skypilot-nightly 1.0.0.dev2024053101__py3-none-any.whl → 1.0.0.dev2025022801__py3-none-any.whl

sky/clouds/oci.py

@@ -4,21 +4,37 @@ History:
  - Hysun He (hysun.he@oracle.com) @ Apr, 2023: Initial implementation
  - Hysun He (hysun.he@oracle.com) @ May 4, 2023: Support use the default
    image_id (configurable) if no image_id specified in the task yaml.
+ - Hysun He (hysun.he@oracle.com) @ Oct 12, 2024:
+   get_credential_file_mounts(): bug fix for sky config
+   file path resolution (by os.path.expanduser) when construct the file
+   mounts. This bug will cause the created workder nodes located in different
+   compartment and VCN than the header node if user specifies compartment_id
+   in the sky config file, because the ~/.sky/config.yaml is not sync-ed to the
+   remote machine.
+   The workaround is set the sky config file path using ENV before running
+   the sky launch: export SKYPILOT_CONFIG=/home/ubuntu/.sky/config.yaml
+ - Hysun He (hysun.he@oracle.com) @ Oct 12, 2024:
+   make_deploy_resources_variables(): Bug fix for specify the image_id as
+   the ocid of the image in the task.yaml file, in this case the image_id
+   for the node config should be set to the ocid instead of a dict.
+ - Hysun He (hysun.he@oracle.com) @ Oct 13, 2024:
+   Support more OS types additional to ubuntu for OCI resources.
 """
-import json
 import logging
 import os
 import typing
-from typing import Dict, Iterator, List, Optional, Tuple
+from typing import Dict, Iterator, List, Optional, Tuple, Union
 
 from sky import clouds
 from sky import exceptions
-from sky import status_lib
 from sky.adaptors import oci as oci_adaptor
 from sky.clouds import service_catalog
 from sky.clouds.utils import oci_utils
+from sky.provision.oci.query_utils import query_helper
 from sky.utils import common_utils
+from sky.utils import registry
 from sky.utils import resources_utils
+from sky.utils import status_lib
 from sky.utils import ux_utils
 
 if typing.TYPE_CHECKING:
@@ -30,7 +46,7 @@ logger = logging.getLogger(__name__)
 _tenancy_prefix: Optional[str] = None
 
 
-@clouds.CLOUD_REGISTRY.register
+@registry.CLOUD_REGISTRY.register
 class OCI(clouds.Cloud):
     """OCI: Oracle Cloud Infrastructure """
 
@@ -42,7 +58,12 @@ class OCI(clouds.Cloud):
 
     _INDENT_PREFIX = '    '
 
-    _SUPPORTED_DISK_TIERS = set(resources_utils.DiskTier)
+    _SUPPORTED_DISK_TIERS = (set(resources_utils.DiskTier) -
+                             {resources_utils.DiskTier.ULTRA})
+    _BEST_DISK_TIER = resources_utils.DiskTier.HIGH
+
+    PROVISIONER_VERSION = clouds.ProvisionerVersion.SKYPILOT
+    STATUS_VERSION = clouds.StatusVersion.SKYPILOT
 
     @classmethod
     def _unsupported_features_for_resources(
@@ -55,8 +76,6 @@ class OCI(clouds.Cloud):
                 (f'Docker image is currently not supported on {cls._REPR}. '
                  'You can try running docker command inside the '
                  '`run` section in task.yaml.'),
-            clouds.CloudImplementationFeatures.OPEN_PORTS:
-                (f'Opening ports is currently not supported on {cls._REPR}.'),
         }
         if resources.use_spot:
             features[clouds.CloudImplementationFeatures.STOP] = (
@@ -176,7 +195,7 @@ class OCI(clouds.Cloud):
     def get_accelerators_from_instance_type(
         cls,
         instance_type: str,
-    ) -> Optional[Dict[str, int]]:
+    ) -> Optional[Dict[str, Union[int, float]]]:
         return service_catalog.get_accelerators_from_instance_type(
             instance_type, clouds='oci')
 
@@ -187,19 +206,18 @@ class OCI(clouds.Cloud):
     def make_deploy_resources_variables(
             self,
             resources: 'resources_lib.Resources',
-            cluster_name_on_cloud: str,
+            cluster_name: resources_utils.ClusterName,
             region: Optional['clouds.Region'],
             zones: Optional[List['clouds.Zone']],
+            num_nodes: int,
             dryrun: bool = False) -> Dict[str, Optional[str]]:
-        del cluster_name_on_cloud, dryrun  # Unused.
+        del cluster_name, dryrun  # Unused.
         assert region is not None, resources
 
         acc_dict = self.get_accelerators_from_instance_type(
             resources.instance_type)
-        if acc_dict is not None:
-            custom_resources = json.dumps(acc_dict, separators=(',', ':'))
-        else:
-            custom_resources = None
+        custom_resources = resources_utils.make_ray_custom_resources_str(
+            acc_dict)
 
         image_str = self._get_image_id(resources.image_id, region.name,
                                        resources.instance_type)
@@ -209,10 +227,20 @@ class OCI(clouds.Cloud):
             listing_id = image_cols[1]
             res_ver = image_cols[2]
         else:
-            image_id = resources.image_id
+            # Oct.12,2024 by HysunHe: Bug fix - resources.image_id is an
+            # dict. The image_id here should be the ocid format.
+            image_id = image_str
             listing_id = None
             res_ver = None
 
+        os_type = None
+        if ':' in image_id:
+            # OS type provided in the --image-id. This is the case where
+            # custom image's ocid provided in the --image-id parameter.
+            #  - ocid1.image...aaa:oraclelinux (os type is oraclelinux)
+            #  - ocid1.image...aaa (OS not provided)
+            image_id, os_type = image_id.replace(' ', '').split(':')
+
         cpus = resources.cpus
         instance_type_arr = resources.instance_type.split(
             oci_utils.oci_config.INSTANCE_TYPE_RES_SPERATOR)
@@ -278,10 +306,24 @@ class OCI(clouds.Cloud):
             cpus=None if cpus is None else float(cpus),
             disk_tier=resources.disk_tier)
 
+        if os_type is None:
+            # OS type is not determined yet. So try to get it from vms.csv
+            image_str = self._get_image_str(
+                image_id=resources.image_id,
+                instance_type=resources.instance_type,
+                region=region.name)
+
+            # pylint: disable=import-outside-toplevel
+            from sky.clouds.service_catalog import oci_catalog
+            os_type = oci_catalog.get_image_os_from_tag(tag=image_str,
+                                                        region=region.name)
+        logger.debug(f'OS type for the image {image_id} is {os_type}')
+
         return {
             'instance_type': instance_type,
             'custom_resources': custom_resources,
             'region': region.name,
+            'os_type': os_type,
             'cpus': str(cpus),
             'memory': resources.memory,
             'disk_size': resources.disk_size,
@@ -295,11 +337,13 @@ class OCI(clouds.Cloud):
 
     def _get_feasible_launchable_resources(
         self, resources: 'resources_lib.Resources'
-    ) -> Tuple[List['resources_lib.Resources'], List[str]]:
+    ) -> 'resources_utils.FeasibleResources':
         if resources.instance_type is not None:
             assert resources.is_launchable(), resources
             resources = resources.copy(accelerators=None)
-            return ([resources], [])
+            # TODO: Add hints to all return values in this method to help
+            #  users understand why the resources are not launchable.
+            return resources_utils.FeasibleResources([resources], [], None)
 
         def _make(instance_list):
             resource_list = []
@@ -326,9 +370,10 @@ class OCI(clouds.Cloud):
                 disk_tier=resources.disk_tier)
 
             if default_instance_type is None:
-                return ([], [])
+                return resources_utils.FeasibleResources([], [], None)
             else:
-                return (_make([default_instance_type]), [])
+                return resources_utils.FeasibleResources(
+                    _make([default_instance_type]), [], None)
 
         assert len(accelerators) == 1, resources
 
@@ -344,9 +389,11 @@ class OCI(clouds.Cloud):
             zone=resources.zone,
             clouds='oci')
         if instance_list is None:
-            return ([], fuzzy_candidate_list)
+            return resources_utils.FeasibleResources([], fuzzy_candidate_list,
+                                                     None)
 
-        return (_make(instance_list), fuzzy_candidate_list)
+        return resources_utils.FeasibleResources(_make(instance_list),
+                                                 fuzzy_candidate_list, None)
 
     @classmethod
     def check_credentials(cls) -> Tuple[bool, Optional[str]]:
@@ -355,7 +402,7 @@ class OCI(clouds.Cloud):
         short_credential_help_str = (
             'For more details, refer to: '
             # pylint: disable=line-too-long
-            'https://skypilot.readthedocs.io/en/latest/getting-started/installation.html#oracle-cloud-infrastructure-oci'
+            'https://docs.skypilot.co/en/latest/getting-started/installation.html#oracle-cloud-infrastructure-oci'
         )
         credential_help_str = (
             'To configure credentials, go to: '
@@ -401,7 +448,7 @@ class OCI(clouds.Cloud):
             return True, None
         except (oci_adaptor.oci.exceptions.ConfigFileNotFound,
                 oci_adaptor.oci.exceptions.InvalidConfig,
-                oci_adaptor.service_exception()) as e:
+                oci_adaptor.oci.exceptions.ServiceError) as e:
             return False, (
                 f'OCI credential is not correctly set. '
                 f'Check the credential file at {conf_file}\n'
@@ -409,22 +456,42 @@ class OCI(clouds.Cloud):
                 f'{cls._INDENT_PREFIX}Error details: '
                 f'{common_utils.format_exception(e, use_bracket=True)}')
 
+    @classmethod
+    def check_disk_tier(
+            cls, instance_type: Optional[str],
+            disk_tier: Optional[resources_utils.DiskTier]) -> Tuple[bool, str]:
+        del instance_type  # Unused.
+        if disk_tier is None or disk_tier == resources_utils.DiskTier.BEST:
+            return True, ''
+        if disk_tier == resources_utils.DiskTier.ULTRA:
+            return False, ('OCI disk_tier=ultra is not supported now. '
+                           'Please use disk_tier={low, medium, high, best} '
+                           'instead.')
+        return True, ''
+
     def get_credential_file_mounts(self) -> Dict[str, str]:
         """Returns a dict of credential file paths to mount paths."""
-        oci_cfg_file = oci_adaptor.get_config_file()
-        # Pass-in a profile parameter so that multiple profile in oci
-        # config file is supported (2023/06/09).
-        oci_cfg = oci_adaptor.get_oci_config(
-            profile=oci_utils.oci_config.get_profile())
-        api_key_file = oci_cfg[
-            'key_file'] if 'key_file' in oci_cfg else 'BadConf'
-        sky_cfg_file = oci_utils.oci_config.get_sky_user_config_file()
+        try:
+            oci_cfg_file = oci_adaptor.get_config_file()
+            # Pass-in a profile parameter so that multiple profile in oci
+            # config file is supported (2023/06/09).
+            oci_cfg = oci_adaptor.get_oci_config(
+                profile=oci_utils.oci_config.get_profile())
+            api_key_file = oci_cfg[
+                'key_file'] if 'key_file' in oci_cfg else 'BadConf'
+            sky_cfg_file = oci_utils.oci_config.get_sky_user_config_file()
+        # Must catch ImportError before any oci_adaptor.oci.exceptions
+        # because oci_adaptor.oci.exceptions can throw ImportError.
+        except ImportError:
+            return {}
+        except oci_adaptor.oci.exceptions.ConfigFileNotFound:
+            return {}
 
         # OCI config and API key file are mandatory
         credential_files = [oci_cfg_file, api_key_file]
 
         # Sky config file is optional
-        if os.path.exists(sky_cfg_file):
+        if os.path.exists(os.path.expanduser(sky_cfg_file)):
             credential_files.append(sky_cfg_file)
 
         file_mounts = {
@@ -435,7 +502,7 @@ class OCI(clouds.Cloud):
         return file_mounts
 
     @classmethod
-    def get_current_user_identity(cls) -> Optional[List[str]]:
+    def get_user_identities(cls) -> Optional[List[List[str]]]:
         # NOTE: used for very advanced SkyPilot functionality
         # Can implement later if desired
         # If the user switches the compartment_ocid, the existing clusters
@@ -463,59 +530,45 @@ class OCI(clouds.Cloud):
         region_name: str,
         instance_type: str,
     ) -> str:
-        if image_id is None:
-            return self._get_default_image(region_name=region_name,
-                                           instance_type=instance_type)
-        if None in image_id:
-            image_id_str = image_id[None]
-        else:
-            assert region_name in image_id, image_id
-            image_id_str = image_id[region_name]
+        image_id_str = self._get_image_str(image_id=image_id,
+                                           instance_type=instance_type,
+                                           region=region_name)
+
         if image_id_str.startswith('skypilot:'):
             image_id_str = service_catalog.get_image_id_from_tag(image_id_str,
                                                                  region_name,
                                                                  clouds='oci')
-            if image_id_str is None:
-                logger.critical(
-                    '! Real image_id not found! - {region_name}:{image_id}')
-                # Raise ResourcesUnavailableError to make sure the failover
-                # in CloudVMRayBackend will be correctly triggered.
-                # TODO(zhwu): This is a information leakage to the cloud
-                # implementor, we need to find a better way to handle this.
-                raise exceptions.ResourcesUnavailableError(
-                    '! ERR: No image found in catalog for region '
-                    f'{region_name}. Try setting a valid image_id.')
+
+        # Image_id should be impossible be None, except for the case when
+        # user specify an image tag which does not exist in the image.csv
+        # catalog file which only possible in "test" / "evaluation" phase.
+        # Therefore, we use assert here.
+        assert image_id_str is not None
 
         logger.debug(f'Got real image_id {image_id_str}')
         return image_id_str
 
-    def _get_default_image(self, region_name: str, instance_type: str) -> str:
+    def _get_image_str(self, image_id: Optional[Dict[Optional[str], str]],
+                       instance_type: str, region: str):
+        if image_id is None:
+            image_str = self._get_default_image_tag(instance_type)
+        elif None in image_id:
+            image_str = image_id[None]
+        else:
+            assert region in image_id, image_id
+            image_str = image_id[region]
+        return image_str
+
+    def _get_default_image_tag(self, instance_type: str) -> str:
         acc = self.get_accelerators_from_instance_type(instance_type)
 
         if acc is None:
             image_tag = oci_utils.oci_config.get_default_image_tag()
-            image_id_str = service_catalog.get_image_id_from_tag(image_tag,
-                                                                 region_name,
-                                                                 clouds='oci')
         else:
             assert len(acc) == 1, acc
             image_tag = oci_utils.oci_config.get_default_gpu_image_tag()
-            image_id_str = service_catalog.get_image_id_from_tag(image_tag,
-                                                                 region_name,
-                                                                 clouds='oci')
-
-        if image_id_str is not None:
-            logger.debug(
-                f'Got default image_id {image_id_str} from tag {image_tag}')
-            return image_id_str
 
-        # Raise ResourcesUnavailableError to make sure the failover in
-        # CloudVMRayBackend will be correctly triggered.
-        # TODO(zhwu): This is a information leakage to the cloud implementor,
-        # we need to find a better way to handle this.
-        raise exceptions.ResourcesUnavailableError(
-            'ERR: No image found in catalog for region '
-            f'{region_name}. Try update your default image_id settings.')
+        return image_tag
 
     def get_vpu_from_disktier(
             self, cpus: Optional[float],
@@ -559,25 +612,11 @@ class OCI(clouds.Cloud):
                      region: Optional[str], zone: Optional[str],
                      **kwargs) -> List[status_lib.ClusterStatus]:
         del zone, kwargs  # Unused.
-        # Check the lifecycleState definition from the page
-        # https://docs.oracle.com/en-us/iaas/api/#/en/iaas/latest/Instance/
-        status_map = {
-            'PROVISIONING': status_lib.ClusterStatus.INIT,
-            'STARTING': status_lib.ClusterStatus.INIT,
-            'RUNNING': status_lib.ClusterStatus.UP,
-            'STOPPING': status_lib.ClusterStatus.STOPPED,
-            'STOPPED': status_lib.ClusterStatus.STOPPED,
-            'TERMINATED': None,
-            'TERMINATING': None,
-        }
-
-        # pylint: disable=import-outside-toplevel
-        from sky.skylet.providers.oci.query_helper import oci_query_helper
 
         status_list = []
         try:
-            vms = oci_query_helper.query_instances_by_tags(
-                tag_filters=tag_filters, region=region)
+            vms = query_helper.query_instances_by_tags(tag_filters=tag_filters,
+                                                       region=region)
         except Exception as e:  # pylint: disable=broad-except
             with ux_utils.print_exception_no_traceback():
                 raise exceptions.ClusterStatusFetchingError(
@@ -587,9 +626,9 @@ class OCI(clouds.Cloud):
 
         for node in vms:
             vm_status = node.lifecycle_state
-            if vm_status in status_map:
-                sky_status = status_map[vm_status]
-                if sky_status is not None:
-                    status_list.append(sky_status)
+            sky_status = oci_utils.oci_config.STATE_MAPPING_OCI_TO_SKY.get(
+                vm_status, None)
+            if sky_status is not None:
+                status_list.append(sky_status)
 
         return status_list

sky/clouds/paperspace.py

@@ -1,14 +1,14 @@
 """ Paperspace Cloud. """
 
-import json
 import typing
-from typing import Dict, Iterator, List, Optional, Tuple
+from typing import Dict, Iterator, List, Optional, Tuple, Union
 
 import requests
 
 from sky import clouds
 from sky.clouds import service_catalog
 from sky.provision.paperspace import utils
+from sky.utils import registry
 from sky.utils import resources_utils
 
 if typing.TYPE_CHECKING:
@@ -20,7 +20,7 @@ _CREDENTIAL_FILES = [
 ]
 
 
-@clouds.CLOUD_REGISTRY.register
+@registry.CLOUD_REGISTRY.register
 class Paperspace(clouds.Cloud):
     """Paperspace GPU Cloud"""
 
@@ -162,7 +162,7 @@ class Paperspace(clouds.Cloud):
 
     @classmethod
     def get_accelerators_from_instance_type(
-            cls, instance_type: str) -> Optional[Dict[str, int]]:
+            cls, instance_type: str) -> Optional[Dict[str, Union[int, float]]]:
         return service_catalog.get_accelerators_from_instance_type(
             instance_type, clouds='paperspace')
 
@@ -173,18 +173,17 @@ class Paperspace(clouds.Cloud):
     def make_deploy_resources_variables(
             self,
             resources: 'resources_lib.Resources',
-            cluster_name_on_cloud: str,
+            cluster_name: resources_utils.ClusterName,
             region: 'clouds.Region',
             zones: Optional[List['clouds.Zone']],
+            num_nodes: int,
             dryrun: bool = False) -> Dict[str, Optional[str]]:
-        del zones, dryrun
+        del zones, dryrun, cluster_name
 
         r = resources
         acc_dict = self.get_accelerators_from_instance_type(r.instance_type)
-        if acc_dict is not None:
-            custom_resources = json.dumps(acc_dict, separators=(',', ':'))
-        else:
-            custom_resources = None
+        custom_resources = resources_utils.make_ray_custom_resources_str(
+            acc_dict)
 
         return {
             'instance_type': resources.instance_type,
@@ -196,11 +195,13 @@ class Paperspace(clouds.Cloud):
             self, resources: 'resources_lib.Resources'):
         """Returns a list of feasible resources for the given resources."""
         if resources.use_spot:
-            return ([], [])
+            # TODO: Add hints to all return values in this method to help
+            #  users understand why the resources are not launchable.
+            return resources_utils.FeasibleResources([], [], None)
         if resources.instance_type is not None:
             assert resources.is_launchable(), resources
             resources = resources.copy(accelerators=None)
-            return ([resources], [])
+            return resources_utils.FeasibleResources([resources], [], None)
 
         def _make(instance_list):
             resource_list = []
@@ -223,9 +224,10 @@ class Paperspace(clouds.Cloud):
                 memory=resources.memory,
                 disk_tier=resources.disk_tier)
             if default_instance_type is None:
-                return ([], [])
+                return resources_utils.FeasibleResources([], [], None)
             else:
-                return (_make([default_instance_type]), [])
+                return resources_utils.FeasibleResources(
+                    _make([default_instance_type]), [], None)
 
         assert len(accelerators) == 1, resources
         acc, acc_count = list(accelerators.items())[0]
@@ -241,8 +243,10 @@ class Paperspace(clouds.Cloud):
                 clouds='paperspace',
             ))
         if instance_list is None:
-            return ([], fuzzy_candidate_list)
-        return (_make(instance_list), fuzzy_candidate_list)
+            return resources_utils.FeasibleResources([], fuzzy_candidate_list,
+                                                     None)
+        return resources_utils.FeasibleResources(_make(instance_list),
+                                                 fuzzy_candidate_list, None)
 
     @classmethod
     def check_credentials(cls) -> Tuple[bool, Optional[str]]:
@@ -255,7 +259,7 @@ class Paperspace(clouds.Cloud):
             return False, (
                 'Failed to access Paperspace Cloud with credentials.\n    '
                 'To configure credentials, follow the instructions at: '
-                'https://skypilot.readthedocs.io/en/latest/getting-started/installation.html#paperspace\n    '
+                'https://docs.skypilot.co/en/latest/getting-started/installation.html#paperspace\n    '
                 'Generate API key and create a json at `~/.paperspace/config.json` with \n     '
                 '    {"apiKey": "[YOUR API KEY]"}\n    '
                 f'Reason: {str(e)}')
@@ -275,7 +279,7 @@ class Paperspace(clouds.Cloud):
         }
 
     @classmethod
-    def get_current_user_identity(cls) -> Optional[List[str]]:
+    def get_user_identities(cls) -> Optional[List[List[str]]]:
         # NOTE: used for very advanced SkyPilot functionality
         # Can implement later if desired
         return None