modal 0.74.45__tar.gz → 0.74.47__tar.gz

{modal-0.74.45 → modal-0.74.47}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 0.74.45
+Version: 0.74.47
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-0.74.45 → modal-0.74.47}/modal/_functions.py

@@ -1049,21 +1049,69 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
         fun._spec = self._spec  # TODO (elias): fix - this is incorrect when using with_options
         return fun
 
+    @live_method
+    async def update_autoscaler(
+        self,
+        *,
+        min_containers: Optional[int] = None,
+        max_containers: Optional[int] = None,
+        buffer_containers: Optional[int] = None,
+        scaledown_window: Optional[int] = None,
+    ) -> None:
+        """Override the current autoscaler behavior for this Function.
+
+        Unspecified parameters will retain their current value, i.e. either the static value
+        from the function decorator, or an override value from a previous call to this method.
+
+        Subsequent deployments of the App containing this Function will reset the autoscaler back to
+        its static configuration.
+
+        Examples:
+
+        ```python notest
+        f = modal.Function.from_name("my-app", "function")
+
+        # Always have at least 2 containers running, with an extra buffer when the Function is active
+        f.update_autoscaler(min_containers=2, buffer_containers=1)
+
+        # Limit this Function to avoid spinning up more than 5 containers
+        f.update_autoscaler(max_containers=5)
+
+        # Extend the scaledown window to increase the amount of time that idle containers stay alive
+        f.update_autoscaler(scaledown_window=300)
+
+        ```
+
+        """
+        if self._is_method:
+            raise InvalidError("Cannot call .update_autoscaler() on a method. Call it on the class instance instead.")
+
+        settings = api_pb2.AutoscalerSettings(
+            min_containers=min_containers,
+            max_containers=max_containers,
+            buffer_containers=buffer_containers,
+            scaledown_window=scaledown_window,
+        )
+        request = api_pb2.FunctionUpdateSchedulingParamsRequest(function_id=self.object_id, settings=settings)
+        await retry_transient_errors(self.client.stub.FunctionUpdateSchedulingParams, request)
+
+        # One idea would be for FunctionUpdateScheduleParams to return the current (coalesced) settings
+        # and then we could return them here (would need some ad hoc dataclass, which I don't love)
+
     @live_method
     async def keep_warm(self, warm_pool_size: int) -> None:
-        """Set the warm pool size for the function.
+        """Set the warm pool size for the Function.
 
-        Please exercise care when using this advanced feature!
-        Setting and forgetting a warm pool on functions can lead to increased costs.
+        DEPRECATED: Please adapt your code to use the more general `update_autoscaler` method instead:
 
         ```python notest
-        # Usage on a regular function.
         f = modal.Function.from_name("my-app", "function")
+
+        # Old pattern (deprecated)
         f.keep_warm(2)
 
-        # Usage on a parametrized function.
-        Model = modal.Cls.from_name("my-app", "Model")
-        Model("fine-tuned-model").keep_warm(2)  # note that this applies to the class instance, not a method
+        # New pattern
+        f.update_autoscaler(min_containers=2)
         ```
         """
         if self._is_method:
@@ -1077,10 +1125,14 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
             """
                 )
             )
-        request = api_pb2.FunctionUpdateSchedulingParamsRequest(
-            function_id=self.object_id, warm_pool_size_override=warm_pool_size
+
+        deprecation_warning(
+            (2025, 5, 5),
+            "The .keep_warm() method has been deprecated in favor of the more general "
+            ".update_autoscaler(min_containers=...) method.",
+            show_source=True,
         )
-        await retry_transient_errors(self.client.stub.FunctionUpdateSchedulingParams, request)
+        await self.update_autoscaler(min_containers=warm_pool_size)
 
     @classmethod
     def _from_name(cls, app_name: str, name: str, namespace, environment_name: Optional[str]):

{modal-0.74.45 → modal-0.74.47}/modal/_utils/grpc_utils.py

@@ -118,7 +118,16 @@ def create_channel(
 
         logger.debug(f"Sending request to {event.method_name}")
 
+    async def recv_trailing_metadata(trailing_metadata: grpclib.events.RecvTrailingMetadata) -> None:
+        # If we receive an auth token from the server, include it in all future requests.
+        # TODO(nathan): This isn't perfect because the metadata isn't propagated when the
+        # process is forked and a new channel is created. This is OK for now since this
+        # token is only used by the experimental input plane
+        if token := trailing_metadata.metadata.get("x-modal-auth-token"):
+            metadata["x-modal-auth-token"] = str(token)
+
     grpclib.events.listen(channel, grpclib.events.SendRequest, send_request)
+    grpclib.events.listen(channel, grpclib.events.RecvTrailingMetadata, recv_trailing_metadata)
 
     return channel
 

{modal-0.74.45 → modal-0.74.47}/modal/client.pyi

@@ -27,7 +27,7 @@ class _Client:
     _snapshotted: bool
 
     def __init__(
-        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.74.45"
+        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.74.47"
     ): ...
     def is_closed(self) -> bool: ...
     @property
@@ -85,7 +85,7 @@ class Client:
     _snapshotted: bool
 
     def __init__(
-        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.74.45"
+        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.74.47"
     ): ...
     def is_closed(self) -> bool: ...
     @property

{modal-0.74.45 → modal-0.74.47}/modal/cls.py

@@ -231,22 +231,70 @@ class _Obj:
         user_cls_instance._modal_functions = instance_methods
         return user_cls_instance
 
+    async def update_autoscaler(
+        self,
+        *,
+        min_containers: Optional[int] = None,
+        max_containers: Optional[int] = None,
+        scaledown_window: Optional[int] = None,
+        buffer_containers: Optional[int] = None,
+    ) -> None:
+        """Override the current autoscaler behavior for this Cls instance.
+
+        Unspecified parameters will retain their current value, i.e. either the static value
+        from the function decorator, or an override value from a previous call to this method.
+
+        Subsequent deployments of the App containing this Cls will reset the autoscaler back to
+        its static configuration.
+
+        Note: When calling this method on a Cls that is defined locally, static type checkers will
+        issue an error, because the object will appear to have the user-defined type.
+
+        Examples:
+
+        ```python notest
+        Model = modal.Cls.from_name("my-app", "Model")
+        model = Model()  # This method is called on an *instance* of the class
+
+        # Always have at least 2 containers running, with an extra buffer when the Function is active
+        model.update_autoscaler(min_containers=2, buffer_containers=1)
+
+        # Limit this Function to avoid spinning up more than 5 containers
+        f.update_autoscaler(max_containers=5)
+        ```
+
+        """
+        return await self._cached_service_function().update_autoscaler(
+            min_containers=min_containers,
+            max_containers=max_containers,
+            scaledown_window=scaledown_window,
+            buffer_containers=buffer_containers,
+        )
+
     async def keep_warm(self, warm_pool_size: int) -> None:
         """Set the warm pool size for the class containers
 
-        Please exercise care when using this advanced feature!
-        Setting and forgetting a warm pool on functions can lead to increased costs.
-
-        Note that all Modal methods and web endpoints of a class share the same set
-        of containers and the warm_pool_size affects that common container pool.
+        DEPRECATED: Please adapt your code to use the more general `update_autoscaler` method instead:
 
         ```python notest
-        # Usage on a parametrized function.
         Model = modal.Cls.from_name("my-app", "Model")
-        Model("fine-tuned-model").keep_warm(2)
+        model = Model()  # This method is called on an *instance* of the class
+
+        # Old pattern (deprecated)
+        model.keep_warm(2)
+
+        # New pattern
+        model.update_autoscaler(min_containers=2)
         ```
+
         """
-        await self._cached_service_function().keep_warm(warm_pool_size)
+        deprecation_warning(
+            (2025, 5, 5),
+            "The .keep_warm() method has been deprecated in favor of the more general "
+            ".update_autoscaler(min_containers=...) method.",
+            show_source=True,
+        )
+        await self._cached_service_function().update_autoscaler(min_containers=warm_pool_size)
 
     def _cached_user_cls_instance(self):
         """Get or construct the local object

{modal-0.74.45 → modal-0.74.47}/modal/cls.pyi

@@ -65,6 +65,14 @@ class _Obj:
     def _cached_service_function(self) -> modal._functions._Function: ...
     def _get_parameter_values(self) -> dict[str, typing.Any]: ...
     def _new_user_cls_instance(self): ...
+    async def update_autoscaler(
+        self,
+        *,
+        min_containers: typing.Optional[int] = None,
+        max_containers: typing.Optional[int] = None,
+        scaledown_window: typing.Optional[int] = None,
+        buffer_containers: typing.Optional[int] = None,
+    ) -> None: ...
     async def keep_warm(self, warm_pool_size: int) -> None: ...
     def _cached_user_cls_instance(self): ...
     def _enter(self): ...
@@ -94,6 +102,26 @@ class Obj:
     def _get_parameter_values(self) -> dict[str, typing.Any]: ...
     def _new_user_cls_instance(self): ...
 
+    class __update_autoscaler_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(
+            self,
+            *,
+            min_containers: typing.Optional[int] = None,
+            max_containers: typing.Optional[int] = None,
+            scaledown_window: typing.Optional[int] = None,
+            buffer_containers: typing.Optional[int] = None,
+        ) -> None: ...
+        async def aio(
+            self,
+            *,
+            min_containers: typing.Optional[int] = None,
+            max_containers: typing.Optional[int] = None,
+            scaledown_window: typing.Optional[int] = None,
+            buffer_containers: typing.Optional[int] = None,
+        ) -> None: ...
+
+    update_autoscaler: __update_autoscaler_spec[typing_extensions.Self]
+
     class __keep_warm_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, warm_pool_size: int) -> None: ...
         async def aio(self, warm_pool_size: int) -> None: ...

{modal-0.74.45 → modal-0.74.47}/modal/experimental/__init__.py

@@ -12,6 +12,7 @@ from .._object import _get_environment_name
 from .._partial_function import _clustered
 from .._runtime.container_io_manager import _ContainerIOManager
 from .._utils.async_utils import synchronize_api, synchronizer
+from .._utils.deprecation import deprecation_warning
 from .._utils.grpc_utils import retry_transient_errors
 from ..client import _Client
 from ..cls import _Obj
@@ -179,6 +180,13 @@ async def update_autoscaler(
     may look different (i.e., it may be a standalone function or a method).
 
     """
+    deprecation_warning(
+        (2025, 5, 5),
+        "The modal.experimental.update_autoscaler(...) function is now deprecated in favor of"
+        " a stable `.update_autoscaler(...) method on the corresponding object.",
+        show_source=True,
+    )
+
     settings = api_pb2.AutoscalerSettings(
         min_containers=min_containers,
         max_containers=max_containers,

{modal-0.74.45 → modal-0.74.47}/modal/functions.pyi

@@ -110,6 +110,26 @@ class Function(
         kwargs: dict[str, typing.Any],
     ) -> Function: ...
 
+    class __update_autoscaler_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(
+            self,
+            *,
+            min_containers: typing.Optional[int] = None,
+            max_containers: typing.Optional[int] = None,
+            buffer_containers: typing.Optional[int] = None,
+            scaledown_window: typing.Optional[int] = None,
+        ) -> None: ...
+        async def aio(
+            self,
+            *,
+            min_containers: typing.Optional[int] = None,
+            max_containers: typing.Optional[int] = None,
+            buffer_containers: typing.Optional[int] = None,
+            scaledown_window: typing.Optional[int] = None,
+        ) -> None: ...
+
+    update_autoscaler: __update_autoscaler_spec[typing_extensions.Self]
+
     class __keep_warm_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, warm_pool_size: int) -> None: ...
         async def aio(self, warm_pool_size: int) -> None: ...
@@ -198,11 +218,11 @@ class Function(
 
     _call_generator_nowait: ___call_generator_nowait_spec[typing_extensions.Self]
 
-    class __remote_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
+    class __remote_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
         def __call__(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> ReturnType_INNER: ...
         async def aio(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> ReturnType_INNER: ...
 
-    remote: __remote_spec[modal._functions.ReturnType, modal._functions.P, typing_extensions.Self]
+    remote: __remote_spec[modal._functions.P, modal._functions.ReturnType, typing_extensions.Self]
 
     class __remote_gen_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, *args, **kwargs) -> typing.Generator[typing.Any, None, None]: ...
@@ -217,19 +237,19 @@ class Function(
         self, *args: modal._functions.P.args, **kwargs: modal._functions.P.kwargs
     ) -> modal._functions.OriginalReturnType: ...
 
-    class ___experimental_spawn_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
+    class ___experimental_spawn_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
         def __call__(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]: ...
         async def aio(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]: ...
 
     _experimental_spawn: ___experimental_spawn_spec[
-        modal._functions.ReturnType, modal._functions.P, typing_extensions.Self
+        modal._functions.P, modal._functions.ReturnType, typing_extensions.Self
     ]
 
-    class __spawn_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
+    class __spawn_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
         def __call__(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]: ...
         async def aio(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]: ...
 
-    spawn: __spawn_spec[modal._functions.ReturnType, modal._functions.P, typing_extensions.Self]
+    spawn: __spawn_spec[modal._functions.P, modal._functions.ReturnType, typing_extensions.Self]
 
     def get_raw_f(self) -> collections.abc.Callable[..., typing.Any]: ...
 

{modal-0.74.45 → modal-0.74.47}/modal.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 0.74.45
+Version: 0.74.47
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-0.74.45 → modal-0.74.47}/modal_version/_version_generated.py

@@ -1,4 +1,4 @@
 # Copyright Modal Labs 2025
 
 # Note: Reset this value to -1 whenever you make a minor `0.X` release of the client.
-build_number = 45  # git: fe9ccef
+build_number = 47  # git: 8f5251c