skypilot-nightly 1.0.0.dev20251019__py3-none-any.whl → 1.0.0.dev20251022__py3-none-any.whl

sky/utils/asyncio_utils.py

@@ -2,17 +2,77 @@
 
 import asyncio
 import functools
+from typing import Set
+
+_background_tasks: Set[asyncio.Task] = set()
 
 
 def shield(func):
     """Shield the decorated async function from cancellation.
 
-    Note that filelock.AsyncFileLock is not cancellation safe, thus the
-    function calls filelock.AsyncFileLock must be shielded.
+    If the outter coroutine is cancelled, the inner decorated function
+    will be protected from cancellation by asyncio.shield(). And we will
+    maintain a reference to the the inner task to avoid it get GCed before
+    it is done.
+
+    For example, filelock.AsyncFileLock is not cancellation safe. The
+    following code:
+
+        async def fn_with_lock():
+            async with filelock.AsyncFileLock('lock'):
+                await asyncio.sleep(1)
+
+    is equivalent to:
+
+        # The lock may leak if the cancellation happens in
+        # lock.acquire() or lock.release()
+        async def fn_with_lock():
+            lock = filelock.AsyncFileLock('lock')
+            await lock.acquire()
+            try:
+                await asyncio.sleep(1)
+            finally:
+                await lock.release()
+
+    Shilding the function ensures there is no cancellation will happen in the
+    function, thus the lock will be released properly:
+
+        @shield
+        async def fn_with_lock()
+
+    Note that the resource acquisition and release should usually be protected
+    in one @shield block but not separately, e.g.:
+
+        lock = filelock.AsyncFileLock('lock')
+
+        @shield
+        async def acquire():
+            await lock.acquire()
+
+        @shield
+        async def release():
+            await lock.release()
+
+        async def fn_with_lock():
+            await acquire()
+            try:
+                do_something()
+            finally:
+                await release()
+
+    The above code is not safe because if `fn_with_lock` is cancelled,
+    `acquire()` and `release()` will be executed in the background
+    concurrently and causes race conditions.
     """
 
     @functools.wraps(func)
     async def async_wrapper(*args, **kwargs):
-        return await asyncio.shield(func(*args, **kwargs))
+        task = asyncio.create_task(func(*args, **kwargs))
+        try:
+            return await asyncio.shield(task)
+        except asyncio.CancelledError:
+            _background_tasks.add(task)
+            task.add_done_callback(lambda _: _background_tasks.discard(task))
+            raise
 
     return async_wrapper

sky/utils/db/db_utils.py

@@ -382,21 +382,28 @@ def get_max_connections():
 
 @typing.overload
 def get_engine(
-        db_name: str,
+        db_name: Optional[str],
         async_engine: Literal[False] = False) -> sqlalchemy.engine.Engine:
     ...
 
 
 @typing.overload
-def get_engine(db_name: str,
+def get_engine(db_name: Optional[str],
                async_engine: Literal[True]) -> sqlalchemy_async.AsyncEngine:
     ...
 
 
 def get_engine(
-    db_name: str,
+    db_name: Optional[str],
     async_engine: bool = False
 ) -> Union[sqlalchemy.engine.Engine, sqlalchemy_async.AsyncEngine]:
+    """Get the engine for the given database name.
+
+    Args:
+        db_name: The name of the database. ONLY used for SQLite. On Postgres,
+        we use a single database, which we get from the connection string.
+        async_engine: Whether to return an async engine.
+    """
     conn_string = None
     if os.environ.get(constants.ENV_VAR_IS_SKYPILOT_SERVER) is not None:
         conn_string = os.environ.get(constants.ENV_VAR_DB_CONNECTION_URI)
@@ -429,6 +436,7 @@ def get_engine(
                             max_overflow=0))
             engine = _postgres_engine_cache[conn_string]
     else:
+        assert db_name is not None, 'db_name must be provided for SQLite'
         db_path = os.path.expanduser(f'~/.sky/{db_name}.db')
         pathlib.Path(db_path).parents[0].mkdir(parents=True, exist_ok=True)
         if async_engine:

sky/utils/db/migration_utils.py

@@ -19,15 +19,19 @@ DB_INIT_LOCK_TIMEOUT_SECONDS = 10
 
 GLOBAL_USER_STATE_DB_NAME = 'state_db'
 GLOBAL_USER_STATE_VERSION = '010'
-GLOBAL_USER_STATE_LOCK_PATH = '~/.sky/locks/.state_db.lock'
+GLOBAL_USER_STATE_LOCK_PATH = f'~/.sky/locks/.{GLOBAL_USER_STATE_DB_NAME}.lock'
 
 SPOT_JOBS_DB_NAME = 'spot_jobs_db'
 SPOT_JOBS_VERSION = '003'
-SPOT_JOBS_LOCK_PATH = '~/.sky/locks/.spot_jobs_db.lock'
+SPOT_JOBS_LOCK_PATH = f'~/.sky/locks/.{SPOT_JOBS_DB_NAME}.lock'
 
 SERVE_DB_NAME = 'serve_db'
 SERVE_VERSION = '001'
-SERVE_LOCK_PATH = '~/.sky/locks/.serve_db.lock'
+SERVE_LOCK_PATH = f'~/.sky/locks/.{SERVE_DB_NAME}.lock'
+
+SKYPILOT_CONFIG_DB_NAME = 'sky_config_db'
+SKYPILOT_CONFIG_VERSION = '001'
+SKYPILOT_CONFIG_LOCK_PATH = f'~/.sky/locks/.{SKYPILOT_CONFIG_DB_NAME}.lock'
 
 
 @contextlib.contextmanager

sky/volumes/server/server.py

@@ -25,7 +25,7 @@ async def volume_list(request: fastapi.Request) -> None:
         'env_vars': auth_user.to_env_vars()
     } if auth_user else {}
     request_body = payloads.RequestBody(**auth_user_env_vars_kwargs)
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='volume_list',
         request_body=request_body,
@@ -38,7 +38,7 @@ async def volume_list(request: fastapi.Request) -> None:
 async def volume_delete(request: fastapi.Request,
                         volume_delete_body: payloads.VolumeDeleteBody) -> None:
     """Deletes a volume."""
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='volume_delete',
         request_body=volume_delete_body,
@@ -112,7 +112,7 @@ async def volume_apply(request: fastapi.Request,
             raise fastapi.HTTPException(
                 status_code=400,
                 detail='Runpod network volume is only supported on Runpod')
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='volume_apply',
         request_body=volume_apply_body,

sky/workspaces/server.py

@@ -22,7 +22,7 @@ async def get(request: fastapi.Request) -> None:
     } if auth_user else {}
     request_body = payloads.RequestBody(**auth_user_env_vars_kwargs)
 
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='workspaces.get',
         request_body=request_body,
@@ -35,7 +35,7 @@ async def get(request: fastapi.Request) -> None:
 async def update(request: fastapi.Request,
                  update_workspace_body: payloads.UpdateWorkspaceBody) -> None:
     """Updates a specific workspace configuration."""
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='workspaces.update',
         request_body=update_workspace_body,
@@ -48,7 +48,7 @@ async def update(request: fastapi.Request,
 async def create(request: fastapi.Request,
                  create_workspace_body: payloads.CreateWorkspaceBody) -> None:
     """Creates a new workspace configuration."""
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='workspaces.create',
         request_body=create_workspace_body,
@@ -61,7 +61,7 @@ async def create(request: fastapi.Request,
 async def delete(request: fastapi.Request,
                  delete_workspace_body: payloads.DeleteWorkspaceBody) -> None:
     """Deletes a workspace configuration."""
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='workspaces.delete',
         request_body=delete_workspace_body,
@@ -78,7 +78,7 @@ async def get_config(request: fastapi.Request) -> None:
         'env_vars': auth_user.to_env_vars()
     } if auth_user else {}
     get_config_body = payloads.GetConfigBody(**auth_user_env_vars_kwargs)
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='workspaces.get_config',
         request_body=get_config_body,
@@ -91,7 +91,7 @@ async def get_config(request: fastapi.Request) -> None:
 async def update_config(request: fastapi.Request,
                         update_config_body: payloads.UpdateConfigBody) -> None:
     """Updates the entire SkyPilot configuration."""
-    executor.schedule_request(
+    await executor.schedule_request_async(
         request_id=request.state.request_id,
         request_name='workspaces.update_config',
         request_body=update_config_body,

{skypilot_nightly-1.0.0.dev20251019.dist-info → skypilot_nightly-1.0.0.dev20251022.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: skypilot-nightly
-Version: 1.0.0.dev20251019
+Version: 1.0.0.dev20251022
 Summary: SkyPilot: Run AI on Any Infra — Unified, Faster, Cheaper.
 Author: SkyPilot Team
 License: Apache 2.0
@@ -371,51 +371,51 @@ Requires-Dist: protobuf<7.0.0,>=5.26.1; extra == "shadeform"
 Requires-Dist: aiosqlite; extra == "shadeform"
 Requires-Dist: greenlet; extra == "shadeform"
 Provides-Extra: all
+Requires-Dist: casbin; extra == "all"
+Requires-Dist: google-api-python-client>=2.69.0; extra == "all"
 Requires-Dist: ibm-vpc; extra == "all"
+Requires-Dist: google-cloud-storage; extra == "all"
+Requires-Dist: aiohttp; extra == "all"
+Requires-Dist: azure-identity>=1.19.0; extra == "all"
 Requires-Dist: pyvmomi==8.0.1.0.2; extra == "all"
-Requires-Dist: cudo-compute>=0.1.10; extra == "all"
+Requires-Dist: msrestazure; extra == "all"
+Requires-Dist: ibm-platform-services>=0.48.0; extra == "all"
+Requires-Dist: oci; extra == "all"
 Requires-Dist: pyjwt; extra == "all"
-Requires-Dist: ibm-cloud-sdk-core; extra == "all"
-Requires-Dist: grpcio>=1.63.0; extra == "all"
+Requires-Dist: runpod>=1.6.1; extra == "all"
 Requires-Dist: docker; extra == "all"
-Requires-Dist: pyopenssl<24.3.0,>=23.2.0; extra == "all"
-Requires-Dist: sqlalchemy_adapter; extra == "all"
-Requires-Dist: azure-mgmt-compute>=33.0.0; extra == "all"
-Requires-Dist: awscli>=1.27.10; extra == "all"
+Requires-Dist: azure-common; extra == "all"
 Requires-Dist: nebius>=0.2.47; extra == "all"
-Requires-Dist: websockets; extra == "all"
-Requires-Dist: azure-cli>=2.65.0; extra == "all"
-Requires-Dist: casbin; extra == "all"
-Requires-Dist: aiohttp; extra == "all"
-Requires-Dist: anyio; extra == "all"
+Requires-Dist: ibm-cos-sdk; extra == "all"
 Requires-Dist: protobuf<7.0.0,>=5.26.1; extra == "all"
-Requires-Dist: colorama<0.4.5; extra == "all"
-Requires-Dist: google-api-python-client>=2.69.0; extra == "all"
 Requires-Dist: ecsapi>=0.2.0; extra == "all"
-Requires-Dist: azure-common; extra == "all"
-Requires-Dist: ibm-cos-sdk; extra == "all"
-Requires-Dist: azure-storage-blob>=12.23.1; extra == "all"
-Requires-Dist: azure-core>=1.31.0; extra == "all"
-Requires-Dist: botocore>=1.29.10; extra == "all"
-Requires-Dist: tomli; python_version < "3.11" and extra == "all"
-Requires-Dist: kubernetes!=32.0.0,>=20.0.0; extra == "all"
-Requires-Dist: oci; extra == "all"
-Requires-Dist: azure-identity>=1.19.0; extra == "all"
-Requires-Dist: ray[default]>=2.6.1; extra == "all"
-Requires-Dist: msrestazure; extra == "all"
-Requires-Dist: aiosqlite; extra == "all"
-Requires-Dist: pydo>=0.3.0; extra == "all"
-Requires-Dist: python-dateutil; extra == "all"
-Requires-Dist: google-cloud-storage; extra == "all"
-Requires-Dist: greenlet; extra == "all"
-Requires-Dist: runpod>=1.6.1; extra == "all"
+Requires-Dist: anyio; extra == "all"
+Requires-Dist: ibm-cloud-sdk-core; extra == "all"
+Requires-Dist: sqlalchemy_adapter; extra == "all"
 Requires-Dist: boto3>=1.26.1; extra == "all"
+Requires-Dist: pyopenssl<24.3.0,>=23.2.0; extra == "all"
+Requires-Dist: aiosqlite; extra == "all"
 Requires-Dist: vastai-sdk>=0.1.12; extra == "all"
-Requires-Dist: azure-core>=1.24.0; extra == "all"
-Requires-Dist: ibm-platform-services>=0.48.0; extra == "all"
+Requires-Dist: pydo>=0.3.0; extra == "all"
+Requires-Dist: botocore>=1.29.10; extra == "all"
+Requires-Dist: msgraph-sdk; extra == "all"
+Requires-Dist: awscli>=1.27.10; extra == "all"
 Requires-Dist: azure-mgmt-network>=27.0.0; extra == "all"
+Requires-Dist: azure-mgmt-compute>=33.0.0; extra == "all"
+Requires-Dist: grpcio>=1.63.0; extra == "all"
+Requires-Dist: kubernetes!=32.0.0,>=20.0.0; extra == "all"
+Requires-Dist: azure-storage-blob>=12.23.1; extra == "all"
+Requires-Dist: colorama<0.4.5; extra == "all"
+Requires-Dist: azure-cli>=2.65.0; extra == "all"
+Requires-Dist: cudo-compute>=0.1.10; extra == "all"
+Requires-Dist: tomli; python_version < "3.11" and extra == "all"
+Requires-Dist: websockets; extra == "all"
+Requires-Dist: azure-core>=1.24.0; extra == "all"
+Requires-Dist: greenlet; extra == "all"
+Requires-Dist: python-dateutil; extra == "all"
+Requires-Dist: ray[default]>=2.6.1; extra == "all"
+Requires-Dist: azure-core>=1.31.0; extra == "all"
 Requires-Dist: passlib; extra == "all"
-Requires-Dist: msgraph-sdk; extra == "all"
 Provides-Extra: remote
 Requires-Dist: grpcio>=1.63.0; extra == "remote"
 Requires-Dist: protobuf<7.0.0,>=5.26.1; extra == "remote"
@@ -479,6 +479,7 @@ Dynamic: summary
 ----
 
 :fire: *News* :fire:
+- [Jul 2025] Run **RL training for LLMs** with SkyRL on your Kubernetes or clouds: [**example**](./llm/skyrl/)
 - [Oct 2025] Train and serve [Andrej Karpathy's](https://x.com/karpathy/status/1977755427569111362) **nanochat** - the best ChatGPT that $100 can buy: [**example**](./llm/nanochat)
 - [Oct 2025] Run large-scale **LLM training with TorchTitan** on any AI infra: [**example**](./examples/training/torchtitan)
 - [Sep 2025] Scaling AI infrastructure at Abridge - **10x faster development** with SkyPilot: [**blog**](https://blog.skypilot.co/abridge/)
@@ -543,8 +544,7 @@ pip install "skypilot-nightly[kubernetes,aws,gcp,azure,oci,nebius,lambda,runpod,
 </p>
 
 Current supported infra: Kubernetes, AWS, GCP, Azure, OCI, Nebius, Lambda Cloud, RunPod, Fluidstack,
-Cudo, Digital Ocean, Paperspace, Cloudflare, Samsung, IBM, Vast.ai,
-VMware vSphere, Seeweb.
+Cudo, Digital Ocean, Paperspace, Cloudflare, Samsung, IBM, Vast.ai, VMware vSphere, Seeweb, Prime Intellect.
 <p align="center">
   <img alt="SkyPilot" src="https://raw.githubusercontent.com/skypilot-org/skypilot/master/docs/source/images/cloud-logos-light.png" width=85%>
 </p>