modal 1.0.6.dev14__tar.gz → 1.0.6.dev16__tar.gz

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.0.6.dev14
+Version: 1.0.6.dev16
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/modal/_container_entrypoint.py

@@ -15,7 +15,6 @@ if telemetry_socket:
     instrument_imports(telemetry_socket)
 
 import asyncio
-import concurrent.futures
 import inspect
 import queue
 import signal
@@ -49,7 +48,6 @@ from ._runtime.container_io_manager import (
     ContainerIOManager,
     IOContext,
     UserException,
-    _ContainerIOManager,
 )
 
 if TYPE_CHECKING:
@@ -198,21 +196,16 @@ def call_function(
 
                 # Send up to this many outputs at a time.
                 generator_queue: asyncio.Queue[Any] = await container_io_manager._queue_create.aio(1024)
-                generator_output_task = asyncio.create_task(
-                    container_io_manager.generator_output_task.aio(
-                        function_call_ids[0],
-                        io_context.finalized_function.data_format,
-                        generator_queue,
-                    )
-                )
-
-                item_count = 0
-                async for value in res:
-                    await container_io_manager._queue_put.aio(generator_queue, value)
-                    item_count += 1
+                async with container_io_manager.generator_output_sender(
+                    function_call_ids[0],
+                    io_context.finalized_function.data_format,
+                    generator_queue,
+                ):
+                    item_count = 0
+                    async for value in res:
+                        await container_io_manager._queue_put.aio(generator_queue, value)
+                        item_count += 1
 
-                await container_io_manager._queue_put.aio(generator_queue, _ContainerIOManager._GENERATOR_STOP_SENTINEL)
-                await generator_output_task  # Wait to finish sending generator outputs.
                 message = api_pb2.GeneratorDone(items_total=item_count)
                 await container_io_manager.push_outputs.aio(
                     io_context,
@@ -249,20 +242,17 @@ def call_function(
 
                 # Send up to this many outputs at a time.
                 generator_queue: asyncio.Queue[Any] = container_io_manager._queue_create(1024)
-                generator_output_task: concurrent.futures.Future = container_io_manager.generator_output_task(  # type: ignore
+
+                with container_io_manager.generator_output_sender(
                     function_call_ids[0],
                     io_context.finalized_function.data_format,
                     generator_queue,
-                    _future=True,  # type: ignore  # Synchronicity magic to return a future.
-                )
-
-                item_count = 0
-                for value in res:
-                    container_io_manager._queue_put(generator_queue, value)
-                    item_count += 1
+                ):
+                    item_count = 0
+                    for value in res:
+                        container_io_manager._queue_put(generator_queue, value)
+                        item_count += 1
 
-                container_io_manager._queue_put(generator_queue, _ContainerIOManager._GENERATOR_STOP_SENTINEL)
-                generator_output_task.result()  # Wait to finish sending generator outputs.
                 message = api_pb2.GeneratorDone(items_total=item_count)
                 container_io_manager.push_outputs(io_context, started_at, message, api_pb2.DATA_FORMAT_GENERATOR_DONE)
             else:

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/modal/_runtime/container_io_manager.py

@@ -290,7 +290,6 @@ class _ContainerIOManager:
 
     _client: _Client
 
-    _GENERATOR_STOP_SENTINEL: ClassVar[Sentinel] = Sentinel()
     _singleton: ClassVar[Optional["_ContainerIOManager"]] = None
 
     def _init(self, container_args: api_pb2.ContainerArguments, client: _Client):
@@ -508,33 +507,47 @@ class _ContainerIOManager:
         req = api_pb2.FunctionCallPutDataRequest(function_call_id=function_call_id, data_chunks=data_chunks)
         await retry_transient_errors(self._client.stub.FunctionCallPutDataOut, req)
 
-    async def generator_output_task(self, function_call_id: str, data_format: int, message_rx: asyncio.Queue) -> None:
-        """Task that feeds generator outputs into a function call's `data_out` stream."""
-        index = 1
-        received_sentinel = False
-        while not received_sentinel:
-            message = await message_rx.get()
-            if message is self._GENERATOR_STOP_SENTINEL:
-                break
-            # ASGI 'http.response.start' and 'http.response.body' msgs are observed to be separated by 1ms.
-            # If we don't sleep here for 1ms we end up with an extra call to .put_data_out().
-            if index == 1:
-                await asyncio.sleep(0.001)
-            serialized_messages = [serialize_data_format(message, data_format)]
-            total_size = len(serialized_messages[0]) + 512
-            while total_size < 16 * 1024 * 1024:  # 16 MiB, maximum size in a single message
-                try:
-                    message = message_rx.get_nowait()
-                except asyncio.QueueEmpty:
-                    break
-                if message is self._GENERATOR_STOP_SENTINEL:
-                    received_sentinel = True
+    @asynccontextmanager
+    async def generator_output_sender(
+        self, function_call_id: str, data_format: int, message_rx: asyncio.Queue
+    ) -> AsyncGenerator[None, None]:
+        """Runs background task that feeds generator outputs into a function call's `data_out` stream."""
+        GENERATOR_STOP_SENTINEL = Sentinel()
+
+        async def generator_output_task():
+            index = 1
+            received_sentinel = False
+            while not received_sentinel:
+                message = await message_rx.get()
+                if message is GENERATOR_STOP_SENTINEL:
                     break
-                else:
-                    serialized_messages.append(serialize_data_format(message, data_format))
-                    total_size += len(serialized_messages[-1]) + 512  # 512 bytes for estimated framing overhead
-            await self.put_data_out(function_call_id, index, data_format, serialized_messages)
-            index += len(serialized_messages)
+                # ASGI 'http.response.start' and 'http.response.body' msgs are observed to be separated by 1ms.
+                # If we don't sleep here for 1ms we end up with an extra call to .put_data_out().
+                if index == 1:
+                    await asyncio.sleep(0.001)
+                serialized_messages = [serialize_data_format(message, data_format)]
+                total_size = len(serialized_messages[0]) + 512
+                while total_size < 16 * 1024 * 1024:  # 16 MiB, maximum size in a single message
+                    try:
+                        message = message_rx.get_nowait()
+                    except asyncio.QueueEmpty:
+                        break
+                    if message is GENERATOR_STOP_SENTINEL:
+                        received_sentinel = True
+                        break
+                    else:
+                        serialized_messages.append(serialize_data_format(message, data_format))
+                        total_size += len(serialized_messages[-1]) + 512  # 512 bytes for estimated framing overhead
+                await self.put_data_out(function_call_id, index, data_format, serialized_messages)
+                index += len(serialized_messages)
+
+        task = asyncio.create_task(generator_output_task())
+        try:
+            yield
+        finally:
+            # gracefully stop the task after all current inputs have been sent
+            await message_rx.put(GENERATOR_STOP_SENTINEL)
+            await task
 
     async def _queue_create(self, size: int) -> asyncio.Queue:
         """Create a queue, on the synchronicity event loop (needed on Python 3.8 and 3.9)."""

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/modal/_runtime/container_io_manager.pyi

@@ -106,7 +106,6 @@ class _ContainerIOManager:
     _is_interactivity_enabled: bool
     _fetching_inputs: bool
     _client: modal.client._Client
-    _GENERATOR_STOP_SENTINEL: typing.ClassVar[Sentinel]
     _singleton: typing.ClassVar[typing.Optional[_ContainerIOManager]]
 
     def _init(self, container_args: modal_proto.api_pb2.ContainerArguments, client: modal.client._Client): ...
@@ -148,10 +147,10 @@ class _ContainerIOManager:
         """
         ...
 
-    async def generator_output_task(
+    def generator_output_sender(
         self, function_call_id: str, data_format: int, message_rx: asyncio.queues.Queue
-    ) -> None:
-        """Task that feeds generator outputs into a function call's `data_out` stream."""
+    ) -> typing.AsyncContextManager[None]:
+        """Runs background task that feeds generator outputs into a function call's `data_out` stream."""
         ...
 
     async def _queue_create(self, size: int) -> asyncio.queues.Queue:
@@ -268,7 +267,6 @@ class ContainerIOManager:
     _is_interactivity_enabled: bool
     _fetching_inputs: bool
     _client: modal.client.Client
-    _GENERATOR_STOP_SENTINEL: typing.ClassVar[Sentinel]
     _singleton: typing.ClassVar[typing.Optional[ContainerIOManager]]
 
     def __init__(self, /, *args, **kwargs):
@@ -367,16 +365,20 @@ class ContainerIOManager:
 
     put_data_out: __put_data_out_spec[typing_extensions.Self]
 
-    class __generator_output_task_spec(typing_extensions.Protocol[SUPERSELF]):
-        def __call__(self, /, function_call_id: str, data_format: int, message_rx: asyncio.queues.Queue) -> None:
-            """Task that feeds generator outputs into a function call's `data_out` stream."""
+    class __generator_output_sender_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(
+            self, /, function_call_id: str, data_format: int, message_rx: asyncio.queues.Queue
+        ) -> synchronicity.combined_types.AsyncAndBlockingContextManager[None]:
+            """Runs background task that feeds generator outputs into a function call's `data_out` stream."""
             ...
 
-        async def aio(self, /, function_call_id: str, data_format: int, message_rx: asyncio.queues.Queue) -> None:
-            """Task that feeds generator outputs into a function call's `data_out` stream."""
+        def aio(
+            self, /, function_call_id: str, data_format: int, message_rx: asyncio.queues.Queue
+        ) -> typing.AsyncContextManager[None]:
+            """Runs background task that feeds generator outputs into a function call's `data_out` stream."""
             ...
 
-    generator_output_task: __generator_output_task_spec[typing_extensions.Self]
+    generator_output_sender: __generator_output_sender_spec[typing_extensions.Self]
 
     class ___queue_create_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, /, size: int) -> asyncio.queues.Queue:

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/modal/_utils/blob_utils.py

@@ -188,6 +188,14 @@ def get_content_length(data: BinaryIO) -> int:
     return content_length - pos
 
 
+async def _measure_endpoint_latency(item: str) -> int:
+    latency_ms = 0
+    t0 = time.monotonic_ns()
+    async with ClientSessionRegistry.get_session().head(item) as _:
+        latency_ms = (time.monotonic_ns() - t0) // 1_000_000
+    return latency_ms
+
+
 async def _blob_upload_with_fallback(items, blob_ids: list[str], callback) -> tuple[str, bool, int]:
     r2_latency_ms = 0
     r2_failed = False
@@ -197,10 +205,14 @@ async def _blob_upload_with_fallback(items, blob_ids: list[str], callback) -> tu
         if idx == 0 and len(items) > 1 and random.random() > HEALTHY_R2_UPLOAD_PERCENTAGE:
             continue
         try:
-            init_time = time.monotonic_ns()
-            await callback(item)
             if blob_id.endswith(":r2"):
-                r2_latency_ms = (time.monotonic_ns() - init_time) // 1_000_000
+                # measure the time it takes to contact the bucket endpoint
+                r2_latency_ms, _ = await asyncio.gather(
+                    _measure_endpoint_latency(item),
+                    callback(item),
+                )
+            else:
+                await callback(item)
             return blob_id, r2_failed, r2_latency_ms
         except Exception as _:
             if blob_id.endswith(":r2"):

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/modal/client.pyi

@@ -31,7 +31,7 @@ class _Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.0.6.dev14",
+        version: str = "1.0.6.dev16",
     ):
         """mdmd:hidden
         The Modal client object is not intended to be instantiated directly by users.
@@ -160,7 +160,7 @@ class Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.0.6.dev14",
+        version: str = "1.0.6.dev16",
     ):
         """mdmd:hidden
         The Modal client object is not intended to be instantiated directly by users.

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/modal/functions.pyi

@@ -428,7 +428,7 @@ class Function(
 
     _call_generator: ___call_generator_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:
             """Calls the function remotely, executing it with the given arguments and returning the execution's result."""
             ...
@@ -437,7 +437,7 @@ class Function(
             """Calls the function remotely, executing it with the given arguments and returning the execution's result."""
             ...
 
-    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]:
@@ -464,7 +464,7 @@ class Function(
         """
         ...
 
-    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]:
             """[Experimental] Calls the function with the given arguments, without waiting for the results.
 
@@ -488,7 +488,7 @@ class Function(
             ...
 
     _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_map_inner_spec(typing_extensions.Protocol[P_INNER, SUPERSELF]):
@@ -497,7 +497,7 @@ class Function(
 
     _spawn_map_inner: ___spawn_map_inner_spec[modal._functions.P, 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]:
             """Calls the function with the given arguments, without waiting for the results.
 
@@ -518,7 +518,7 @@ class Function(
             """
             ...
 
-    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]:
         """Return the inner Python object wrapped by this Modal Function."""

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/modal.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.0.6.dev14
+Version: 1.0.6.dev16
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.0.6.dev14 → modal-1.0.6.dev16}/modal_version/__init__.py

@@ -1,4 +1,4 @@
 # Copyright Modal Labs 2025
 """Supplies the current version of the modal client library."""
 
-__version__ = "1.0.6.dev14"
+__version__ = "1.0.6.dev16"