[OCI] Re-Implementation with the new provision API framework.

sky/backends/cloud_vm_ray_backend.py

@@ -4003,12 +4003,11 @@ def teardown_no_lock(self,
             region = config['provider']['region']
 
             # pylint: disable=import-outside-toplevel
-            from ray.autoscaler.tags import TAG_RAY_CLUSTER_NAME
-
-            from sky.skylet.providers.oci.query_helper import oci_query_helper
+            from sky.provision.constants import TAG_RAY_CLUSTER_NAME
+            from sky.provision.oci.query_utils import query_helper
 
             # 0: All terminated successfully, failed count otherwise
-            returncode = oci_query_helper.terminate_instances_by_tags(
+            returncode = query_helper.terminate_instances_by_tags(
                 {TAG_RAY_CLUSTER_NAME: cluster_name_on_cloud}, region)
 
             # To avoid undefined local variables error.

sky/clouds/oci.py

@@ -32,6 +32,7 @@
 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 resources_utils
 from sky.utils import ux_utils
@@ -61,6 +62,9 @@ class OCI(clouds.Cloud):
                              {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(
         cls, resources: 'resources_lib.Resources'
@@ -436,7 +440,7 @@ def check_credentials(cls) -> Tuple[bool, Optional[str]]:
             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'
@@ -596,32 +600,19 @@ def query_status(cls, name: str, tag_filters: Dict[str, str],
                      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(
                     f'Failed to query OCI cluster {name!r} status. '
                     'Details: '
                     f'{common_utils.format_exception(e, use_bracket=True)}')
 
+        status_map = oci_utils.oci_config.STATE_MAPPING_OCI_TO_SKY
         for node in vms:
             vm_status = node.lifecycle_state
             if vm_status in status_map:

sky/clouds/service_catalog/oci_catalog.py

@@ -66,7 +66,7 @@ def _get_df() -> 'pd.DataFrame':
             logger.debug(f'It is OK goes here when testing: {str(e)}')
             subscribed_regions = []
 
-        except oci_adaptor.service_exception() as e:
+        except oci_adaptor.oci.exceptions.ServiceError as e:
             # Should never expect going here. However, we still catch
             # it so that if any OCI call failed, the program can still
             # proceed with try-and-error way.

sky/clouds/utils/oci_utils.py

@@ -5,13 +5,14 @@
  - Hysun He (hysun.he@oracle.com) @ Oct, 2024: Add default image OS
    configuration.
 """
-import logging
 import os
 
+from sky import sky_logging
 from sky import skypilot_config
+from sky import status_lib
 from sky.utils import resources_utils
 
-logger = logging.getLogger(__name__)
+logger = sky_logging.init_logger(__name__)
 
 
 class OCIConfig:
@@ -77,6 +78,16 @@ class OCIConfig:
         resources_utils.DiskTier.HIGH: DISK_TIER_HIGH,
     }
 
+    STATE_MAPPING_OCI_TO_SKY = {
+        '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,
+    }
+
     @classmethod
     def get_compartment(cls, region):
         # Allow task(cluster)-specific compartment/VCN parameters.

sky/provision/__init__.py

@@ -20,6 +20,7 @@
 from sky.provision import gcp
 from sky.provision import kubernetes
 from sky.provision import lambda_cloud
+from sky.provision import oci
 from sky.provision import runpod
 from sky.provision import vsphere
 from sky.utils import command_runner

sky/provision/oci/__init__.py

@@ -0,0 +1,15 @@
+"""OCI provisioner for SkyPilot.
+
+History:
+ - Hysun He (hysun.he@oracle.com) @ Oct.16, 2024: Initial implementation
+"""
+
+from sky.provision.oci.config import bootstrap_instances
+from sky.provision.oci.instance import cleanup_ports
+from sky.provision.oci.instance import get_cluster_info
+from sky.provision.oci.instance import open_ports
+from sky.provision.oci.instance import query_instances
+from sky.provision.oci.instance import run_instances
+from sky.provision.oci.instance import stop_instances
+from sky.provision.oci.instance import terminate_instances
+from sky.provision.oci.instance import wait_instances

sky/provision/oci/config.py

@@ -0,0 +1,47 @@
+"""OCI configuration bootstrapping.
+
+Creates the resource group and deploys the configuration template to OCI for
+a cluster to be launched.
+
+History:
+ - Hysun He (hysun.he@oracle.com) @ Oct.16, 2024: Initial implementation
+"""
+
+from sky import sky_logging
+from sky.adaptors import oci as oci_adaptor
+from sky.clouds.utils import oci_utils
+from sky.provision import common
+from sky.provision.oci.query_utils import query_helper
+
+logger = sky_logging.init_logger(__name__)
+
+
+@common.log_function_start_end
+def bootstrap_instances(
+        region: str, cluster_name_on_cloud: str,
+        config: common.ProvisionConfig) -> common.ProvisionConfig:
+    """See sky/provision/__init__.py"""
+    # OCI module import and oci client
+    oci_adaptor.get_core_client(region, oci_utils.oci_config.get_profile())
+
+    # Find / create a compartment for creating instances.
+    compartment = query_helper.find_compartment(region)
+    assert compartment is not None
+
+    # Find the configured VCN, or create a new one.
+    vcn = query_helper.find_create_vcn_subnet(region)
+    assert vcn is not None
+
+    node_config = config.node_config
+
+    # Subscribe the image if it is from Marketplace listing.
+    query_helper.subscribe_image(
+        compartment_id=compartment,
+        listing_id=node_config['AppCatalogListingId'],
+        resource_version=node_config['ResourceVersion'],
+        region=region,
+    )
+
+    logger.info(f'Using cluster name: {cluster_name_on_cloud}')
+
+    return config