Release 0.9.3 (#5598)

* Release 0.9.3

* Pin click version for correct argument parsing (#5597)

* Pin click version for correct argument parsing

Signed-off-by: Aylei <rayingecho@gmail.com>

* Loosen constrain

Signed-off-by: Aylei <rayingecho@gmail.com>

* Correct issue link

Signed-off-by: Aylei <rayingecho@gmail.com>

---------

Signed-off-by: Aylei <rayingecho@gmail.com>

* Cherry-pick: Fix worker exits on duplicate scheduling (#5611) (#5613)

Fix worker exits on duplicate scheduling (#5611)

Signed-off-by: Aylei <rayingecho@gmail.com>

---------

Signed-off-by: Aylei <rayingecho@gmail.com>
Co-authored-by: GitHub Action <action@github.com>
Co-authored-by: Aylei <rayingecho@gmail.com>

Author: 3 people

Dockerfile

@@ -26,7 +26,7 @@ RUN conda install -c conda-forge google-cloud-sdk && \
     # Install uv and skypilot
     curl -LsSf https://astral.sh/uv/install.sh | sh && \
     ~/.local/bin/uv pip install --prerelease allow azure-cli --system && \
-    ~/.local/bin/uv pip install skypilot-nightly[all] --system && \
+    ~/.local/bin/uv pip install skypilot[all]==0.9.3 --system && \
     # Cleanup all caches to reduce the image size
     conda clean -afy && \
     ~/.local/bin/uv cache clean && \

charts/skypilot/values.yaml

@@ -1,5 +1,5 @@
 apiService:
-  image: berkeleyskypilot/skypilot:0.9.2
+  image: berkeleyskypilot/skypilot:0.9.3
   # Number of replicas to deploy - replicas > 1 is not well tested, and requires
   # a PVC that supports ReadWriteMany (see accessMode in storage section below).
   replicas: 1

sky/__init__.py

@@ -35,7 +35,7 @@ def _get_git_commit():
 
 
 __commit__ = _get_git_commit()
-__version__ = '1.0.0-dev0'
+__version__ = '0.9.3'
 __root_dir__ = os.path.dirname(os.path.abspath(__file__))
 
 

sky/exceptions.py

@@ -568,3 +568,8 @@ def from_managed_job_status(
 
         # Should not hit this case, but included to avoid errors
         return cls.FAILED
+
+
+class RequestAlreadyExistsError(Exception):
+    """Raised when a request is already exists."""
+    pass

sky/server/requests/executor.py

@@ -34,6 +34,7 @@
 
 import setproctitle
 
+from sky import exceptions
 from sky import global_user_state
 from sky import models
 from sky import sky_logging
@@ -446,7 +447,8 @@ def prepare_request(
                                    cluster_name=request_cluster_name)
 
     if not api_requests.create_if_not_exists(request):
-        raise RuntimeError(f'Request {request_id} already exists.')
+        raise exceptions.RequestAlreadyExistsError(
+            f'Request {request_id} already exists.')
 
     request.log_path.touch()
     return request

sky/server/server.py

@@ -142,14 +142,19 @@ async def lifespan(app: fastapi.FastAPI):  # pylint: disable=redefined-outer-nam
     del app  # unused
     # Startup: Run background tasks
     for event in requests_lib.INTERNAL_REQUEST_DAEMONS:
-        executor.schedule_request(
-            request_id=event.id,
-            request_name=event.name,
-            request_body=payloads.RequestBody(),
-            func=event.event_fn,
-            schedule_type=requests_lib.ScheduleType.SHORT,
-            is_skypilot_system=True,
-        )
+        try:
+            executor.schedule_request(
+                request_id=event.id,
+                request_name=event.name,
+                request_body=payloads.RequestBody(),
+                func=event.event_fn,
+                schedule_type=requests_lib.ScheduleType.SHORT,
+                is_skypilot_system=True,
+            )
+        except exceptions.RequestAlreadyExistsError:
+            # Lifespan will be executed in each uvicorn worker process, we
+            # can safely ignore the error if the task is already scheduled.
+            logger.debug(f'Request {event.id} already exists.')
     asyncio.create_task(cleanup_upload_ids())
     yield
     # Shutdown: Add any cleanup code here if needed

sky/setup_files/dependencies.py

@@ -12,7 +12,10 @@
     'wheel<0.46.0',  # https://github.com/skypilot-org/skypilot/issues/5153
     'cachetools',
     # NOTE: ray requires click>=7.0.
-    'click >= 7.0',
+    # click 8.2.0 has a bug in parsing the command line arguments:
+    # https://github.com/pallets/click/issues/2894
+    # TODO(aylei): remove this once the bug is fixed in click.
+    'click >= 7.0, < 8.2.0',
     'colorama',
     'cryptography',
     # Jinja has a bug in older versions because of the lack of pinning