skypilot-nightly 1.0.0.dev20250521__py3-none-any.whl → 1.0.0.dev20250523__py3-none-any.whl

sky/utils/log_utils.py

@@ -190,7 +190,7 @@ class SkyLocalUpLineProcessor(LineProcessor):
 
 
 class SkyRemoteUpLineProcessor(LineProcessor):
-    """A processor for deploy_remote_cluster.sh log lines."""
+    """A processor for deploy_remote_cluster.py log lines."""
 
     def __init__(self, log_path: str, is_local: bool):
         self.log_path = log_path
@@ -291,6 +291,219 @@ class SkyRemoteUpLineProcessor(LineProcessor):
         self.status_display.stop()
 
 
+class SkySSHUpLineProcessor(LineProcessor):
+    """A processor for deploy_remote_cluster.py log lines for SSH clusters"""
+
+    def __init__(self, log_path: str, is_local: bool):
+        self.log_path = log_path
+        self.is_local = is_local
+        self.current_cluster: Optional[str] = None
+        self.is_cleanup_mode = False
+
+    def __enter__(self) -> None:
+        status = rich_utils.safe_status(
+            ux_utils.spinner_message('Preparing to set up SSH Node Pools',
+                                     log_path=self.log_path,
+                                     is_local=self.is_local))
+        self.status_display = status
+        self.status_display.start()
+
+    def process_line(self, log_line: str) -> None:
+        # Detect cleanup mode
+        if 'SKYPILOT_CLEANUP_MODE:' in log_line:
+            self.is_cleanup_mode = True
+            if self.current_cluster:
+                self.status_display.update(
+                    ux_utils.spinner_message(
+                        f'Cleaning up Node Pool: \\[{self.current_cluster}]',
+                        log_path=self.log_path,
+                        is_local=self.is_local))
+
+        # Cluster detection message
+        if 'SKYPILOT_CLUSTER_INFO:' in log_line:
+            clusters_part = log_line.split('SKYPILOT_CLUSTER_INFO:',
+                                           1)[1].strip()
+            if clusters_part.startswith('Found'):
+                logger.info(f'{colorama.Style.RESET_ALL}'
+                            f'{colorama.Fore.CYAN}{clusters_part}'
+                            f'{colorama.Style.RESET_ALL}')
+
+        # Current cluster being operated on
+        if 'SKYPILOT_CURRENT_CLUSTER:' in log_line:
+            self.current_cluster = log_line.split('SKYPILOT_CURRENT_CLUSTER:',
+                                                  1)[1].strip()
+
+            if self.is_cleanup_mode:
+                self.status_display.update(
+                    ux_utils.spinner_message(
+                        f'Cleaning up Node Pool: {self.current_cluster}',
+                        log_path=self.log_path,
+                        is_local=self.is_local))
+                logger.info(f'{colorama.Fore.CYAN}\nCleaning up Node Pool: '
+                            f'{self.current_cluster}{colorama.Style.RESET_ALL}')
+            else:
+                self.status_display.update(
+                    ux_utils.spinner_message(
+                        f'Deploying SkyPilot \\[{self.current_cluster}]',
+                        log_path=self.log_path,
+                        is_local=self.is_local))
+                logger.info(f'{colorama.Style.RESET_ALL}'
+                            f'{colorama.Fore.CYAN}\nSetting up Node Pool: '
+                            f'{self.current_cluster}{colorama.Style.RESET_ALL}')
+
+        # Handle cluster completion marker
+        if 'SKYPILOT_CLUSTER_COMPLETED:' in log_line:
+            if self.is_cleanup_mode:
+                logger.info(
+                    f'{ux_utils.INDENT_LAST_SYMBOL}{colorama.Fore.GREEN}'
+                    f'✔ Node Pool {self.current_cluster} cleaned up '
+                    f'successfully.{colorama.Style.RESET_ALL}')
+            else:
+                logger.info(
+                    f'{ux_utils.INDENT_LAST_SYMBOL}{colorama.Fore.GREEN}'
+                    f'✔ Node Pool {self.current_cluster} deployed successfully.'
+                    f'{colorama.Style.RESET_ALL}')
+
+        # Pre-flight checks
+        if 'Checking SSH connection to head node' in log_line:
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Style.DIM}'
+                        'Checking SSH connection to head node...'
+                        f'{colorama.Style.RESET_ALL}')
+
+        if log_line.startswith('SSH connection successful'):
+            node_name = log_line.split('(')[-1].split(')')[0]
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.GREEN}'
+                        '✔ SSH connection established to head node '
+                        f'{node_name}.{colorama.Style.RESET_ALL}')
+
+        # Kubernetes installation steps
+        if 'Deploying Kubernetes on head node' in log_line:
+            current_cluster_str = f' \\[{self.current_cluster}]' if (
+                self.current_cluster) else ''
+            self.status_display.update(
+                ux_utils.spinner_message(
+                    'Deploying SkyPilot runtime on head node'
+                    f'{current_cluster_str}',
+                    log_path=self.log_path,
+                    is_local=self.is_local))
+
+        if 'K3s deployed on head node' in log_line:
+            node_name = log_line.split('(')[-1].split(')')[0]
+            logger.info(
+                f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.GREEN}'
+                f'✔ SkyPilot runtime successfully deployed on head node '
+                f'{node_name}.{colorama.Style.RESET_ALL}')
+
+        # Worker nodes
+        if 'Deploying Kubernetes on worker node' in log_line:
+            self.status_display.update(
+                ux_utils.spinner_message(
+                    'Deploying SkyPilot runtime on worker nodes' +
+                    (f' \\[{self.current_cluster}]'
+                     if self.current_cluster else ''),
+                    log_path=self.log_path,
+                    is_local=self.is_local))
+
+        if 'Kubernetes deployed on worker node' in log_line:
+            node_name = log_line.split('(')[-1].split(')')[0]
+            logger.info(
+                f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.GREEN}'
+                '✔ SkyPilot runtime successfully deployed on worker node '
+                f'{node_name}.{colorama.Style.RESET_ALL}')
+
+        if 'Failed to deploy K3s on worker node' in log_line:
+            node_name = log_line.split('(')[-1].split(')')[0]
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.RED}'
+                        '✗ Failed to deploy K3s on worker node '
+                        f'{node_name}.{colorama.Style.RESET_ALL}')
+
+        # Cluster configuration
+        if 'Configuring local kubectl to connect to the cluster...' in log_line:
+            self.status_display.update(
+                ux_utils.spinner_message('Setting up SkyPilot configuration' +
+                                         (f' \\[{self.current_cluster}]'
+                                          if self.current_cluster else ''),
+                                         log_path=self.log_path,
+                                         is_local=self.is_local))
+
+        if 'kubectl configured to connect to the cluster.' in log_line:
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.GREEN}'
+                        '✔ SkyPilot configuration complete.'
+                        f'{colorama.Style.RESET_ALL}')
+
+        # GPU operator installation
+        if 'Installing Nvidia GPU Operator...' in log_line:
+            self.status_display.update(
+                ux_utils.spinner_message('Configuring Nvidia GPUs' +
+                                         (f' \\[{self.current_cluster}]'
+                                          if self.current_cluster else ''),
+                                         log_path=self.log_path,
+                                         is_local=self.is_local))
+
+        if 'GPU Operator installed.' in log_line:
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.GREEN}'
+                        '✔ Nvidia GPUs configured successfully.'
+                        f'{colorama.Style.RESET_ALL}')
+
+        # Cleanup steps
+        if 'Cleaning up head node' in log_line:
+            self.status_display.update(
+                ux_utils.spinner_message('Cleaning up head node' +
+                                         (f' \\[{self.current_cluster}]'
+                                          if self.current_cluster else ''),
+                                         log_path=self.log_path,
+                                         is_local=self.is_local))
+
+        if 'Cleaning up worker node' in log_line:
+            self.status_display.update(
+                ux_utils.spinner_message('Cleaning up worker nodes' +
+                                         (f' \\[{self.current_cluster}]'
+                                          if self.current_cluster else ''),
+                                         log_path=self.log_path,
+                                         is_local=self.is_local))
+
+        # Handle node cleanup success messages
+        if 'Node' in log_line and 'cleaned up successfully' in log_line:
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.GREEN}'
+                        f'{log_line.strip()}{colorama.Style.RESET_ALL}')
+
+        if 'Node' in log_line and 'Failed to clean up' in log_line:
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.RED}'
+                        f'{log_line.strip()}{colorama.Style.RESET_ALL}')
+
+        if 'Failed to clean up worker node' in log_line:
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.RED}'
+                        f'{log_line.strip()}{colorama.Style.RESET_ALL}')
+
+        # Final status for the cluster deployment
+        if 'Cluster deployment completed.' in log_line:
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.GREEN}'
+                        '✔ SkyPilot runtime is up.'
+                        f'{colorama.Style.RESET_ALL}')
+
+        if 'Failed to deploy Kubernetes on the following nodes:' in log_line:
+            logger.info(log_line.strip())
+
+        if 'already exists in history. ' in log_line:
+            node_name = log_line.split('(')[-1].split(')')[0]
+            logger.info(f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.YELLOW}'
+                        '✔ SkyPilot runtime already deployed on worker node '
+                        f'{node_name}. Skipping.{colorama.Style.RESET_ALL}')
+
+        if 'Failed to setup TCP forwarding on head node' in log_line:
+            node_name = log_line.split('(')[-1].split(')')[0]
+            logger.info(
+                f'{ux_utils.INDENT_SYMBOL}{colorama.Fore.RED}'
+                f'✗ Failed to setup TCP forwarding on head node {node_name}.'
+                f'{colorama.Style.RESET_ALL}')
+
+    def __exit__(self, except_type: Optional[Type[BaseException]],
+                 except_value: Optional[BaseException],
+                 traceback: Optional[types.TracebackType]) -> None:
+        del except_type, except_value, traceback  # unused
+        self.status_display.stop()
+
+
 def create_table(field_names: List[str], **kwargs) -> prettytable.PrettyTable:
     """Creates table with default style."""
     border = kwargs.pop('border', False)

sky/utils/resources_utils.py

@@ -50,6 +50,20 @@ class DiskTier(enum.Enum):
         return types.index(self) <= types.index(other)
 
 
+class StorageType(enum.Enum):
+    """Storage type."""
+    # Durable network storage, e.g. GCP persistent disks
+    NETWORK = 'network'
+    # Local instance storage, e.g. GCP local SSDs
+    INSTANCE = 'instance'
+
+
+class DiskAttachMode(enum.Enum):
+    """Disk attach mode."""
+    READ_ONLY = 'read_only'
+    READ_WRITE = 'read_write'
+
+
 @dataclasses.dataclass
 class ClusterName:
     display_name: str

sky/utils/schemas.py

@@ -188,6 +188,35 @@ def _get_single_resources_schema():
                     }
                 }],
             },
+            'volumes': {
+                'type': 'array',
+                'items': {
+                    'type': 'object',
+                    'properties': {
+                        'disk_size': {
+                            'type': 'integer',
+                        },
+                        'disk_tier': {
+                            'type': 'string',
+                        },
+                        'path': {
+                            'type': 'string',
+                        },
+                        'auto_delete': {
+                            'type': 'boolean',
+                        },
+                        'storage_type': {
+                            'type': 'string',
+                        },
+                        'name': {
+                            'type': 'string',
+                        },
+                        'attach_mode': {
+                            'type': 'string',
+                        },
+                    },
+                },
+            },
             'disk_size': {
                 'type': 'integer',
             },
@@ -372,6 +401,23 @@ def get_storage_schema():
                     mode.value for mode in storage.StorageMode
                 ]
             },
+            'config': {
+                'type': 'object',
+                'properties': {
+                    'disk_size': {
+                        'type': 'integer',
+                    },
+                    'disk_tier': {
+                        'type': 'string',
+                    },
+                    'storage_type': {
+                        'type': 'string',
+                    },
+                    'attach_mode': {
+                        'type': 'string',
+                    },
+                },
+            },
             '_is_sky_managed': {
                 'type': 'boolean',
             },
@@ -1043,6 +1089,27 @@ def get_config_schema():
                     'fabric': {
                         'type': 'string',
                     },
+                    'filesystems': {
+                        'type': 'array',
+                        'items': {
+                            'type': 'object',
+                            'additionalProperties': False,
+                            'properties': {
+                                'filesystem_id': {
+                                    'type': 'string',
+                                },
+                                'attach_mode': {
+                                    'type': 'string',
+                                    'case_sensitive_enum': [
+                                        'READ_WRITE', 'READ_ONLY'
+                                    ]
+                                },
+                                'mount_path': {
+                                    'type': 'string',
+                                }
+                            }
+                        }
+                    },
                 }
             },
         }

sky/utils/ux_utils.py

@@ -161,7 +161,8 @@ def finishing_message(message: str,
     follow_up_message = follow_up_message if (follow_up_message
                                               is not None) else ''
     success_prefix = (f'{colorama.Style.RESET_ALL}{colorama.Fore.GREEN}✓ '
-                      f'{message}{colorama.Style.RESET_ALL}{follow_up_message}')
+                      f'{message}{colorama.Style.RESET_ALL}{follow_up_message}'
+                      f'{colorama.Style.RESET_ALL}')
     if log_path is None:
         return success_prefix
     path_hint = log_path_hint(log_path, is_local)

{skypilot_nightly-1.0.0.dev20250521.dist-info → skypilot_nightly-1.0.0.dev20250523.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: skypilot-nightly
-Version: 1.0.0.dev20250521
+Version: 1.0.0.dev20250523
 Summary: SkyPilot: Run AI on Any Infra — Unified, Faster, Cheaper.
 Author: SkyPilot Team
 License: Apache 2.0
@@ -87,6 +87,9 @@ Requires-Dist: ray[default]!=2.6.0,>=2.2.0; extra == "oci"
 Provides-Extra: kubernetes
 Requires-Dist: kubernetes!=32.0.0,>=20.0.0; extra == "kubernetes"
 Requires-Dist: websockets; extra == "kubernetes"
+Provides-Extra: ssh
+Requires-Dist: kubernetes!=32.0.0,>=20.0.0; extra == "ssh"
+Requires-Dist: websockets; extra == "ssh"
 Provides-Extra: remote
 Requires-Dist: grpcio!=1.48.0,>=1.32.0; python_version < "3.10" and extra == "remote"
 Requires-Dist: grpcio!=1.48.0,>=1.42.0; python_version >= "3.10" and extra == "remote"
@@ -142,6 +145,8 @@ Requires-Dist: oci; extra == "all"
 Requires-Dist: ray[default]!=2.6.0,>=2.2.0; extra == "all"
 Requires-Dist: kubernetes!=32.0.0,>=20.0.0; extra == "all"
 Requires-Dist: websockets; extra == "all"
+Requires-Dist: kubernetes!=32.0.0,>=20.0.0; extra == "all"
+Requires-Dist: websockets; extra == "all"
 Requires-Dist: grpcio!=1.48.0,>=1.32.0; python_version < "3.10" and extra == "all"
 Requires-Dist: grpcio!=1.48.0,>=1.42.0; python_version >= "3.10" and extra == "all"
 Requires-Dist: protobuf!=3.19.5,>=3.15.3; extra == "all"