skypilot-nightly 1.0.0.dev20250212__tar.gz → 1.0.0.dev20250213__tar.gz

{skypilot_nightly-1.0.0.dev20250212/skypilot_nightly.egg-info → skypilot_nightly-1.0.0.dev20250213}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: skypilot-nightly
-Version: 1.0.0.dev20250212
+Version: 1.0.0.dev20250213
 Summary: SkyPilot: An intercloud broker for the clouds
 Author: SkyPilot Team
 License: Apache 2.0

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/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 = '1fe3fab0e7a3242f32039d55b456603350dc4196'
+_SKYPILOT_COMMIT_SHA = 'c49961417a83b049b3f3435a252c8ec5ea0fb5e6'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20250212'
+__version__ = '1.0.0.dev20250213'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/clouds/kubernetes.py

@@ -464,7 +464,9 @@ class Kubernetes(clouds.Cloud):
         # CPU resources on the node instead within the pod.
         custom_ray_options = {
             'object-store-memory': 500000000,
-            'num-cpus': str(int(cpus)),
+            # 'num-cpus' must be an integer, but we should not set it to 0 if
+            # cpus is <1.
+            'num-cpus': str(max(int(cpus), 1)),
         }
         deploy_vars = {
             'instance_type': resources.instance_type,

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/clouds/runpod.py

@@ -177,6 +177,11 @@ class RunPod(clouds.Cloud):
         hourly_cost = self.instance_type_to_hourly_cost(
             instance_type=instance_type, use_spot=use_spot)
 
+        # default to root
+        docker_username_for_runpod = (resources.docker_username_for_runpod
+                                      if resources.docker_username_for_runpod
+                                      is not None else 'root')
+
         return {
             'instance_type': instance_type,
             'custom_resources': custom_resources,
@@ -184,6 +189,7 @@ class RunPod(clouds.Cloud):
             'image_id': image_id,
             'use_spot': use_spot,
             'bid_per_gpu': str(hourly_cost),
+            'docker_username_for_runpod': docker_username_for_runpod,
         }
 
     def _get_feasible_launchable_resources(

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/clouds/service_catalog/data_fetchers/fetch_lambda_cloud.py

@@ -28,6 +28,7 @@ REGIONS = [
     'asia-northeast-2',
     'us-east-1',
     'us-east-2',
+    'us-east-3',
     'us-west-2',
     'us-west-1',
     'us-south-1',

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/provision/provisioner.py

@@ -450,6 +450,13 @@ def _post_provision_setup(
         logger.info(f'{indent_str}{colorama.Style.DIM}{vm_str}{plural} {verb} '
                     f'up.{colorama.Style.RESET_ALL}')
 
+        # It's promised by the cluster config that docker_config does not
+        # exist for docker-native clouds, i.e. they provide docker containers
+        # instead of full VMs, like Kubernetes and RunPod, as it requires some
+        # special handlings to run docker inside their docker virtualization.
+        # For their Docker image settings, we do them when provisioning the
+        # cluster. See provision/{cloud}/instance.py:get_cluster_info for more
+        # details.
         if docker_config:
             status.update(
                 ux_utils.spinner_message(

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/provision/runpod/utils.py

@@ -186,7 +186,7 @@ def delete_pod_template(template_name: str) -> None:
         runpod.runpod.api.graphql.run_graphql_query(
             f'mutation {{deleteTemplate(templateName: "{template_name}")}}')
     except runpod.runpod.error.QueryError as e:
-        logger.warning(f'Failed to delete template {template_name}: {e}'
+        logger.warning(f'Failed to delete template {template_name}: {e} '
                        'Please delete it manually.')
 
 
@@ -195,8 +195,9 @@ def delete_register_auth(registry_auth_id: str) -> None:
     try:
         runpod.runpod.delete_container_registry_auth(registry_auth_id)
     except runpod.runpod.error.QueryError as e:
-        logger.warning(f'Failed to delete registry auth {registry_auth_id}: {e}'
-                       'Please delete it manually.')
+        logger.warning(
+            f'Failed to delete registry auth {registry_auth_id}: {e} '
+            'Please delete it manually.')
 
 
 def _create_template_for_docker_login(

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/resources.py

@@ -67,6 +67,7 @@ class Resources:
         # Internal use only.
         # pylint: disable=invalid-name
         _docker_login_config: Optional[docker_utils.DockerLoginConfig] = None,
+        _docker_username_for_runpod: Optional[str] = None,
         _is_image_managed: Optional[bool] = None,
         _requires_fuse: Optional[bool] = None,
         _cluster_config_overrides: Optional[Dict[str, Any]] = None,
@@ -148,6 +149,9 @@ class Resources:
           _docker_login_config: the docker configuration to use. This includes
             the docker username, password, and registry server. If None, skip
             docker login.
+          _docker_username_for_runpod: the login username for the docker
+            containers. This is used by RunPod to set the ssh user for the
+            docker containers.
           _requires_fuse: whether the task requires FUSE mounting support. This
             is used internally by certain cloud implementations to do additional
             setup for FUSE mounting. This flag also safeguards against using
@@ -234,6 +238,12 @@ class Resources:
 
         self._docker_login_config = _docker_login_config
 
+        # TODO(andyl): This ctor param seems to be unused.
+        # We always use `Task.set_resources` and `Resources.copy` to set the
+        # `docker_username_for_runpod`. But to keep the consistency with
+        # `_docker_login_config`, we keep it here.
+        self._docker_username_for_runpod = _docker_username_for_runpod
+
         self._requires_fuse = _requires_fuse
 
         self._cluster_config_overrides = _cluster_config_overrides
@@ -479,6 +489,10 @@ class Resources:
     def requires_fuse(self, value: Optional[bool]) -> None:
         self._requires_fuse = value
 
+    @property
+    def docker_username_for_runpod(self) -> Optional[str]:
+        return self._docker_username_for_runpod
+
     def _set_cpus(
         self,
         cpus: Union[None, int, float, str],
@@ -1065,6 +1079,10 @@ class Resources:
         cloud_specific_variables = self.cloud.make_deploy_resources_variables(
             self, cluster_name, region, zones, num_nodes, dryrun)
 
+        # TODO(andyl): Should we print some warnings if users' envs share
+        # same names with the cloud specific variables, but not enabled
+        # since it's not on the particular cloud?
+
         # Docker run options
         docker_run_options = skypilot_config.get_nested(
             ('docker', 'run_options'),
@@ -1277,6 +1295,9 @@ class Resources:
             labels=override.pop('labels', self.labels),
             _docker_login_config=override.pop('_docker_login_config',
                                               self._docker_login_config),
+            _docker_username_for_runpod=override.pop(
+                '_docker_username_for_runpod',
+                self._docker_username_for_runpod),
             _is_image_managed=override.pop('_is_image_managed',
                                            self._is_image_managed),
             _requires_fuse=override.pop('_requires_fuse', self._requires_fuse),
@@ -1438,6 +1459,8 @@ class Resources:
         resources_fields['labels'] = config.pop('labels', None)
         resources_fields['_docker_login_config'] = config.pop(
             '_docker_login_config', None)
+        resources_fields['_docker_username_for_runpod'] = config.pop(
+            '_docker_username_for_runpod', None)
         resources_fields['_is_image_managed'] = config.pop(
             '_is_image_managed', None)
         resources_fields['_requires_fuse'] = config.pop('_requires_fuse', None)
@@ -1486,6 +1509,9 @@ class Resources:
         if self._docker_login_config is not None:
             config['_docker_login_config'] = dataclasses.asdict(
                 self._docker_login_config)
+        if self._docker_username_for_runpod is not None:
+            config['_docker_username_for_runpod'] = (
+                self._docker_username_for_runpod)
         add_if_not_none('_cluster_config_overrides',
                         self._cluster_config_overrides)
         if self._is_image_managed is not None:

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/skylet/constants.py

@@ -110,6 +110,8 @@ DOCKER_LOGIN_ENV_VARS = {
     DOCKER_SERVER_ENV_VAR,
 }
 
+RUNPOD_DOCKER_USERNAME_ENV_VAR = 'SKYPILOT_RUNPOD_DOCKER_USERNAME'
+
 # Commands for disable GPU ECC, which can improve the performance of the GPU
 # for some workloads by 30%. This will only be applied when a user specify
 # `nvidia_gpus.disable_ecc: true` in ~/.sky/config.yaml.

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/task.py

@@ -121,6 +121,9 @@ def _check_docker_login_config(task_envs: Dict[str, str]) -> bool:
 
     If any of the docker login env vars is set, all of them must be set.
 
+    Returns:
+        True if there is a valid docker login config in task_envs.
+        False otherwise.
     Raises:
         ValueError: if any of the docker login env vars is set, but not all of
             them are set.
@@ -168,6 +171,23 @@ def _with_docker_login_config(
     return type(resources)(new_resources)
 
 
+def _with_docker_username_for_runpod(
+    resources: Union[Set['resources_lib.Resources'],
+                     List['resources_lib.Resources']],
+    task_envs: Dict[str, str],
+) -> Union[Set['resources_lib.Resources'], List['resources_lib.Resources']]:
+    docker_username_for_runpod = task_envs.get(
+        constants.RUNPOD_DOCKER_USERNAME_ENV_VAR)
+
+    # We should not call r.copy() if docker_username_for_runpod is None,
+    # to prevent `DummyResources` instance becoming a `Resources` instance.
+    if docker_username_for_runpod is None:
+        return resources
+    return (type(resources)(
+        r.copy(_docker_username_for_runpod=docker_username_for_runpod)
+        for r in resources))
+
+
 class Task:
     """Task: a computation to be run on the cloud."""
 
@@ -582,6 +602,8 @@ class Task:
         if _check_docker_login_config(self._envs):
             self.resources = _with_docker_login_config(self.resources,
                                                        self._envs)
+        self.resources = _with_docker_username_for_runpod(
+            self.resources, self._envs)
         return self
 
     @property
@@ -647,6 +669,9 @@ class Task:
             resources = {resources}
         # TODO(woosuk): Check if the resources are None.
         self.resources = _with_docker_login_config(resources, self.envs)
+        # Only have effect on RunPod.
+        self.resources = _with_docker_username_for_runpod(
+            self.resources, self.envs)
 
         # Evaluate if the task requires FUSE and set the requires_fuse flag
         for _, storage_obj in self.storage_mounts.items():

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213}/sky/templates/runpod-ray.yml.j2

@@ -25,7 +25,7 @@ provider:
   {%- endif %}
 
 auth:
-  ssh_user: root
+  ssh_user: {{docker_username_for_runpod}}
   ssh_private_key: {{ssh_private_key}}
 
 available_node_types:

{skypilot_nightly-1.0.0.dev20250212 → skypilot_nightly-1.0.0.dev20250213/skypilot_nightly.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: skypilot-nightly
-Version: 1.0.0.dev20250212
+Version: 1.0.0.dev20250213
 Summary: SkyPilot: An intercloud broker for the clouds
 Author: SkyPilot Team
 License: Apache 2.0