skypilot-nightly 1.0.0.dev20250825__py3-none-any.whl → 1.0.0.dev20250827__py3-none-any.whl

sky/server/rest.py

@@ -48,6 +48,13 @@ _session.headers[constants.API_VERSION_HEADER] = str(constants.API_VERSION)
 _session.headers[constants.VERSION_HEADER] = (
     versions.get_local_readable_version())
 
+# Enumerate error types that might be transient and can be addressed by
+# retrying.
+_transient_errors = [
+    requests.exceptions.RequestException,
+    ConnectionError,
+]
+
 
 class RetryContext:
 
@@ -87,14 +94,10 @@ def retry_transient_errors(max_retries: int = 3,
         if isinstance(e, requests.exceptions.HTTPError):
             # Only server error is considered as transient.
             return e.response.status_code >= 500
-        if isinstance(e, exceptions.ClientError):
-            return False
-        # It is hard to enumerate all other errors that are transient, e.g.
-        # broken pipe, connection refused, etc. Instead, it is safer to assume
-        # all other errors might be transient since we only retry for 3 times
-        # by default. For permanent errors that we do not know now, we can
-        # exclude them here in the future.
-        return True
+        for error in _transient_errors:
+            if isinstance(e, error):
+                return True
+        return False
 
     def decorator(func: F) -> F:
 

sky/server/server.py

@@ -7,7 +7,6 @@ import contextlib
 import datetime
 import hashlib
 import json
-import logging
 import multiprocessing
 import os
 import pathlib
@@ -24,7 +23,6 @@ import zipfile
 import aiofiles
 import fastapi
 from fastapi.middleware import cors
-from passlib.hash import apr_md5_crypt
 import starlette.middleware.base
 import uvloop
 
@@ -69,7 +67,6 @@ from sky.utils import common_utils
 from sky.utils import context
 from sky.utils import context_utils
 from sky.utils import dag_utils
-from sky.utils import env_options
 from sky.utils import status_lib
 from sky.utils import subprocess_utils
 from sky.volumes.server import server as volumes_rest
@@ -85,31 +82,6 @@ P = ParamSpec('P')
 
 _SERVER_USER_HASH_KEY = 'server_user_hash'
 
-
-def _add_timestamp_prefix_for_server_logs() -> None:
-    server_logger = sky_logging.init_logger('sky.server')
-    # Clear existing handlers first to prevent duplicates
-    server_logger.handlers.clear()
-    # Disable propagation to avoid the root logger of SkyPilot being affected
-    server_logger.propagate = False
-    # Add date prefix to the log message printed by loggers under
-    # server.
-    stream_handler = logging.StreamHandler(sys.stdout)
-    if env_options.Options.SHOW_DEBUG_INFO.get():
-        stream_handler.setLevel(logging.DEBUG)
-    else:
-        stream_handler.setLevel(logging.INFO)
-    stream_handler.flush = sys.stdout.flush  # type: ignore
-    stream_handler.setFormatter(sky_logging.FORMATTER)
-    server_logger.addHandler(stream_handler)
-    # Add date prefix to the log message printed by uvicorn.
-    for name in ['uvicorn', 'uvicorn.access']:
-        uvicorn_logger = logging.getLogger(name)
-        uvicorn_logger.handlers.clear()
-        uvicorn_logger.addHandler(stream_handler)
-
-
-_add_timestamp_prefix_for_server_logs()
 logger = sky_logging.init_logger(__name__)
 
 # TODO(zhwu): Streaming requests, such log tailing after sky launch or sky logs,
@@ -148,7 +120,7 @@ def _try_set_basic_auth_user(request: fastapi.Request):
         username_encoded = username.encode('utf8')
         db_username_encoded = user.name.encode('utf8')
         if (username_encoded == db_username_encoded and
-                apr_md5_crypt.verify(password, user.password)):
+                common.crypt_ctx.verify(password, user.password)):
             request.state.auth_user = user
             break
 
@@ -248,7 +220,7 @@ class BasicAuthMiddleware(starlette.middleware.base.BaseHTTPMiddleware):
             username_encoded = username.encode('utf8')
             db_username_encoded = user.name.encode('utf8')
             if (username_encoded == db_username_encoded and
-                    apr_md5_crypt.verify(password, user.password)):
+                    common.crypt_ctx.verify(password, user.password)):
                 valid_user = True
                 request.state.auth_user = user
                 await authn.override_user_info_in_request_body(request, user)
@@ -852,6 +824,15 @@ async def upload_zip_file(request: fastapi.Request, user_hash: str,
         chunk_index: The chunk index, starting from 0.
         total_chunks: The total number of chunks.
     """
+    # Field _body would be set if the request body has been received, fail fast
+    # to surface potential memory issues, i.e. catch the issue in our smoke
+    # test.
+    # pylint: disable=protected-access
+    if hasattr(request, '_body'):
+        raise fastapi.HTTPException(
+            status_code=500,
+            detail='Upload request body should not be received before streaming'
+        )
     # Add the upload id to the cleanup list.
     upload_ids_to_cleanup[(upload_id,
                            user_hash)] = (datetime.datetime.now() +
@@ -919,8 +900,9 @@ async def upload_zip_file(request: fastapi.Request, user_hash: str,
         zip_file_path.rename(zip_file_path.with_suffix(''))
         missing_chunks = get_missing_chunks(total_chunks)
         if missing_chunks:
-            return payloads.UploadZipFileResponse(status='uploading',
-                                                  missing_chunks=missing_chunks)
+            return payloads.UploadZipFileResponse(
+                status=responses.UploadStatus.UPLOADING.value,
+                missing_chunks=missing_chunks)
         zip_file_path = client_file_mounts_dir / f'{upload_id}.zip'
         async with aiofiles.open(zip_file_path, 'wb') as zip_file:
             for chunk in range(total_chunks):
@@ -937,7 +919,8 @@ async def upload_zip_file(request: fastapi.Request, user_hash: str,
     unzip_file(zip_file_path, client_file_mounts_dir)
     if total_chunks > 1:
         shutil.rmtree(chunk_dir)
-    return payloads.UploadZipFileResponse(status='completed')
+    return payloads.UploadZipFileResponse(
+        status=responses.UploadStatus.COMPLETED.value)
 
 
 def _is_relative_to(path: pathlib.Path, parent: pathlib.Path) -> bool:
@@ -1859,6 +1842,8 @@ if __name__ == '__main__':
 
     from sky.server import uvicorn as skyuvicorn
 
+    skyuvicorn.add_timestamp_prefix_for_server_logs()
+
     # Initialize global user state db
     global_user_state.initialize_and_get_db()
     # Initialize request db

sky/server/uvicorn.py

@@ -4,8 +4,10 @@ This module is a wrapper around uvicorn to customize the behavior of the
 server.
 """
 import asyncio
+import logging
 import os
 import signal
+import sys
 import threading
 import time
 from types import FrameType
@@ -21,6 +23,7 @@ from sky.server import state
 from sky.server.requests import requests as requests_lib
 from sky.skylet import constants
 from sky.utils import context_utils
+from sky.utils import env_options
 from sky.utils import subprocess_utils
 
 logger = sky_logging.init_logger(__name__)
@@ -47,6 +50,35 @@ _RETRIABLE_REQUEST_NAMES = [
 ]
 
 
+def add_timestamp_prefix_for_server_logs() -> None:
+    """Configure logging for API server.
+
+    Note: we only do this in the main API server process and uvicorn processes,
+    to avoid affecting executor logs (including in modules like
+    sky.server.requests) that may get sent to the client.
+    """
+    server_logger = sky_logging.init_logger('sky.server')
+    # Clear existing handlers first to prevent duplicates
+    server_logger.handlers.clear()
+    # Disable propagation to avoid the root logger of SkyPilot being affected
+    server_logger.propagate = False
+    # Add date prefix to the log message printed by loggers under
+    # server.
+    stream_handler = logging.StreamHandler(sys.stdout)
+    if env_options.Options.SHOW_DEBUG_INFO.get():
+        stream_handler.setLevel(logging.DEBUG)
+    else:
+        stream_handler.setLevel(logging.INFO)
+    stream_handler.flush = sys.stdout.flush  # type: ignore
+    stream_handler.setFormatter(sky_logging.FORMATTER)
+    server_logger.addHandler(stream_handler)
+    # Add date prefix to the log message printed by uvicorn.
+    for name in ['uvicorn', 'uvicorn.access']:
+        uvicorn_logger = logging.getLogger(name)
+        uvicorn_logger.handlers.clear()
+        uvicorn_logger.addHandler(stream_handler)
+
+
 class Server(uvicorn.Server):
     """Server wrapper for uvicorn.
 
@@ -162,6 +194,7 @@ class Server(uvicorn.Server):
 
     def run(self, *args, **kwargs):
         """Run the server process."""
+        add_timestamp_prefix_for_server_logs()
         context_utils.hijack_sys_attrs()
         # Use default loop policy of uvicorn (use uvloop if available).
         self.config.setup_event_loop()

sky/setup_files/dependencies.py

@@ -65,6 +65,7 @@ install_requires = [
     # Required for API server metrics
     'prometheus_client>=0.8.0',
     'passlib',
+    'bcrypt',
     'pyjwt',
     'gitpython',
     'types-paramiko',

sky/sky_logging.py

@@ -202,7 +202,10 @@ def set_sky_logging_levels(level: int):
 
 
 def logging_enabled(logger: logging.Logger, level: int) -> bool:
-    return logger.level <= level
+    # Note(cooperc): This may return true in a lot of cases where we won't
+    # actually log anything, since the log level is set on the handler in
+    # _setup_logger.
+    return logger.getEffectiveLevel() <= level
 
 
 @contextlib.contextmanager

sky/skylet/events.py

@@ -7,7 +7,6 @@ import time
 import traceback
 
 import psutil
-import yaml
 
 from sky import clouds
 from sky import sky_logging
@@ -24,6 +23,7 @@ from sky.utils import cluster_utils
 from sky.utils import common_utils
 from sky.utils import registry
 from sky.utils import ux_utils
+from sky.utils import yaml_utils
 
 # Seconds of sleep between the processing of skylet events.
 EVENT_CHECKING_INTERVAL_SECONDS = 20
@@ -309,7 +309,7 @@ class AutostopEvent(SkyletEvent):
         else:
             yaml_str = self._CATCH_NODES.sub(r'cache_stopped_nodes: true',
                                              yaml_str)
-        config = yaml.safe_load(yaml_str)
+        config = yaml_utils.safe_load(yaml_str)
         # Set the private key with the existed key on the remote instance.
         config['auth']['ssh_private_key'] = '~/ray_bootstrap_key.pem'
         # NOTE: We must do this, otherwise with ssh_proxy_command still under

sky/skypilot_config.py

@@ -75,6 +75,7 @@ from sky.utils import config_utils
 from sky.utils import context
 from sky.utils import schemas
 from sky.utils import ux_utils
+from sky.utils import yaml_utils
 from sky.utils.db import db_utils
 from sky.utils.kubernetes import config_map_utils
 
@@ -532,7 +533,7 @@ def _parse_dotlist(dotlist: List[str]) -> config_utils.Config:
         if len(key) == 0 or len(value) == 0:
             raise ValueError(f'Invalid config override: {arg}. '
                              'Please use the format: key=value')
-        value = yaml.safe_load(value)
+        value = yaml_utils.safe_load(value)
         nested_keys = tuple(key.split('.'))
         config.set_nested(nested_keys, value)
     return config
@@ -585,7 +586,8 @@ def _reload_config_as_server() -> None:
                     row = session.query(config_yaml_table).filter_by(
                         key=key).first()
                 if row:
-                    db_config = config_utils.Config(yaml.safe_load(row.value))
+                    db_config = config_utils.Config(
+                        yaml_utils.safe_load(row.value))
                     db_config.pop_nested(('db',), None)
                     return db_config
                 return None

sky/ssh_node_pools/core.py

@@ -5,6 +5,8 @@ from typing import Any, Dict, List
 
 import yaml
 
+from sky.utils import yaml_utils
+
 
 class SSHNodePoolManager:
     """Manager for SSH Node Pool configurations."""
@@ -21,7 +23,7 @@ class SSHNodePoolManager:
 
         try:
             with open(self.config_path, 'r', encoding='utf-8') as f:
-                return yaml.safe_load(f) or {}
+                return yaml_utils.safe_load(f) or {}
         except Exception as e:
             raise RuntimeError(
                 f'Failed to read SSH Node Pool config: {e}') from e

sky/task.py

@@ -4,7 +4,6 @@ import inspect
 import json
 import os
 import re
-import typing
 from typing import (Any, Callable, Dict, Iterable, List, Optional, Set, Tuple,
                     Union)
 
@@ -15,7 +14,6 @@ from sky import dag as dag_lib
 from sky import exceptions
 from sky import resources as resources_lib
 from sky import sky_logging
-from sky.adaptors import common as adaptors_common
 from sky.data import data_utils
 from sky.data import storage as storage_lib
 from sky.provision import docker_utils
@@ -26,11 +24,7 @@ from sky.utils import registry
 from sky.utils import schemas
 from sky.utils import ux_utils
 from sky.utils import volume as volume_lib
-
-if typing.TYPE_CHECKING:
-    import yaml
-else:
-    yaml = adaptors_common.LazyImport('yaml')
+from sky.utils import yaml_utils
 
 logger = sky_logging.init_logger(__name__)
 
@@ -836,7 +830,7 @@ class Task:
             #  https://github.com/yaml/pyyaml/issues/165#issuecomment-430074049
             # to raise errors on duplicate keys.
             user_specified_yaml = f.read()
-            config = yaml.safe_load(user_specified_yaml)
+            config = yaml_utils.safe_load(user_specified_yaml)
 
         if isinstance(config, str):
             with ux_utils.print_exception_no_traceback():
@@ -1611,7 +1605,7 @@ class Task:
         if use_user_specified_yaml:
             if self._user_specified_yaml is None:
                 return self._to_yaml_config(redact_secrets=True)
-            config = yaml.safe_load(self._user_specified_yaml)
+            config = yaml_utils.safe_load(self._user_specified_yaml)
             if config.get('secrets') is not None:
                 config['secrets'] = {k: '<redacted>' for k in config['secrets']}
             return config

sky/users/server.py

@@ -10,11 +10,11 @@ from typing import Any, Dict, Generator, List
 
 import fastapi
 import filelock
-from passlib.hash import apr_md5_crypt
 
 from sky import global_user_state
 from sky import models
 from sky import sky_logging
+from sky.server import common as server_common
 from sky.server.requests import payloads
 from sky.skylet import constants
 from sky.users import permission
@@ -86,7 +86,7 @@ async def user_create(user_create_body: payloads.UserCreateBody) -> None:
         role = rbac.get_default_role()
 
     # Create user
-    password_hash = apr_md5_crypt.hash(password)
+    password_hash = server_common.crypt_ctx.hash(password)
     user_hash = hashlib.md5(
         username.encode()).hexdigest()[:common_utils.USER_HASH_LENGTH]
     with _user_lock(user_hash):
@@ -146,7 +146,7 @@ async def user_update(request: fastapi.Request,
 
     with _user_lock(user_info.id):
         if password:
-            password_hash = apr_md5_crypt.hash(password)
+            password_hash = server_common.crypt_ctx.hash(password)
             global_user_state.add_or_update_user(
                 models.User(id=user_info.id,
                             name=user_info.name,
@@ -271,13 +271,13 @@ async def user_import(
                 creation_errors.append(f'{username}: User already exists')
                 continue
 
-            # Check if password is already hashed (APR1 hash)
-            if password.startswith('$apr1$'):
+            # Check if password is already hashed
+            if server_common.crypt_ctx.identify(password) is not None:
                 # Password is already hashed, use it directly
                 password_hash = password
             else:
                 # Password is plain text, hash it
-                password_hash = apr_md5_crypt.hash(password)
+                password_hash = server_common.crypt_ctx.hash(password)
 
             user_hash = hashlib.md5(
                 username.encode()).hexdigest()[:common_utils.USER_HASH_LENGTH]

sky/utils/common_utils.py

@@ -30,6 +30,7 @@ from sky.usage import constants as usage_constants
 from sky.utils import annotations
 from sky.utils import ux_utils
 from sky.utils import validator
+from sky.utils import yaml_utils
 
 if typing.TYPE_CHECKING:
     import jinja2
@@ -577,13 +578,13 @@ def read_yaml(path: Optional[str]) -> Dict[str, Any]:
     if path is None:
         raise ValueError('Attempted to read a None YAML.')
     with open(path, 'r', encoding='utf-8') as f:
-        config = yaml.safe_load(f)
+        config = yaml_utils.safe_load(f)
     return config
 
 
 def read_yaml_all_str(yaml_str: str) -> List[Dict[str, Any]]:
     stream = io.StringIO(yaml_str)
-    config = yaml.safe_load_all(stream)
+    config = yaml_utils.safe_load_all(stream)
     configs = list(config)
     if not configs:
         # Empty YAML file.

sky/utils/yaml_utils.py

@@ -0,0 +1,35 @@
+"""YAML utilities."""
+from typing import Any, TYPE_CHECKING
+
+from sky.adaptors import common
+
+if TYPE_CHECKING:
+    import yaml
+else:
+    yaml = common.LazyImport('yaml')
+
+_csafe_loader_import_error = False
+
+
+def safe_load(stream) -> Any:
+    global _csafe_loader_import_error
+    if _csafe_loader_import_error:
+        return yaml.load(stream, Loader=yaml.SafeLoader)
+
+    try:
+        return yaml.load(stream, Loader=yaml.CSafeLoader)
+    except ImportError:
+        _csafe_loader_import_error = True
+        return yaml.load(stream, Loader=yaml.SafeLoader)
+
+
+def safe_load_all(stream) -> Any:
+    global _csafe_loader_import_error
+    if _csafe_loader_import_error:
+        return yaml.load_all(stream, Loader=yaml.SafeLoader)
+
+    try:
+        return yaml.load_all(stream, Loader=yaml.CSafeLoader)
+    except ImportError:
+        _csafe_loader_import_error = True
+        return yaml.load_all(stream, Loader=yaml.SafeLoader)

sky/volumes/volume.py

@@ -125,14 +125,19 @@ class Volume:
 
     def _validate_config(self) -> None:
         """Validate the volume config."""
+        assert self.cloud is not None, 'Cloud must be specified'
+        cloud_obj = registry.CLOUD_REGISTRY.from_str(self.cloud)
+        assert cloud_obj is not None
+
+        valid, err_msg = cloud_obj.is_volume_name_valid(self.name)
+        if not valid:
+            raise ValueError(f'Invalid volume name: {err_msg}')
+
         if not self.resource_name and not self.size:
             raise ValueError('Size is required for new volumes. '
                              'Please specify the size in the YAML file or '
                              'use the --size flag.')
         if self.labels:
-            assert self.cloud is not None, 'Cloud must be specified'
-            cloud_obj = registry.CLOUD_REGISTRY.from_str(self.cloud)
-            assert cloud_obj is not None
             for key, value in self.labels.items():
                 valid, err_msg = cloud_obj.is_label_valid(key, value)
                 if not valid:

{skypilot_nightly-1.0.0.dev20250825.dist-info → skypilot_nightly-1.0.0.dev20250827.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: skypilot-nightly
-Version: 1.0.0.dev20250825
+Version: 1.0.0.dev20250827
 Summary: SkyPilot: Run AI on Any Infra — Unified, Faster, Cheaper.
 Author: SkyPilot Team
 License: Apache 2.0
@@ -53,6 +53,7 @@ Requires-Dist: casbin
 Requires-Dist: sqlalchemy_adapter
 Requires-Dist: prometheus_client>=0.8.0
 Requires-Dist: passlib
+Requires-Dist: bcrypt
 Requires-Dist: pyjwt
 Requires-Dist: gitpython
 Requires-Dist: types-paramiko