modal 1.2.1.dev14__tar.gz → 1.2.1.dev15__tar.gz

{modal-1.2.1.dev14 → modal-1.2.1.dev15}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.2.1.dev14
+Version: 1.2.1.dev15
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.2.1.dev14 → modal-1.2.1.dev15}/modal/client.pyi

@@ -33,7 +33,7 @@ class _Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.2.1.dev14",
+        version: str = "1.2.1.dev15",
     ):
         """mdmd:hidden
         The Modal client object is not intended to be instantiated directly by users.
@@ -164,7 +164,7 @@ class Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.2.1.dev14",
+        version: str = "1.2.1.dev15",
     ):
         """mdmd:hidden
         The Modal client object is not intended to be instantiated directly by users.

{modal-1.2.1.dev14 → modal-1.2.1.dev15}/modal/experimental/flash.py

@@ -321,7 +321,7 @@ class _FlashPrometheusAutoscaler:
 
     async def _compute_target_containers(self, current_replicas: int) -> int:
         """
-        Gets internal metrics from container to autoscale up or down.
+        Gets metrics from container to autoscale up or down.
         """
         containers = await self._get_all_containers()
         if len(containers) > current_replicas:
@@ -334,7 +334,7 @@ class _FlashPrometheusAutoscaler:
         if current_replicas == 0:
             return 1
 
-        # Get metrics based on autoscaler type (prometheus or internal)
+        # Get metrics based on autoscaler type
         sum_metric, n_containers_with_metrics = await self._get_scaling_info(containers)
 
         desired_replicas = self._calculate_desired_replicas(
@@ -406,39 +406,26 @@ class _FlashPrometheusAutoscaler:
         return desired_replicas
 
     async def _get_scaling_info(self, containers) -> tuple[float, int]:
-        """Get metrics using either internal container metrics API or prometheus HTTP endpoints."""
-        if self.metrics_endpoint == "internal":
-            container_metrics_results = await asyncio.gather(
-                *[self._get_container_metrics(container.task_id) for container in containers]
-            )
-            container_metrics_list = []
-            for container_metric in container_metrics_results:
-                if container_metric is None:
-                    continue
-                container_metrics_list.append(getattr(container_metric.metrics, self.target_metric))
-
-            sum_metric = sum(container_metrics_list)
-            n_containers_with_metrics = len(container_metrics_list)
-        else:
-            sum_metric = 0
-            n_containers_with_metrics = 0
-
-            container_metrics_list = await asyncio.gather(
-                *[
-                    self._get_metrics(f"https://{container.host}:{container.port}/{self.metrics_endpoint}")
-                    for container in containers
-                ]
-            )
+        """Get metrics using container exposed metrics endpoints."""
+        sum_metric = 0
+        n_containers_with_metrics = 0
+
+        container_metrics_list = await asyncio.gather(
+            *[
+                self._get_metrics(f"https://{container.host}:{container.port}/{self.metrics_endpoint}")
+                for container in containers
+            ]
+        )
 
-            for container_metrics in container_metrics_list:
-                if (
-                    container_metrics is None
-                    or self.target_metric not in container_metrics
-                    or len(container_metrics[self.target_metric]) == 0
-                ):
-                    continue
-                sum_metric += container_metrics[self.target_metric][0].value
-                n_containers_with_metrics += 1
+        for container_metrics in container_metrics_list:
+            if (
+                container_metrics is None
+                or self.target_metric not in container_metrics
+                or len(container_metrics[self.target_metric]) == 0
+            ):
+                continue
+            sum_metric += container_metrics[self.target_metric][0].value
+            n_containers_with_metrics += 1
 
         return sum_metric, n_containers_with_metrics
 
@@ -474,15 +461,6 @@ class _FlashPrometheusAutoscaler:
 
         return metrics
 
-    async def _get_container_metrics(self, container_id: str) -> Optional[api_pb2.TaskGetAutoscalingMetricsResponse]:
-        req = api_pb2.TaskGetAutoscalingMetricsRequest(task_id=container_id)
-        try:
-            resp = await retry_transient_errors(self.client.stub.TaskGetAutoscalingMetrics, req)
-            return resp
-        except Exception as e:
-            logger.warning(f"[Modal Flash] Error getting metrics for container {container_id}: {e}")
-            return None
-
     async def _get_all_containers(self):
         req = api_pb2.FlashContainerListRequest(function_id=self.fn.object_id)
         resp = await retry_transient_errors(self.client.stub.FlashContainerList, req)
@@ -572,14 +550,10 @@ async def flash_prometheus_autoscaler(
     app_name: str,
     cls_name: str,
     # Endpoint to fetch metrics from. Must be in Prometheus format. Example: "/metrics"
-    # If metrics_endpoint is "internal", we will use containers' internal metrics to autoscale instead.
     metrics_endpoint: str,
     # Target metric to autoscale on. Example: "vllm:num_requests_running"
-    # If metrics_endpoint is "internal", target_metrics options are: [cpu_usage_percent, memory_usage_percent]
     target_metric: str,
     # Target metric value. Example: 25
-    # If metrics_endpoint is "internal", target_metric_value is a percentage value between 0.1 and 1.0 (inclusive),
-    # indicating container's usage of that metric.
     target_metric_value: float,
     min_containers: Optional[int] = None,
     max_containers: Optional[int] = None,

{modal-1.2.1.dev14 → modal-1.2.1.dev15}/modal/experimental/flash.pyi

@@ -1,5 +1,4 @@
 import modal.client
-import modal_proto.api_pb2
 import subprocess
 import typing
 import typing_extensions
@@ -139,7 +138,7 @@ class _FlashPrometheusAutoscaler:
     async def start(self): ...
     async def _run_autoscaler_loop(self): ...
     async def _compute_target_containers(self, current_replicas: int) -> int:
-        """Gets internal metrics from container to autoscale up or down."""
+        """Gets metrics from container to autoscale up or down."""
         ...
 
     def _calculate_desired_replicas(
@@ -154,13 +153,10 @@ class _FlashPrometheusAutoscaler:
         ...
 
     async def _get_scaling_info(self, containers) -> tuple[float, int]:
-        """Get metrics using either internal container metrics API or prometheus HTTP endpoints."""
+        """Get metrics using container exposed metrics endpoints."""
         ...
 
     async def _get_metrics(self, url: str) -> typing.Optional[dict[str, list[typing.Any]]]: ...
-    async def _get_container_metrics(
-        self, container_id: str
-    ) -> typing.Optional[modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse]: ...
     async def _get_all_containers(self): ...
     async def _set_target_slots(self, target_slots: int): ...
     def _make_scaling_decision(
@@ -226,11 +222,11 @@ class FlashPrometheusAutoscaler:
 
     class ___compute_target_containers_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, /, current_replicas: int) -> int:
-            """Gets internal metrics from container to autoscale up or down."""
+            """Gets metrics from container to autoscale up or down."""
             ...
 
         async def aio(self, /, current_replicas: int) -> int:
-            """Gets internal metrics from container to autoscale up or down."""
+            """Gets metrics from container to autoscale up or down."""
             ...
 
     _compute_target_containers: ___compute_target_containers_spec[typing_extensions.Self]
@@ -248,11 +244,11 @@ class FlashPrometheusAutoscaler:
 
     class ___get_scaling_info_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, /, containers) -> tuple[float, int]:
-            """Get metrics using either internal container metrics API or prometheus HTTP endpoints."""
+            """Get metrics using container exposed metrics endpoints."""
             ...
 
         async def aio(self, /, containers) -> tuple[float, int]:
-            """Get metrics using either internal container metrics API or prometheus HTTP endpoints."""
+            """Get metrics using container exposed metrics endpoints."""
             ...
 
     _get_scaling_info: ___get_scaling_info_spec[typing_extensions.Self]
@@ -263,16 +259,6 @@ class FlashPrometheusAutoscaler:
 
     _get_metrics: ___get_metrics_spec[typing_extensions.Self]
 
-    class ___get_container_metrics_spec(typing_extensions.Protocol[SUPERSELF]):
-        def __call__(
-            self, /, container_id: str
-        ) -> typing.Optional[modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse]: ...
-        async def aio(
-            self, /, container_id: str
-        ) -> typing.Optional[modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse]: ...
-
-    _get_container_metrics: ___get_container_metrics_spec[typing_extensions.Self]
-
     class ___get_all_containers_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, /): ...
         async def aio(self, /): ...

{modal-1.2.1.dev14 → modal-1.2.1.dev15}/modal/functions.pyi

@@ -401,7 +401,7 @@ class Function(
 
     _call_generator: ___call_generator_spec[typing_extensions.Self]
 
-    class __remote_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
+    class __remote_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
         def __call__(self, /, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> ReturnType_INNER:
             """Calls the function remotely, executing it with the given arguments and returning the execution's result."""
             ...
@@ -410,7 +410,7 @@ class Function(
             """Calls the function remotely, executing it with the given arguments and returning the execution's result."""
             ...
 
-    remote: __remote_spec[modal._functions.P, modal._functions.ReturnType, typing_extensions.Self]
+    remote: __remote_spec[modal._functions.ReturnType, modal._functions.P, typing_extensions.Self]
 
     class __remote_gen_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, /, *args, **kwargs) -> typing.Generator[typing.Any, None, None]:
@@ -437,7 +437,7 @@ class Function(
         """
         ...
 
-    class ___experimental_spawn_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
+    class ___experimental_spawn_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
         def __call__(self, /, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]:
             """[Experimental] Calls the function with the given arguments, without waiting for the results.
 
@@ -461,7 +461,7 @@ class Function(
             ...
 
     _experimental_spawn: ___experimental_spawn_spec[
-        modal._functions.P, modal._functions.ReturnType, typing_extensions.Self
+        modal._functions.ReturnType, modal._functions.P, typing_extensions.Self
     ]
 
     class ___spawn_map_inner_spec(typing_extensions.Protocol[P_INNER, SUPERSELF]):
@@ -470,7 +470,7 @@ class Function(
 
     _spawn_map_inner: ___spawn_map_inner_spec[modal._functions.P, typing_extensions.Self]
 
-    class __spawn_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
+    class __spawn_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
         def __call__(self, /, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]:
             """Calls the function with the given arguments, without waiting for the results.
 
@@ -491,7 +491,7 @@ class Function(
             """
             ...
 
-    spawn: __spawn_spec[modal._functions.P, modal._functions.ReturnType, typing_extensions.Self]
+    spawn: __spawn_spec[modal._functions.ReturnType, modal._functions.P, typing_extensions.Self]
 
     def get_raw_f(self) -> collections.abc.Callable[..., typing.Any]:
         """Return the inner Python object wrapped by this Modal Function."""

{modal-1.2.1.dev14 → modal-1.2.1.dev15}/modal.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.2.1.dev14
+Version: 1.2.1.dev15
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.2.1.dev14 → modal-1.2.1.dev15}/modal_proto/api.proto

@@ -3143,14 +3143,6 @@ message TaskCurrentInputsResponse {
   repeated string input_ids = 1;
 }
 
-message TaskGetAutoscalingMetricsRequest {
-  string task_id = 1;
-}
-
-message TaskGetAutoscalingMetricsResponse {
-  AutoscalingMetrics metrics = 1;
-}
-
 // Used to get a JWT and URL for direct access to a task command router
 // running on the modal-worker, so the client can issue exec commands (and other
 // operations as they become available) directly to the worker.
@@ -3778,7 +3770,6 @@ service ModalClient {
   // Tasks
   rpc TaskClusterHello(TaskClusterHelloRequest) returns (TaskClusterHelloResponse);
   rpc TaskCurrentInputs(google.protobuf.Empty) returns (TaskCurrentInputsResponse);
-  rpc TaskGetAutoscalingMetrics(TaskGetAutoscalingMetricsRequest) returns (TaskGetAutoscalingMetricsResponse); // Used for flash autoscaling
   rpc TaskGetCommandRouterAccess(TaskGetCommandRouterAccessRequest) returns (TaskGetCommandRouterAccessResponse);
   rpc TaskList(TaskListRequest) returns (TaskListResponse);
   rpc TaskResult(TaskResultRequest) returns (google.protobuf.Empty);

{modal-1.2.1.dev14 → modal-1.2.1.dev15}/modal_proto/api_grpc.py

@@ -610,10 +610,6 @@ class ModalClientBase(abc.ABC):
     async def TaskCurrentInputs(self, stream: 'grpclib.server.Stream[google.protobuf.empty_pb2.Empty, modal_proto.api_pb2.TaskCurrentInputsResponse]') -> None:
         pass
 
-    @abc.abstractmethod
-    async def TaskGetAutoscalingMetrics(self, stream: 'grpclib.server.Stream[modal_proto.api_pb2.TaskGetAutoscalingMetricsRequest, modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse]') -> None:
-        pass
-
     @abc.abstractmethod
     async def TaskGetCommandRouterAccess(self, stream: 'grpclib.server.Stream[modal_proto.api_pb2.TaskGetCommandRouterAccessRequest, modal_proto.api_pb2.TaskGetCommandRouterAccessResponse]') -> None:
         pass
@@ -1608,12 +1604,6 @@ class ModalClientBase(abc.ABC):
                 google.protobuf.empty_pb2.Empty,
                 modal_proto.api_pb2.TaskCurrentInputsResponse,
             ),
-            '/modal.client.ModalClient/TaskGetAutoscalingMetrics': grpclib.const.Handler(
-                self.TaskGetAutoscalingMetrics,
-                grpclib.const.Cardinality.UNARY_UNARY,
-                modal_proto.api_pb2.TaskGetAutoscalingMetricsRequest,
-                modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse,
-            ),
             '/modal.client.ModalClient/TaskGetCommandRouterAccess': grpclib.const.Handler(
                 self.TaskGetCommandRouterAccess,
                 grpclib.const.Cardinality.UNARY_UNARY,
@@ -2664,12 +2654,6 @@ class ModalClientStub:
             google.protobuf.empty_pb2.Empty,
             modal_proto.api_pb2.TaskCurrentInputsResponse,
         )
-        self.TaskGetAutoscalingMetrics = grpclib.client.UnaryUnaryMethod(
-            channel,
-            '/modal.client.ModalClient/TaskGetAutoscalingMetrics',
-            modal_proto.api_pb2.TaskGetAutoscalingMetricsRequest,
-            modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse,
-        )
         self.TaskGetCommandRouterAccess = grpclib.client.UnaryUnaryMethod(
             channel,
             '/modal.client.ModalClient/TaskGetCommandRouterAccess',