skypilot-nightly 1.0.0.dev20250413__py3-none-any.whl → 1.0.0.dev20250421__py3-none-any.whl

sky/server/requests/payloads.py

@@ -6,7 +6,6 @@ with the backend functions. The benefit of having the default values in the
 payloads is that a user can find the default values in the Restful API docs.
 """
 import getpass
-import json
 import os
 import typing
 from typing import Any, Dict, List, Optional, Tuple, Union
@@ -47,7 +46,7 @@ def request_body_env_vars() -> dict:
     # Remove the path to config file, as the config content is included in the
     # request body and will be merged with the config on the server side.
     env_vars.pop(skypilot_config.ENV_VAR_SKYPILOT_CONFIG, None)
-    env_vars.pop(skypilot_config.ENV_VAR_USER_CONFIG, None)
+    env_vars.pop(skypilot_config.ENV_VAR_GLOBAL_CONFIG, None)
     env_vars.pop(skypilot_config.ENV_VAR_PROJECT_CONFIG, None)
     return env_vars
 
@@ -56,20 +55,9 @@ def get_override_skypilot_config_from_client() -> Dict[str, Any]:
     """Returns the override configs from the client."""
     config = skypilot_config.to_dict()
     # Remove the API server config, as we should not specify the SkyPilot
-    # server endpoint on the server side. This avoids the warning below.
+    # server endpoint on the server side. This avoids the warning at
+    # server-side.
     config.pop_nested(('api_server',), default_value=None)
-    ignored_key_values = {}
-    for nested_key in constants.SKIPPED_CLIENT_OVERRIDE_KEYS:
-        value = config.pop_nested(nested_key, default_value=None)
-        if value is not None:
-            ignored_key_values['.'.join(nested_key)] = value
-    if ignored_key_values:
-        logger.debug(f'The following keys ({json.dumps(ignored_key_values)}) '
-                     'are specified in the client SkyPilot config at '
-                     f'{skypilot_config.loaded_config_path()!r}. '
-                     'This will be ignored. If you want to specify it, '
-                     'please modify it on server side or contact your '
-                     'administrator.')
     return config
 
 
@@ -420,6 +408,15 @@ class ServeLogsBody(RequestBody):
     follow: bool = True
 
 
+class ServeDownloadLogsBody(RequestBody):
+    """The request body for the serve download logs endpoint."""
+    service_name: str
+    local_dir: str
+    targets: Optional[Union[str, serve.ServiceComponent,
+                            List[Union[str, serve.ServiceComponent]]]]
+    replica_ids: Optional[List[int]] = None
+
+
 class ServeStatusBody(RequestBody):
     """The request body for the serve status endpoint."""
     service_names: Optional[Union[str, List[str]]]

sky/server/requests/serializers/decoders.py

@@ -188,8 +188,5 @@ def decode_job_status(
 
 @register_decoders('kubernetes_node_info')
 def decode_kubernetes_node_info(
-        return_value: Dict[str, Any]) -> Dict[str, models.KubernetesNodeInfo]:
-    return {
-        node_name: models.KubernetesNodeInfo(**node_info)
-        for node_name, node_info in return_value.items()
-    }
+        return_value: Dict[str, Any]) -> models.KubernetesNodesInfo:
+    return models.KubernetesNodesInfo.from_dict(return_value)

sky/server/requests/serializers/encoders.py

@@ -159,8 +159,5 @@ def encode_job_status(return_value: Dict[int, Any]) -> Dict[int, str]:
 
 @register_encoder('kubernetes_node_info')
 def encode_kubernetes_node_info(
-        return_value: Dict[str, 'models.KubernetesNodeInfo']) -> Dict[str, Any]:
-    return {
-        node_name: dataclasses.asdict(node_info)
-        for node_name, node_info in return_value.items()
-    }
+        return_value: 'models.KubernetesNodesInfo') -> Dict[str, Any]:
+    return return_value.to_dict()

sky/server/server.py

@@ -150,7 +150,21 @@ async def lifespan(app: fastapi.FastAPI):  # pylint: disable=redefined-outer-nam
     # Shutdown: Add any cleanup code here if needed
 
 
+# Add a new middleware class to handle /internal/dashboard prefix
+class InternalDashboardPrefixMiddleware(
+        starlette.middleware.base.BaseHTTPMiddleware):
+    """Middleware to handle /internal/dashboard prefix in requests."""
+
+    async def dispatch(self, request: fastapi.Request, call_next):
+        path = request.url.path
+        if path.startswith('/internal/dashboard/'):
+            # Remove /internal/dashboard prefix and update request scope
+            request.scope['path'] = path.replace('/internal/dashboard/', '/', 1)
+        return await call_next(request)
+
+
 app = fastapi.FastAPI(prefix='/api/v1', debug=True, lifespan=lifespan)
+app.add_middleware(InternalDashboardPrefixMiddleware)
 app.add_middleware(
     cors.CORSMiddleware,
     # TODO(zhwu): in production deployment, we should restrict the allowed
@@ -210,7 +224,7 @@ async def kubernetes_node_info(
         request: fastapi.Request,
         kubernetes_node_info_body: payloads.KubernetesNodeInfoRequestBody
 ) -> None:
-    """Gets Kubernetes node information."""
+    """Gets Kubernetes nodes information and hints."""
     executor.schedule_request(
         request_id=request.state.request_id,
         request_name='kubernetes_node_info',
@@ -1101,6 +1115,35 @@ async def complete_storage_name(incomplete: str,) -> List[str]:
     return global_user_state.get_storage_names_start_with(incomplete)
 
 
+# Add a route to serve static files
+@app.get('/{full_path:path}')
+async def serve_static_or_dashboard(full_path: str):
+    """Serves static files for any unmatched routes.
+
+    Handles the /dashboard prefix from Next.js configuration.
+    """
+    # Check if the path starts with 'dashboard/' and remove it if it does
+    if full_path.startswith('dashboard/'):
+        full_path = full_path[len('dashboard/'):]
+
+    # Try to serve the file directly from the out directory first
+    file_path = os.path.join(server_constants.DASHBOARD_DIR, full_path)
+    if os.path.isfile(file_path):
+        return fastapi.responses.FileResponse(file_path)
+
+    # If file not found, serve the index.html for client-side routing.
+    # For example, the non-matched arbitrary route (/ or /test) from
+    # client will be redirected to the index.html.
+    index_path = os.path.join(server_constants.DASHBOARD_DIR, 'index.html')
+    try:
+        with open(index_path, 'r', encoding='utf-8') as f:
+            content = f.read()
+        return fastapi.responses.HTMLResponse(content=content)
+    except Exception as e:
+        logger.error(f'Error serving dashboard: {e}')
+        raise fastapi.HTTPException(status_code=500, detail=str(e))
+
+
 if __name__ == '__main__':
     import uvicorn
 

sky/setup_files/MANIFEST.in

@@ -15,3 +15,4 @@ include sky/jobs/dashboard/static/*
 include sky/templates/*
 include sky/utils/kubernetes/*
 include sky/server/html/*
+recursive-include sky/dashboard/out *

sky/setup_files/dependencies.py

@@ -53,6 +53,7 @@ install_requires = [
     'aiofiles',
     'httpx',
     'setproctitle',
+    'omegaconf>=2.4.0dev3,<2.5',
 ]
 
 local_ray = [

sky/sky_logging.py

@@ -18,6 +18,12 @@ _FORMAT = '%(levelname).1s %(asctime)s %(filename)s:%(lineno)d] %(message)s'
 _DATE_FORMAT = '%m-%d %H:%M:%S'
 _SENSITIVE_LOGGER = ['sky.provisioner', 'sky.optimizer']
 
+DEBUG = logging.DEBUG
+INFO = logging.INFO
+WARNING = logging.WARNING
+ERROR = logging.ERROR
+CRITICAL = logging.CRITICAL
+
 
 def _show_logging_prefix():
     return env_options.Options.SHOW_DEBUG_INFO.get(
@@ -97,8 +103,8 @@ def _setup_logger():
 def reload_logger():
     """Reload the logger.
 
-    This is useful when the logging configuration is changed.
-    e.g., the logging level is changed or stdout/stderr is reset.
+    This ensures that the logger takes the new environment variables,
+    such as SKYPILOT_DEBUG.
     """
     global _default_handler
     _root_logger.removeHandler(_default_handler)
@@ -127,6 +133,10 @@ def set_logging_level(logger: str, level: int):
         logger.setLevel(original_level)
 
 
+def logging_enabled(logger: logging.Logger, level: int) -> bool:
+    return logger.level <= level
+
+
 @contextlib.contextmanager
 def silent():
     """Make all sky_logging.print() and logger.{info, warning...} silent.

sky/skylet/constants.py

@@ -117,7 +117,7 @@ RUNPOD_DOCKER_USERNAME_ENV_VAR = 'SKYPILOT_RUNPOD_DOCKER_USERNAME'
 
 # Commands for disable GPU ECC, which can improve the performance of the GPU
 # for some workloads by 30%. This will only be applied when a user specify
-# `nvidia_gpus.disable_ecc: true` in ~/.sky/skyconfig.yaml.
+# `nvidia_gpus.disable_ecc: true` in ~/.sky/config.yaml.
 # Running this command will reboot the machine, introducing overhead for
 # provisioning the machine.
 # https://portal.nutanix.com/page/documents/kbs/details?targetId=kA00e000000LKjOCAW
@@ -299,11 +299,6 @@ FILE_MOUNTS_WORKDIR_SUBPATH = 'job-{run_id}/workdir'
 FILE_MOUNTS_SUBPATH = 'job-{run_id}/local-file-mounts/{i}'
 FILE_MOUNTS_TMP_SUBPATH = 'job-{run_id}/tmp-files'
 
-# The default idle timeout for SkyPilot controllers. This include jobs
-# controller and sky serve controller.
-# TODO(tian): Refactor to controller_utils. Current blocker: circular import.
-CONTROLLER_IDLE_MINUTES_TO_AUTOSTOP = 10
-
 # Due to the CPU/memory usage of the controller process launched with sky jobs (
 # use ray job under the hood), we need to reserve some CPU/memory for each jobs/
 # serve controller process.
@@ -337,7 +332,7 @@ RCLONE_LOG_DIR = '~/.sky/rclone_log'
 RCLONE_CACHE_DIR = '~/.cache/rclone'
 RCLONE_CACHE_REFRESH_INTERVAL = 10
 
-# The keys that can be overridden in the `~/.sky/skyconfig.yaml` file. The
+# The keys that can be overridden in the `~/.sky/config.yaml` file. The
 # overrides are specified in task YAMLs.
 OVERRIDEABLE_CONFIG_KEYS_IN_TASK: List[Tuple[str, ...]] = [
     ('docker', 'run_options'),
@@ -367,3 +362,6 @@ LOCAL_SKYPILOT_CONFIG_PATH_PLACEHOLDER = 'skypilot:local_skypilot_config_path'
 
 # Path to the generated cluster config yamls and ssh configs.
 SKY_USER_FILE_PATH = '~/.sky/generated'
+
+# Environment variable that is set to 'true' if this is a skypilot server.
+ENV_VAR_IS_SKYPILOT_SERVER = 'IS_SKYPILOT_SERVER'

sky/skylet/job_lib.py

@@ -630,7 +630,7 @@ def update_job_status(job_ids: List[int],
                         'it to FAILED_DRIVER')
                     status = JobStatus.FAILED_DRIVER
             elif job_pid < 0:
-                # TODO(zhwu): Backward compatibility, remove after 0.9.0.
+                # TODO(zhwu): Backward compatibility, remove after 0.10.0.
                 # We set the job status to PENDING instead of actually
                 # checking ray job status and let the status in job table
                 # take effect in the later max.
@@ -882,7 +882,7 @@ def cancel_jobs_encoded_results(jobs: Optional[List[int]],
                 # child processes.
             elif job['pid'] < 0:
                 try:
-                    # TODO(zhwu): Backward compatibility, remove after 0.9.0.
+                    # TODO(zhwu): Backward compatibility, remove after 0.10.0.
                     # The job was submitted with ray job submit before #4318.
                     job_client = _create_ray_job_submission_client()
                     job_client.stop_job(_make_ray_job_id(job['job_id']))
@@ -1008,7 +1008,7 @@ class JobLibCodeGen:
             # Print cancelled IDs. Caller should parse by decoding.
             'print(cancelled, flush=True)',
         ]
-        # TODO(zhwu): Backward compatibility, remove after 0.9.0.
+        # TODO(zhwu): Backward compatibility, remove after 0.12.0.
         if user_hash is None:
             code = [
                 (f'cancelled = job_lib.cancel_jobs_encoded_results('