skypilot-nightly 1.0.0.dev20241227__py3-none-any.whl → 1.0.0.dev20250124__py3-none-any.whl

sky/clouds/utils/oci_utils.py

@@ -6,6 +6,12 @@ History:
    configuration.
  - Hysun He (hysun.he@oracle.com) @ Nov.12, 2024: Add the constant
    SERVICE_PORT_RULE_TAG
+ - Hysun He (hysun.he@oracle.com) @ Jan.01, 2025: Set the default image
+   from ubuntu 20.04 to ubuntu 22.04, including:
+   - GPU: skypilot:gpu-ubuntu-2004 -> skypilot:gpu-ubuntu-2204
+   - CPU: skypilot:cpu-ubuntu-2004 -> skypilot:cpu-ubuntu-2204
+ - Hysun He (hysun.he@oracle.com) @ Jan.01, 2025: Support reuse existing
+   VCN for SkyServe.
 """
 import os
 
@@ -105,8 +111,15 @@ class OCIConfig:
             ('oci', region, 'compartment_ocid'), default_compartment_ocid)
         return compartment
 
+    @classmethod
+    def get_vcn_ocid(cls, region):
+        # Will reuse the regional VCN if specified.
+        vcn = skypilot_config.get_nested(('oci', region, 'vcn_ocid'), None)
+        return vcn
+
     @classmethod
     def get_vcn_subnet(cls, region):
+        # Will reuse the subnet if specified.
         vcn = skypilot_config.get_nested(('oci', region, 'vcn_subnet'), None)
         return vcn
 
@@ -117,7 +130,7 @@ class OCIConfig:
         # the sky's user-config file (if not specified, use the hardcode one at
         # last)
         return skypilot_config.get_nested(('oci', 'default', 'image_tag_gpu'),
-                                          'skypilot:gpu-ubuntu-2004')
+                                          'skypilot:gpu-ubuntu-2204')
 
     @classmethod
     def get_default_image_tag(cls) -> str:
@@ -125,7 +138,7 @@ class OCIConfig:
         # set the default image tag in the sky's user-config file. (if not
         # specified, use the hardcode one at last)
         return skypilot_config.get_nested(
-            ('oci', 'default', 'image_tag_general'), 'skypilot:cpu-ubuntu-2004')
+            ('oci', 'default', 'image_tag_general'), 'skypilot:cpu-ubuntu-2204')
 
     @classmethod
     def get_sky_user_config_file(cls) -> str:

sky/core.py

@@ -915,8 +915,11 @@ def storage_delete(name: str) -> None:
     handle = global_user_state.get_handle_from_storage_name(name)
     if handle is None:
         raise ValueError(f'Storage name {name!r} not found.')
-    else:
-        storage_object = data.Storage(name=handle.storage_name,
-                                      source=handle.source,
-                                      sync_on_reconstruction=False)
-        storage_object.delete()
+
+    assert handle.storage_name == name, (
+        f'In global_user_state, storage name {name!r} does not match '
+        f'handle.storage_name {handle.storage_name!r}')
+    storage_object = data.Storage(name=handle.storage_name,
+                                  source=handle.source,
+                                  sync_on_reconstruction=False)
+    storage_object.delete()

sky/data/data_transfer.py

@@ -200,3 +200,40 @@ def _add_bucket_iam_member(bucket_name: str, role: str, member: str) -> None:
     bucket.set_iam_policy(policy)
 
     logger.debug(f'Added {member} with role {role} to {bucket_name}.')
+
+
+def s3_to_oci(s3_bucket_name: str, oci_bucket_name: str) -> None:
+    """Creates a one-time transfer from Amazon S3 to OCI Object Storage.
+    Args:
+      s3_bucket_name: str; Name of the Amazon S3 Bucket
+      oci_bucket_name: str; Name of the OCI Bucket
+    """
+    # TODO(HysunHe): Implement sync with other clouds (s3, gs)
+    raise NotImplementedError('Moving data directly from S3 to OCI bucket '
+                              'is currently not supported. Please specify '
+                              'a local source for the storage object.')
+
+
+def gcs_to_oci(gs_bucket_name: str, oci_bucket_name: str) -> None:
+    """Creates a one-time transfer from Google Cloud Storage to
+    OCI Object Storage.
+    Args:
+      gs_bucket_name: str; Name of the Google Cloud Storage Bucket
+      oci_bucket_name: str; Name of the OCI Bucket
+    """
+    # TODO(HysunHe): Implement sync with other clouds (s3, gs)
+    raise NotImplementedError('Moving data directly from GCS to OCI bucket '
+                              'is currently not supported. Please specify '
+                              'a local source for the storage object.')
+
+
+def r2_to_oci(r2_bucket_name: str, oci_bucket_name: str) -> None:
+    """Creates a one-time transfer from Cloudflare R2 to OCI Bucket.
+    Args:
+      r2_bucket_name: str; Name of the Cloudflare R2 Bucket
+      oci_bucket_name: str; Name of the OCI Bucket
+    """
+    raise NotImplementedError(
+        'Moving data directly from Cloudflare R2 to OCI '
+        'bucket is currently not supported. Please specify '
+        'a local source for the storage object.')

sky/data/data_utils.py

@@ -523,10 +523,14 @@ def get_gsutil_command() -> Tuple[str, str]:
 
 def run_upload_cli(command: str, access_denied_message: str, bucket_name: str,
                    log_path: str):
-    returncode, stdout, stderr = log_lib.run_with_log(command,
-                                                      log_path,
-                                                      shell=True,
-                                                      require_outputs=True)
+    returncode, stdout, stderr = log_lib.run_with_log(
+        command,
+        log_path,
+        shell=True,
+        require_outputs=True,
+        # We need to use bash as some of the cloud commands uses bash syntax,
+        # such as [[ ... ]]
+        executable='/bin/bash')
     if access_denied_message in stderr:
         with ux_utils.print_exception_no_traceback():
             raise PermissionError('Failed to upload files to '
@@ -730,3 +734,14 @@ class Rclone():
                 lines_to_keep.append(line)
 
         return lines_to_keep
+
+
+def split_oci_path(oci_path: str) -> Tuple[str, str]:
+    """Splits OCI Path into Bucket name and Relative Path to Bucket
+    Args:
+      oci_path: str; OCI Path, e.g. oci://imagenet/train/
+    """
+    path_parts = oci_path.replace('oci://', '').split('/')
+    bucket = path_parts.pop(0)
+    key = '/'.join(path_parts)
+    return bucket, key

sky/data/mounting_utils.py

@@ -19,6 +19,7 @@ BLOBFUSE2_VERSION = '2.2.0'
 _BLOBFUSE_CACHE_ROOT_DIR = '~/.sky/blobfuse2_cache'
 _BLOBFUSE_CACHE_DIR = ('~/.sky/blobfuse2_cache/'
                        '{storage_account_name}_{container_name}')
+RCLONE_VERSION = 'v1.68.2'
 
 
 def get_s3_mount_install_cmd() -> str:
@@ -30,12 +31,19 @@ def get_s3_mount_install_cmd() -> str:
     return install_cmd
 
 
-def get_s3_mount_cmd(bucket_name: str, mount_path: str) -> str:
+# pylint: disable=invalid-name
+def get_s3_mount_cmd(bucket_name: str,
+                     mount_path: str,
+                     _bucket_sub_path: Optional[str] = None) -> str:
     """Returns a command to mount an S3 bucket using goofys."""
+    if _bucket_sub_path is None:
+        _bucket_sub_path = ''
+    else:
+        _bucket_sub_path = f':{_bucket_sub_path}'
     mount_cmd = ('goofys -o allow_other '
                  f'--stat-cache-ttl {_STAT_CACHE_TTL} '
                  f'--type-cache-ttl {_TYPE_CACHE_TTL} '
-                 f'{bucket_name} {mount_path}')
+                 f'{bucket_name}{_bucket_sub_path} {mount_path}')
     return mount_cmd
 
 
@@ -49,15 +57,20 @@ def get_gcs_mount_install_cmd() -> str:
     return install_cmd
 
 
-def get_gcs_mount_cmd(bucket_name: str, mount_path: str) -> str:
+# pylint: disable=invalid-name
+def get_gcs_mount_cmd(bucket_name: str,
+                      mount_path: str,
+                      _bucket_sub_path: Optional[str] = None) -> str:
     """Returns a command to mount a GCS bucket using gcsfuse."""
-
+    bucket_sub_path_arg = f'--only-dir {_bucket_sub_path} '\
+        if _bucket_sub_path else ''
     mount_cmd = ('gcsfuse -o allow_other '
                  '--implicit-dirs '
                  f'--stat-cache-capacity {_STAT_CACHE_CAPACITY} '
                  f'--stat-cache-ttl {_STAT_CACHE_TTL} '
                  f'--type-cache-ttl {_TYPE_CACHE_TTL} '
                  f'--rename-dir-limit {_RENAME_DIR_LIMIT} '
+                 f'{bucket_sub_path_arg}'
                  f'{bucket_name} {mount_path}')
     return mount_cmd
 
@@ -78,10 +91,12 @@ def get_az_mount_install_cmd() -> str:
     return install_cmd
 
 
+# pylint: disable=invalid-name
 def get_az_mount_cmd(container_name: str,
                      storage_account_name: str,
                      mount_path: str,
-                     storage_account_key: Optional[str] = None) -> str:
+                     storage_account_key: Optional[str] = None,
+                     _bucket_sub_path: Optional[str] = None) -> str:
     """Returns a command to mount an AZ Container using blobfuse2.
 
     Args:
@@ -90,6 +105,7 @@ def get_az_mount_cmd(container_name: str,
             belongs to.
         mount_path: Path where the container will be mounting.
         storage_account_key: Access key for the given storage account.
+        _bucket_sub_path: Sub path of the mounting container.
 
     Returns:
         str: Command used to mount AZ container with blobfuse2.
@@ -106,25 +122,44 @@ def get_az_mount_cmd(container_name: str,
     cache_path = _BLOBFUSE_CACHE_DIR.format(
         storage_account_name=storage_account_name,
         container_name=container_name)
+    # The line below ensures the cache directory is new before mounting to
+    # avoid "config error in file_cache [temp directory not empty]" error, which
+    # can occur after stopping and starting the same cluster on Azure.
+    # This helps ensure a clean state for blobfuse2 operations.
+    remote_boot_time_cmd = 'date +%s -d "$(uptime -s)"'
+    if _bucket_sub_path is None:
+        bucket_sub_path_arg = ''
+    else:
+        bucket_sub_path_arg = f'--subdirectory={_bucket_sub_path}/ '
+    # TODO(zpoint): clear old cache that has been created in the previous boot.
     mount_cmd = (f'AZURE_STORAGE_ACCOUNT={storage_account_name} '
                  f'{key_env_var} '
                  f'blobfuse2 {mount_path} --allow-other --no-symlinks '
                  '-o umask=022 -o default_permissions '
-                 f'--tmp-path {cache_path} '
+                 f'--tmp-path {cache_path}_$({remote_boot_time_cmd}) '
+                 f'{bucket_sub_path_arg}'
                  f'--container-name {container_name}')
     return mount_cmd
 
 
-def get_r2_mount_cmd(r2_credentials_path: str, r2_profile_name: str,
-                     endpoint_url: str, bucket_name: str,
-                     mount_path: str) -> str:
+# pylint: disable=invalid-name
+def get_r2_mount_cmd(r2_credentials_path: str,
+                     r2_profile_name: str,
+                     endpoint_url: str,
+                     bucket_name: str,
+                     mount_path: str,
+                     _bucket_sub_path: Optional[str] = None) -> str:
     """Returns a command to install R2 mount utility goofys."""
+    if _bucket_sub_path is None:
+        _bucket_sub_path = ''
+    else:
+        _bucket_sub_path = f':{_bucket_sub_path}'
     mount_cmd = (f'AWS_SHARED_CREDENTIALS_FILE={r2_credentials_path} '
                  f'AWS_PROFILE={r2_profile_name} goofys -o allow_other '
                  f'--stat-cache-ttl {_STAT_CACHE_TTL} '
                  f'--type-cache-ttl {_TYPE_CACHE_TTL} '
                  f'--endpoint {endpoint_url} '
-                 f'{bucket_name} {mount_path}')
+                 f'{bucket_name}{_bucket_sub_path} {mount_path}')
     return mount_cmd
 
 
@@ -136,9 +171,12 @@ def get_cos_mount_install_cmd() -> str:
     return install_cmd
 
 
-def get_cos_mount_cmd(rclone_config_data: str, rclone_config_path: str,
-                      bucket_rclone_profile: str, bucket_name: str,
-                      mount_path: str) -> str:
+def get_cos_mount_cmd(rclone_config_data: str,
+                      rclone_config_path: str,
+                      bucket_rclone_profile: str,
+                      bucket_name: str,
+                      mount_path: str,
+                      _bucket_sub_path: Optional[str] = None) -> str:
     """Returns a command to mount an IBM COS bucket using rclone."""
     # creates a fusermount soft link on older (<22) Ubuntu systems for
     # rclone's mount utility.
@@ -150,14 +188,60 @@ def get_cos_mount_cmd(rclone_config_data: str, rclone_config_path: str,
                                 'mkdir -p ~/.config/rclone/ && '
                                 f'echo "{rclone_config_data}" >> '
                                 f'{rclone_config_path}')
+    if _bucket_sub_path is None:
+        sub_path_arg = f'{bucket_name}/{_bucket_sub_path}'
+    else:
+        sub_path_arg = f'/{bucket_name}'
     # --daemon will keep the mounting process running in the background.
     mount_cmd = (f'{configure_rclone_profile} && '
                  'rclone mount '
-                 f'{bucket_rclone_profile}:{bucket_name} {mount_path} '
+                 f'{bucket_rclone_profile}:{sub_path_arg} {mount_path} '
                  '--daemon')
     return mount_cmd
 
 
+def get_rclone_install_cmd() -> str:
+    """ RClone installation for both apt-get and rpm.
+    This would be common command.
+    """
+    # pylint: disable=line-too-long
+    install_cmd = (
+        f'(which dpkg > /dev/null 2>&1 && (which rclone > /dev/null || (cd ~ > /dev/null'
+        f' && curl -O https://downloads.rclone.org/{RCLONE_VERSION}/rclone-{RCLONE_VERSION}-linux-amd64.deb'
+        f' && sudo dpkg -i rclone-{RCLONE_VERSION}-linux-amd64.deb'
+        f' && rm -f rclone-{RCLONE_VERSION}-linux-amd64.deb)))'
+        f' || (which rclone > /dev/null || (cd ~ > /dev/null'
+        f' && curl -O https://downloads.rclone.org/{RCLONE_VERSION}/rclone-{RCLONE_VERSION}-linux-amd64.rpm'
+        f' && sudo yum --nogpgcheck install rclone-{RCLONE_VERSION}-linux-amd64.rpm -y'
+        f' && rm -f rclone-{RCLONE_VERSION}-linux-amd64.rpm))')
+    return install_cmd
+
+
+def get_oci_mount_cmd(mount_path: str, store_name: str, region: str,
+                      namespace: str, compartment: str, config_file: str,
+                      config_profile: str) -> str:
+    """ OCI specific RClone mount command for oci object storage. """
+    # pylint: disable=line-too-long
+    mount_cmd = (
+        f'sudo chown -R `whoami` {mount_path}'
+        f' && rclone config create oos_{store_name} oracleobjectstorage'
+        f' provider user_principal_auth namespace {namespace}'
+        f' compartment {compartment} region {region}'
+        f' oci-config-file {config_file}'
+        f' oci-config-profile {config_profile}'
+        f' && sed -i "s/oci-config-file/config_file/g;'
+        f' s/oci-config-profile/config_profile/g" ~/.config/rclone/rclone.conf'
+        f' && ([ ! -f /bin/fusermount3 ] && sudo ln -s /bin/fusermount /bin/fusermount3 || true)'
+        f' && (grep -q {mount_path} /proc/mounts || rclone mount oos_{store_name}:{store_name} {mount_path} --daemon --allow-non-empty)'
+    )
+    return mount_cmd
+
+
+def get_rclone_version_check_cmd() -> str:
+    """ RClone version check. This would be common command. """
+    return f'rclone --version | grep -q {RCLONE_VERSION}'
+
+
 def _get_mount_binary(mount_cmd: str) -> str:
     """Returns mounting binary in string given as the mount command.
 
@@ -209,7 +293,7 @@ def get_mounting_script(
     script = textwrap.dedent(f"""
         #!/usr/bin/env bash
         set -e
-                             
+
         {command_runner.ALIAS_SUDO_TO_EMPTY_FOR_ROOT_CMD}
 
         MOUNT_PATH={mount_path}