
     return server_common.get_request_id(response)
 
@@ -790,6 +801,7 @@ def down(cluster_name: str, purge: bool = False) -> server_common.RequestId:
         f'{server_common.get_server_url()}/down',
         json=json.loads(body.model_dump_json()),
         timeout=5,
+        cookies=server_common.get_api_cookie_jar(),
     )
     return server_common.get_request_id(response)
 
@@ -838,6 +850,7 @@ def stop(cluster_name: str, purge: bool = False) -> server_common.RequestId:
         f'{server_common.get_server_url()}/stop',
         json=json.loads(body.model_dump_json()),
         timeout=5,
+        cookies=server_common.get_api_cookie_jar(),
     )
     return server_common.get_request_id(response)
 
@@ -907,6 +920,7 @@ def autostop(
         f'{server_common.get_server_url()}/autostop',
         json=json.loads(body.model_dump_json()),
         timeout=5,
+        cookies=server_common.get_api_cookie_jar(),
     )
     return server_common.get_request_id(response)
 
@@ -966,7 +980,8 @@ def queue(cluster_name: str,
         all_users=all_users,
     )
     response = requests.post(f'{server_common.get_server_url()}/queue',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1007,7 +1022,8 @@ def job_status(cluster_name: str,
         job_ids=job_ids,
     )
     response = requests.post(f'{server_common.get_server_url()}/job_status',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1060,7 +1076,8 @@ def cancel(
         try_cancel_if_cluster_is_init=_try_cancel_if_cluster_is_init,
     )
     response = requests.post(f'{server_common.get_server_url()}/cancel',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1155,7 +1172,8 @@ def status(
         all_users=all_users,
     )
     response = requests.post(f'{server_common.get_server_url()}/status',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1189,7 +1207,8 @@ def endpoints(
         port=port,
     )
     response = requests.post(f'{server_common.get_server_url()}/endpoints',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1227,7 +1246,8 @@ def cost_report() -> server_common.RequestId:  # pylint: disable=redefined-built
               'total_cost': (float) cost given resources and usage intervals,
             }
     """
-    response = requests.get(f'{server_common.get_server_url()}/cost_report')
+    response = requests.get(f'{server_common.get_server_url()}/cost_report',
+                            cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1256,7 +1276,8 @@ def storage_ls() -> server_common.RequestId:
                 }
         ]
     """
-    response = requests.get(f'{server_common.get_server_url()}/storage/ls')
+    response = requests.get(f'{server_common.get_server_url()}/storage/ls',
+                            cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1280,7 +1301,8 @@ def storage_delete(name: str) -> server_common.RequestId:
     """
     body = payloads.StorageBody(name=name)
     response = requests.post(f'{server_common.get_server_url()}/storage/delete',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1318,7 +1340,8 @@ def local_up(gpus: bool,
                                 context_name=context_name,
                                 password=password)
     response = requests.post(f'{server_common.get_server_url()}/local_up',
-                             json=json.loads(body.model_dump_json()))
+                             json=json.loads(body.model_dump_json()),
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1334,7 +1357,8 @@ def local_down() -> server_common.RequestId:
         with ux_utils.print_exception_no_traceback():
             raise ValueError('sky local down is only supported when running '
                              'SkyPilot locally.')
-    response = requests.post(f'{server_common.get_server_url()}/local_down')
+    response = requests.post(f'{server_common.get_server_url()}/local_down',
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1358,7 +1382,8 @@ def realtime_kubernetes_gpu_availability(
     response = requests.post(
         f'{server_common.get_server_url()}/'
         'realtime_kubernetes_gpu_availability',
-        json=json.loads(body.model_dump_json()))
+        json=json.loads(body.model_dump_json()),
+        cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1389,7 +1414,8 @@ def kubernetes_node_info(
     body = payloads.KubernetesNodeInfoRequestBody(context=context)
     response = requests.post(
         f'{server_common.get_server_url()}/kubernetes_node_info',
-        json=json.loads(body.model_dump_json()))
+        json=json.loads(body.model_dump_json()),
+        cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1418,7 +1444,8 @@ def status_kubernetes() -> server_common.RequestId:
         - context: Kubernetes context used to fetch the cluster information.
     """
     response = requests.get(
-        f'{server_common.get_server_url()}/status_kubernetes')
+        f'{server_common.get_server_url()}/status_kubernetes',
+        cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1444,7 +1471,8 @@ def get(request_id: str) -> Any:
     response = requests.get(
         f'{server_common.get_server_url()}/api/get?request_id={request_id}',
         timeout=(client_common.API_SERVER_REQUEST_CONNECTION_TIMEOUT_SECONDS,
-                 None))
+                 None),
+        cookies=server_common.get_api_cookie_jar())
     request_task = None
     if response.status_code == 200:
         request_task = requests_lib.Request.decode(
@@ -1523,7 +1551,8 @@ def stream_and_get(
         params=params,
         timeout=(client_common.API_SERVER_REQUEST_CONNECTION_TIMEOUT_SECONDS,
                  None),
-        stream=True)
+        stream=True,
+        cookies=server_common.get_api_cookie_jar())
     if response.status_code in [404, 400]:
         detail = response.json().get('detail')
         with ux_utils.print_exception_no_traceback():
@@ -1579,7 +1608,8 @@ def api_cancel(request_ids: Optional[Union[str, List[str]]] = None,
 
     response = requests.post(f'{server_common.get_server_url()}/api/cancel',
                              json=json.loads(body.model_dump_json()),
-                             timeout=5)
+                             timeout=5,
+                             cookies=server_common.get_api_cookie_jar())
     return server_common.get_request_id(response)
 
 
@@ -1607,7 +1637,8 @@ def api_status(
         f'{server_common.get_server_url()}/api/status',
         params=server_common.request_body_to_params(body),
         timeout=(client_common.API_SERVER_REQUEST_CONNECTION_TIMEOUT_SECONDS,
-                 None))
+                 None),
+        cookies=server_common.get_api_cookie_jar())
     server_common.handle_request_error(response)
     return [
         requests_lib.RequestPayload(**request) for request in response.json()
@@ -1634,7 +1665,8 @@ def api_info() -> Dict[str, str]:
             }
 
     """
-    response = requests.get(f'{server_common.get_server_url()}/api/health')
+    response = requests.get(f'{server_common.get_server_url()}/api/health',
+                            cookies=server_common.get_api_cookie_jar())
     response.raise_for_status()
     return response.json()
 

sky/cloud_stores.py

@@ -578,13 +578,11 @@ class NebiusCloudStorage(CloudStorage):
         # AWS Sync by default uses 10 threads to upload files to the bucket.
         # To increase parallelism, modify max_concurrent_requests in your
         # aws config file (Default path: ~/.aws/config).
-        endpoint_url = nebius.create_endpoint()
         assert 'nebius://' in source, 'nebius:// is not in source'
         source = source.replace('nebius://', 's3://')
         download_via_awscli = (f'{constants.SKY_REMOTE_PYTHON_ENV}/bin/aws s3 '
                                'sync --no-follow-symlinks '
                                f'{source} {destination} '
-                               f'--endpoint {endpoint_url} '
                                f'--profile={nebius.NEBIUS_PROFILE_NAME}')
 
         all_commands = list(self._GET_AWSCLI)
@@ -593,12 +591,10 @@ class NebiusCloudStorage(CloudStorage):
 
     def make_sync_file_command(self, source: str, destination: str) -> str:
         """Downloads a file using AWS CLI."""
-        endpoint_url = nebius.create_endpoint()
         assert 'nebius://' in source, 'nebius:// is not in source'
         source = source.replace('nebius://', 's3://')
         download_via_awscli = (f'{constants.SKY_REMOTE_PYTHON_ENV}/bin/aws s3 '
                                f'cp {source} {destination} '
-                               f'--endpoint {endpoint_url} '
                                f'--profile={nebius.NEBIUS_PROFILE_NAME}')
 
         all_commands = list(self._GET_AWSCLI)

sky/clouds/do.py

@@ -280,13 +280,12 @@ class DO(clouds.Cloud):
         return True, None
 
     def get_credential_file_mounts(self) -> Dict[str, str]:
-        if do_utils.CREDENTIALS_PATH is None:
+        credential_path = do_utils.get_credentials_path()
+        if credential_path is None:
             return {}
-        if not os.path.exists(os.path.expanduser(do_utils.CREDENTIALS_PATH)):
+        if not os.path.exists(os.path.expanduser(credential_path)):
             return {}
-        return {
-            f'~/.config/doctl/{_CREDENTIAL_FILE}': do_utils.CREDENTIALS_PATH
-        }
+        return {f'~/.config/doctl/{_CREDENTIAL_FILE}': credential_path}
 
     @classmethod
     def get_current_user_identity(cls) -> Optional[List[str]]:

sky/clouds/gcp.py

@@ -688,9 +688,11 @@ class GCP(clouds.Cloud):
         cls,
         instance_type: str,
     ) -> Optional[Dict[str, Union[int, float]]]:
-        # GCP handles accelerators separately from regular instance types,
-        # hence return none here.
-        return None
+        # GCP handles accelerators separately from regular instance types.
+        # This method supports automatically inferring the GPU type for
+        # the instance type that come with GPUs pre-attached.
+        return service_catalog.get_accelerators_from_instance_type(
+            instance_type, clouds='gcp')
 
     @classmethod
     def get_vcpus_mem_from_instance_type(