skypilot-nightly 1.0.0.dev20250509__py3-none-any.whl → 1.0.0.dev20251107__py3-none-any.whl

sky/clouds/seeweb.py

@@ -0,0 +1,466 @@
+"""Seeweb Cloud
+
+History:
+@ Aug 6, 2025: Initial version of the integration.
+- Francesco Massa
+- Marco Cristofanilli (marco.cATseeweb.it)
+
+"""
+
+from __future__ import annotations
+
+import typing
+from typing import Any, Dict, Iterator, List, Optional, Tuple, Union
+
+from sky import catalog
+from sky import clouds
+from sky.adaptors import seeweb as seeweb_adaptor
+from sky.provision import seeweb as seeweb_provision
+from sky.utils import registry
+from sky.utils import resources_utils
+from sky.utils import ux_utils
+
+if typing.TYPE_CHECKING:
+    from sky import resources as resources_lib
+    from sky.utils import status_lib
+    from sky.utils import volume as volume_lib
+
+# ---------- key file path -----------------
+_SEEWEB_KEY_FILE = '~/.seeweb_cloud/seeweb_keys'
+# (content: ini-like)
+#   api_key = <TOKEN>
+
+
+@registry.CLOUD_REGISTRY.register
+class Seeweb(clouds.Cloud):
+    """Seeweb GPU Cloud."""
+
+    _REPR = 'Seeweb'
+    # Define unsupported features to provide clear error messages
+    # This helps users understand what Seeweb can and cannot do
+    _CLOUD_UNSUPPORTED_FEATURES = {
+        clouds.CloudImplementationFeatures.MULTI_NODE:
+            ('Multi-node not supported. '
+             'Seeweb does not support multi-node clusters.'),
+        clouds.CloudImplementationFeatures.CUSTOM_DISK_TIER:
+            ('Custom disk tiers not supported. '
+             'Seeweb does not support custom disk tiers.'),
+        clouds.CloudImplementationFeatures.STORAGE_MOUNTING:
+            ('Storage mounting not supported. '
+             'Seeweb does not support storage mounting.'),
+        clouds.CloudImplementationFeatures.HIGH_AVAILABILITY_CONTROLLERS:
+            ('High availability controllers not supported. '
+             'Seeweb does not support high availability controllers.'),
+        clouds.CloudImplementationFeatures.SPOT_INSTANCE:
+            ('Spot instances not supported. '
+             'Seeweb does not support spot instances.'),
+        clouds.CloudImplementationFeatures.CLONE_DISK_FROM_CLUSTER:
+            ('Disk cloning not supported. '
+             'Seeweb does not support disk cloning.'),
+        clouds.CloudImplementationFeatures.DOCKER_IMAGE:
+            ('Docker images not supported. '
+             'Seeweb does not support Docker images.'),
+        clouds.CloudImplementationFeatures.IMAGE_ID:
+            ('Custom image IDs not supported. '
+             'Seeweb does not support custom image IDs.'),
+        clouds.CloudImplementationFeatures.CUSTOM_NETWORK_TIER:
+            ('Custom network tiers not supported. '
+             'Seeweb does not support custom network tiers.'),
+        clouds.CloudImplementationFeatures.HOST_CONTROLLERS:
+            ('Host controllers not supported. '
+             'Seeweb does not support host controllers.'),
+        clouds.CloudImplementationFeatures.CUSTOM_MULTI_NETWORK:
+            ('Custom multi-network not supported. '
+             'Seeweb does not support custom multi-network.'),
+    }
+    _MAX_CLUSTER_NAME_LEN_LIMIT = 120
+    _regions: List[clouds.Region] = []
+
+    PROVISIONER_VERSION = clouds.ProvisionerVersion.SKYPILOT
+    STATUS_VERSION = clouds.StatusVersion.SKYPILOT
+
+    # Enable port support with updatable version
+    OPEN_PORTS_VERSION = clouds.OpenPortsVersion.UPDATABLE
+
+    @classmethod
+    def _unsupported_features_for_resources(
+        cls,
+        resources: 'resources_lib.Resources',
+        region: Optional[str] = None,
+    ) -> Dict[clouds.CloudImplementationFeatures, str]:
+        return cls._CLOUD_UNSUPPORTED_FEATURES
+
+    @classmethod
+    def max_cluster_name_length(cls) -> Optional[int]:
+        return cls._MAX_CLUSTER_NAME_LEN_LIMIT
+
+    @classmethod
+    def regions(cls) -> List['clouds.Region']:
+        """Return available regions for Seeweb."""
+        # Get regions from the catalog system
+        # This reads from the CSV files generated by fetch_seeweb.py
+        regions = catalog.regions(clouds='seeweb')
+        return regions
+
+    @classmethod
+    def regions_with_offering(
+        cls,
+        instance_type: str,
+        accelerators: Optional[Dict[str, int]],
+        use_spot: bool,
+        region: Optional[str],
+        zone: Optional[str],
+        resources: Optional['resources_lib.Resources'] = None,
+    ) -> List[clouds.Region]:
+        assert zone is None, 'Seeweb does not support zones.'
+        del zone
+        if use_spot:
+            return []
+
+        # Get regions from catalog based on instance type
+        # This will read the CSV and return only regions
+        # where the instance type exists
+        regions = catalog.get_region_zones_for_instance_type(
+            instance_type, use_spot, 'seeweb')
+
+        if region is not None:
+            regions = [r for r in regions if r.name == region]
+
+        return regions
+
+    @classmethod
+    def zones_provision_loop(
+        cls,
+        *,
+        region: str,
+        num_nodes: int,
+        instance_type: str,
+        accelerators: Optional[Dict[str, int]] = None,
+        use_spot: bool = False,
+    ) -> Iterator[None]:
+        del num_nodes
+        regions = cls.regions_with_offering(instance_type,
+                                            accelerators,
+                                            use_spot,
+                                            region=region,
+                                            zone=None)
+        for r in regions:
+            assert r.zones is None, r
+            yield r.zones
+
+    @classmethod
+    def get_zone_shell_cmd(cls) -> Optional[str]:
+        """Seeweb doesn't support zones."""
+        return None
+
+    def instance_type_to_hourly_cost(
+        self,
+        instance_type: str,
+        use_spot: bool,
+        region: Optional[str],
+        zone: Optional[str],
+    ) -> float:
+        cost = catalog.get_hourly_cost(instance_type,
+                                       use_spot=use_spot,
+                                       region=region,
+                                       zone=zone,
+                                       clouds='seeweb')
+        return cost
+
+    def accelerators_to_hourly_cost(
+        self,
+        accelerators: Dict[str, int],
+        use_spot: bool,
+        region: Optional[str],
+        zone: Optional[str],
+    ) -> float:
+
+        return 0.0
+
+    def get_egress_cost(self, num_gigabytes: float):
+        return 0.0
+
+    def make_deploy_resources_variables(
+        self,
+        resources: 'resources_lib.Resources',
+        cluster_name: resources_utils.ClusterName,
+        region: 'clouds.Region',
+        zones: Optional[List['clouds.Zone']],
+        num_nodes: int,
+        dryrun: bool = False,
+        volume_mounts: Optional[List['volume_lib.VolumeMount']] = None,
+    ) -> Dict[str, Any]:
+        """Create deployment variables for Seeweb."""
+
+        # Note: Spot instances and multi-node are automatically handled by
+        # the framework via _CLOUD_UNSUPPORTED_FEATURES
+
+        resources = resources.assert_launchable()
+
+        acc_dict = self.get_accelerators_from_instance_type(
+            resources.instance_type)
+
+        # Standard custom_resources string for Ray
+        custom_resources = resources_utils.make_ray_custom_resources_str(
+            acc_dict)
+
+        # Seeweb-specific GPU configuration for the provisioner
+        # This tells the provisioner how to configure GPU resources
+        seeweb_gpu_config = None
+        if resources.accelerators:
+            # If the instance has accelerators, prepare GPU configuration
+            accelerator_name = list(resources.accelerators.keys())[0]
+            accelerator_count = resources.accelerators[accelerator_name]
+            seeweb_gpu_config = {
+                'gpu': accelerator_count,
+                'gpu_label': accelerator_name,
+            }
+
+        # Seeweb uses pre-configured images based on instance type
+        # Determine image based on whether the instance type name contains "GPU"
+        if resources.instance_type and 'GPU' in resources.instance_type.upper():
+            # GPU instance - use image with NVIDIA drivers
+            if resources.instance_type in ['ECS1GPU10', 'ECS2GPU10']:
+                # H200 GPU instance - use UEFI image with NVIDIA drivers
+                image_id = 'ubuntu-2204-uefi-nvidia-driver'
+            else:
+                # Other GPU instance - use standard image with NVIDIA drivers
+                image_id = 'ubuntu-2204-nvidia-driver'
+        else:
+            # CPU-only instance - use standard Ubuntu image
+            image_id = 'ubuntu-2204'
+
+        result = {
+            'instance_type': resources.instance_type,
+            'region': region.name,
+            'cluster_name': cluster_name,
+            'custom_resources': custom_resources,
+            'seeweb_gpu_config': seeweb_gpu_config,
+            'image_id': image_id,
+        }
+        return result
+
+    @classmethod
+    def get_vcpus_mem_from_instance_type(
+            cls, instance_type: str) -> Tuple[Optional[float], Optional[float]]:
+        result = catalog.get_vcpus_mem_from_instance_type(instance_type,
+                                                          clouds='seeweb')
+        return result
+
+    @classmethod
+    def get_accelerators_from_instance_type(
+        cls,
+        instance_type: str,
+    ) -> Optional[Dict[str, Union[int, float]]]:
+        result = catalog.get_accelerators_from_instance_type(instance_type,
+                                                             clouds='seeweb')
+        return result
+
+    @classmethod
+    def get_default_instance_type(
+        cls,
+        cpus: Optional[str] = None,
+        memory: Optional[str] = None,
+        disk_tier: Optional[resources_utils.DiskTier] = None,
+        region: Optional[str] = None,
+        zone: Optional[str] = None,
+    ) -> Optional[str]:
+        result = catalog.get_default_instance_type(cpus=cpus,
+                                                   memory=memory,
+                                                   disk_tier=disk_tier,
+                                                   clouds='seeweb')
+        return result
+
+    def _get_feasible_launchable_resources(
+        self, resources: 'resources_lib.Resources'
+    ) -> 'resources_utils.FeasibleResources':
+        """Get feasible resources for Seeweb."""
+        if resources.use_spot:
+            return resources_utils.FeasibleResources(
+                [], [], 'Spot instances not supported on Seeweb')
+
+        if resources.accelerators and len(resources.accelerators) > 1:
+            return resources_utils.FeasibleResources(
+                [], [], 'Multiple accelerator types not supported on Seeweb')
+
+        # If no instance_type is specified, try to get a default one
+        if not resources.instance_type:
+            # If accelerators are specified, try to find instance
+            # type forthat accelerator
+            if resources.accelerators:
+                # Get the first accelerator
+                # (we already checked there's only one)
+                acc_name, acc_count = list(resources.accelerators.items())[0]
+
+                # Use catalog to find instance type for this accelerator
+                # This leverages the catalog system to find suitable instances
+                (
+                    instance_types,
+                    fuzzy_candidates,
+                ) = catalog.get_instance_type_for_accelerator(
+                    acc_name=acc_name,
+                    acc_count=acc_count,
+                    cpus=resources.cpus,
+                    memory=resources.memory,
+                    use_spot=resources.use_spot,
+                    region=resources.region,
+                    zone=resources.zone,
+                    clouds='seeweb',
+                )
+
+                if instance_types and len(instance_types) > 0:
+                    # Use the first (cheapest) instance type
+                    selected_instance_type = instance_types[0]
+                    resources = resources.copy(
+                        instance_type=selected_instance_type)
+                else:
+                    return resources_utils.FeasibleResources(
+                        [],
+                        fuzzy_candidates,
+                        f'No instance type found for accelerator'
+                        f'{acc_name}:{acc_count} on Seeweb',
+                    )
+            else:
+                # No accelerators specified, use default instance type
+                default_instance_type = self.get_default_instance_type(
+                    cpus=resources.cpus,
+                    memory=resources.memory,
+                    region=resources.region,
+                    zone=resources.zone,
+                )
+
+                if default_instance_type:
+                    # Create new resources with the default instance type
+                    resources = resources.copy(
+                        instance_type=default_instance_type)
+                else:
+                    return resources_utils.FeasibleResources(
+                        [],
+                        [],
+                        f'No suitable instance type found for'
+                        f'cpus={resources.cpus}, memory={resources.memory}',
+                    )
+
+        # Check if instance type exists
+        if resources.instance_type:
+            exists = catalog.instance_type_exists(resources.instance_type,
+                                                  clouds='seeweb')
+            if not exists:
+                return resources_utils.FeasibleResources(
+                    [],
+                    [],
+                    f'Instance type {resources.instance_type}'
+                    f' not available on Seeweb',
+                )
+
+        # Set the cloud if not already set
+        if not resources.cloud:
+            resources = resources.copy(cloud=self)
+
+        # Return the resources as feasible
+        return resources_utils.FeasibleResources([resources], [], None)
+
+    @classmethod
+    def _check_compute_credentials(cls) -> Tuple[bool, Optional[str]]:
+        """Check Seeweb compute credentials."""
+        try:
+            result = seeweb_adaptor.check_compute_credentials()
+            return result, None
+        except Exception as e:  # pylint: disable=broad-except
+            return False, str(e)
+
+    @classmethod
+    def _check_storage_credentials(cls) -> Tuple[bool, Optional[str]]:
+        """Check Seeweb storage credentials."""
+        try:
+            result = seeweb_adaptor.check_storage_credentials()
+            return result, None
+        except Exception as e:  # pylint: disable=broad-except
+            return False, str(e)
+
+    @classmethod
+    def get_user_identities(cls) -> Optional[List[List[str]]]:
+        # Seeweb doesn't have user identity concept
+        return None
+
+    @classmethod
+    def query_status(
+        cls,
+        name: str,
+        tag_filters: Dict[str, str],
+        region: Optional[str],
+        zone: Optional[str],
+        **kwargs,
+    ) -> List['status_lib.ClusterStatus']:
+        """Query the status of Seeweb cluster instances."""
+        cluster_name_on_cloud = name
+
+        result = seeweb_provision.instance.query_instances(
+            cluster_name=name,
+            cluster_name_on_cloud=cluster_name_on_cloud,
+            provider_config={},
+            non_terminated_only=True)
+        # Convert Dict[str, Tuple[Optional[ClusterStatus],
+        # Optional[str]]] to List[ClusterStatus]
+        return [status for status, _ in result.values() if status is not None]
+
+    def get_credential_file_mounts(self) -> Dict[str, str]:
+        """Returns the credential files to mount."""
+        # Mount the Seeweb API key file to the remote instance
+        # This allows the provisioner to authenticate with Seeweb API
+        result = {
+            _SEEWEB_KEY_FILE: _SEEWEB_KEY_FILE,
+        }
+        return result
+
+    def instance_type_exists(self, instance_type: str) -> bool:
+        """Returns whether the instance type exists for Seeweb."""
+        result = catalog.instance_type_exists(instance_type, clouds='seeweb')
+        return result
+
+    @classmethod
+    def get_image_size(cls, image_id: str, region: Optional[str]) -> float:
+        """Seeweb doesn't support custom images."""
+        del image_id, region
+        with ux_utils.print_exception_no_traceback():
+            raise ValueError(f'Custom images are not supported on {cls._REPR}. '
+                             'Seeweb clusters use pre-configured images only.')
+
+    # Image-related methods (not supported)
+    @classmethod
+    def create_image_from_cluster(
+        cls,
+        cluster_name: resources_utils.ClusterName,
+        region: Optional[str],
+        zone: Optional[str],
+    ) -> str:
+        del cluster_name, region, zone  # unused
+        with ux_utils.print_exception_no_traceback():
+            raise ValueError(
+                f'Creating images from clusters is not supported on'
+                f' {cls._REPR}. Seeweb does not support custom'
+                f' image creation.')
+
+    @classmethod
+    def maybe_move_image(
+        cls,
+        image_id: str,
+        source_region: str,
+        target_region: str,
+        source_zone: Optional[str],
+        target_zone: Optional[str],
+    ) -> str:
+        del image_id, source_region, target_region, source_zone, target_zone
+        with ux_utils.print_exception_no_traceback():
+            raise ValueError(
+                f'Moving images between regions is not supported on'
+                f' {cls._REPR}. '
+                'Seeweb does not support custom images.')
+
+    @classmethod
+    def delete_image(cls, image_id: str, region: Optional[str]) -> None:
+        del image_id, region
+        with ux_utils.print_exception_no_traceback():
+            raise ValueError(
+                f'Deleting images is not supported on {cls._REPR}. '
+                'Seeweb does not support custom image management.')