skypilot-nightly 1.0.0.dev20241109__py3-none-any.whl → 1.0.0.dev20241110__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 = '42c79e1d0a5e018e275705ada53957573f9a0181'
+_SKYPILOT_COMMIT_SHA = 'dddd65187953a5d6b32f762bea78eed1f109ec3c'
 
 
 def _get_git_commit():
@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0.dev20241109'
+__version__ = '1.0.0.dev20241110'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

sky/backends/cloud_vm_ray_backend.py

@@ -3979,25 +3979,6 @@ class CloudVmRayBackend(backends.Backend['CloudVmRayResourceHandle']):
                 stdout = ''
                 stderr = str(e)
 
-        # Apr, 2023 by Hysun(hysun.he@oracle.com): Added support for OCI
-        # May, 2023 by Hysun: Allow terminate INIT cluster which may have
-        # some instances provisioning in background but not completed.
-        elif (isinstance(cloud, clouds.OCI) and terminate and
-              prev_cluster_status in (status_lib.ClusterStatus.STOPPED,
-                                      status_lib.ClusterStatus.INIT)):
-            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
-
-            # 0: All terminated successfully, failed count otherwise
-            returncode = oci_query_helper.terminate_instances_by_tags(
-                {TAG_RAY_CLUSTER_NAME: cluster_name_on_cloud}, region)
-
-            # To avoid undefined local variables error.
-            stdout = stderr = ''
         else:
             config['provider']['cache_stopped_nodes'] = not terminate
             with tempfile.NamedTemporaryFile('w',

sky/clouds/oci.py

@@ -31,6 +31,7 @@ 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 resources_utils
 from sky.utils import ux_utils
@@ -60,6 +61,9 @@ 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'
@@ -433,7 +437,7 @@ 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'
@@ -597,25 +601,11 @@ 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(
@@ -625,9 +615,9 @@ 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/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 @@ History:
  - 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 +78,19 @@ class OCIConfig:
         resources_utils.DiskTier.HIGH: DISK_TIER_HIGH,
     }
 
+    # Oracle instance's lifecycle state to sky state mapping.
+    # For Oracle VM instance's lifecyle state, please refer to the link:
+    # https://docs.oracle.com/en-us/iaas/api/#/en/iaas/latest/Instance/
+    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/dag.py

@@ -56,21 +56,25 @@ class Dag:
         return self.graph
 
     def is_chain(self) -> bool:
-        # NOTE: this method assumes that the graph has no cycle.
-        is_chain = True
-        visited_zero_out_degree = False
-        for node in self.graph.nodes:
-            out_degree = self.graph.out_degree(node)
-            if out_degree > 1:
-                is_chain = False
-                break
-            elif out_degree == 0:
-                if visited_zero_out_degree:
-                    is_chain = False
-                    break
-                else:
-                    visited_zero_out_degree = True
-        return is_chain
+        """Check if the DAG is a linear chain of tasks."""
+
+        nodes = list(self.graph.nodes)
+
+        if len(nodes) == 0:
+            return True
+
+        in_degrees = [self.graph.in_degree(node) for node in nodes]
+        out_degrees = [self.graph.out_degree(node) for node in nodes]
+
+        # Check out-degrees: all <= 1 and exactly one node has out_degree == 0
+        out_degree_condition = (all(degree <= 1 for degree in out_degrees) and
+                                sum(degree == 0 for degree in out_degrees) == 1)
+
+        # Check in-degrees: all <= 1 and exactly one node has in_degree == 0
+        in_degree_condition = (all(degree <= 1 for degree in in_degrees) and
+                               sum(degree == 0 for degree in in_degrees) == 1)
+
+        return out_degree_condition and in_degree_condition
 
 
 class _DagContext(threading.local):

sky/provision/__init__.py

@@ -20,6 +20,7 @@ from sky.provision import fluidstack
 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 +1,51 @@
+"""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 exceptions
+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)
+
+    # Find the configured VCN, or create a new one.
+    vcn = query_helper.find_create_vcn_subnet(region)
+    if vcn is None:
+        # pylint: disable=line-too-long
+        raise exceptions.ResourcesUnavailableError(
+            'Failed to create a new VCN, possibly you hit the resource limitation.'
+        )
+
+    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