modAL 1.2.7.dev6__tar.gz → 1.2.7.dev9__tar.gz

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.2.7.dev6
+Version: 1.2.7.dev9
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/_functions.py

@@ -694,7 +694,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
         # Experimental: Clustered functions
         cluster_size: Optional[int] = None,
         rdma: Optional[bool] = None,
-        max_inputs: Optional[int] = None,
+        single_use_containers: bool = False,
         ephemeral_disk: Optional[int] = None,
         include_source: bool = True,
         experimental_options: Optional[dict[str, str]] = None,
@@ -810,14 +810,6 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
                 if arg.default is not inspect.Parameter.empty:
                     raise InvalidError(f"Modal batched function {func_name} does not accept default arguments.")
 
-        if max_inputs is not None:
-            if not isinstance(max_inputs, int):
-                raise InvalidError(f"`max_inputs` must be an int, not {type(max_inputs).__name__}")
-            if max_inputs <= 0:
-                raise InvalidError("`max_inputs` must be positive")
-            if max_inputs > 1:
-                raise InvalidError("Only `max_inputs=1` is currently supported")
-
         # Validate volumes
         validated_volumes = validate_volumes(volumes)
         cloud_bucket_mounts = [(k, v) for k, v in validated_volumes if isinstance(v, _CloudBucketMount)]
@@ -1024,7 +1016,8 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
                     object_dependencies=object_dependencies,
                     block_network=block_network,
                     untrusted=restrict_modal_access,
-                    max_inputs=max_inputs or 0,
+                    single_use_containers=single_use_containers,
+                    max_inputs=int(single_use_containers),  # TODO(michael) remove after worker rollover
                     cloud_bucket_mounts=cloud_bucket_mounts_to_proto(cloud_bucket_mounts),
                     scheduler_placement=scheduler_placement,
                     is_class=info.is_service_class(),

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/_runtime/container_io_manager.py

@@ -845,8 +845,9 @@ class _ContainerIOManager:
                     yield inputs
                     yielded = True
 
-                    # We only support max_inputs = 1 at the moment
-                    if final_input_received or self.function_def.max_inputs == 1:
+                    # TODO(michael): Remove use of max_inputs after worker rollover
+                    single_use_container = self.function_def.single_use_containers or self.function_def.max_inputs == 1
+                    if final_input_received or single_use_container:
                         return
             finally:
                 if not yielded:

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/_utils/blob_utils.py

@@ -152,12 +152,13 @@ async def perform_multipart_upload(
     part_etags = await TaskContext.gather(*upload_coros)
 
     # The body of the complete_multipart_upload command needs some data in xml format:
-    completion_body = "<CompleteMultipartUpload>\n"
+    completion_parts = ["<CompleteMultipartUpload>"]
     for part_number, etag in enumerate(part_etags, 1):
-        completion_body += f"""<Part>\n<PartNumber>{part_number}</PartNumber>\n<ETag>"{etag}"</ETag>\n</Part>\n"""
-    completion_body += "</CompleteMultipartUpload>"
+        completion_parts.append(f"""<Part>\n<PartNumber>{part_number}</PartNumber>\n<ETag>"{etag}"</ETag>\n</Part>""")
+    completion_parts.append("</CompleteMultipartUpload>")
+    completion_body = "\n".join(completion_parts)
 
-    # etag of combined object should be md5 hex of concatendated md5 *bytes* from parts + `-{num_parts}`
+    # etag of combined object should be md5 hex of concatenated md5 *bytes* from parts + `-{num_parts}`
     bin_hash_parts = [bytes.fromhex(etag) for etag in part_etags]
 
     expected_multipart_etag = hashlib.md5(b"".join(bin_hash_parts)).hexdigest() + f"-{len(part_etags)}"

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/app.py

@@ -725,9 +725,7 @@ class _App:
         enable_memory_snapshot: bool = False,  # Enable memory checkpointing for faster cold starts.
         block_network: bool = False,  # Whether to block network access
         restrict_modal_access: bool = False,  # Whether to allow this function access to other Modal resources
-        # Maximum number of inputs a container should handle before shutting down.
-        # With `max_inputs = 1`, containers will be single-use.
-        max_inputs: Optional[int] = None,
+        single_use_containers: bool = False,  # When True, containers will shut down after handling a single input
         i6pn: Optional[bool] = None,  # Whether to enable IPv6 container networking within the region.
         # Whether the file or directory containing the Function's source should automatically be included
         # in the container. When unset, falls back to the App-level configuration, or is otherwise True by default.
@@ -742,6 +740,7 @@ class _App:
         concurrency_limit: Optional[int] = None,  # Replaced with `max_containers`
         container_idle_timeout: Optional[int] = None,  # Replaced with `scaledown_window`
         allow_concurrent_inputs: Optional[int] = None,  # Replaced with the `@modal.concurrent` decorator
+        max_inputs: Optional[int] = None,  # Replaced with `single_use_containers`
         _experimental_buffer_containers: Optional[int] = None,  # Now stable API with `buffer_containers`
         _experimental_scheduler_placement: Optional[SchedulerPlacement] = None,  # Replaced in favor of
         # using `region` and `nonpreemptible`
@@ -764,6 +763,20 @@ class _App:
                 "\n\nSee https://modal.com/docs/guide/modal-1-0-migration for more information.",
             )
 
+        if max_inputs is not None:
+            if not isinstance(max_inputs, int):
+                raise InvalidError(f"`max_inputs` must be an int, not {type(max_inputs).__name__}")
+            if max_inputs <= 0:
+                raise InvalidError("`max_inputs` must be positive")
+            if max_inputs > 1:
+                raise InvalidError("Only `max_inputs=1` is currently supported")
+            deprecation_warning(
+                (2025, 12, 16),
+                "The `max_inputs` parameter is deprecated. Please set `single_use_containers=True` instead.",
+                pending=True,
+            )
+            single_use_containers = max_inputs == 1
+
         if _experimental_scheduler_placement is not None:
             deprecation_warning(
                 (2025, 11, 17),
@@ -912,7 +925,7 @@ class _App:
                 enable_memory_snapshot=enable_memory_snapshot,
                 block_network=block_network,
                 restrict_modal_access=restrict_modal_access,
-                max_inputs=max_inputs,
+                single_use_containers=single_use_containers,
                 i6pn_enabled=i6pn_enabled,
                 cluster_size=cluster_size,  # Experimental: Clustered functions
                 rdma=rdma,
@@ -969,9 +982,7 @@ class _App:
         enable_memory_snapshot: bool = False,  # Enable memory checkpointing for faster cold starts.
         block_network: bool = False,  # Whether to block network access
         restrict_modal_access: bool = False,  # Whether to allow this class access to other Modal resources
-        # Limits the number of inputs a container handles before shutting down.
-        # Use `max_inputs = 1` for single-use containers.
-        max_inputs: Optional[int] = None,
+        single_use_containers: bool = False,  # When True, containers will shut down after handling a single input
         i6pn: Optional[bool] = None,  # Whether to enable IPv6 container networking within the region.
         include_source: Optional[bool] = None,  # When `False`, don't automatically add the App source to the container.
         experimental_options: Optional[dict[str, Any]] = None,
@@ -984,6 +995,7 @@ class _App:
         concurrency_limit: Optional[int] = None,  # Replaced with `max_containers`
         container_idle_timeout: Optional[int] = None,  # Replaced with `scaledown_window`
         allow_concurrent_inputs: Optional[int] = None,  # Replaced with the `@modal.concurrent` decorator
+        max_inputs: Optional[int] = None,  # Replaced with `single_use_containers`
         _experimental_buffer_containers: Optional[int] = None,  # Now stable API with `buffer_containers`
         _experimental_scheduler_placement: Optional[SchedulerPlacement] = None,  # Replaced in favor of
         # using `region` and `nonpreemptible`
@@ -1002,6 +1014,20 @@ class _App:
                 "\n\nSee https://modal.com/docs/guide/modal-1-0-migration for more information.",
             )
 
+        if max_inputs is not None:
+            if not isinstance(max_inputs, int):
+                raise InvalidError(f"`max_inputs` must be an int, not {type(max_inputs).__name__}")
+            if max_inputs <= 0:
+                raise InvalidError("`max_inputs` must be positive")
+            if max_inputs > 1:
+                raise InvalidError("Only `max_inputs=1` is currently supported")
+            deprecation_warning(
+                (2025, 12, 16),
+                "The `max_inputs` parameter is deprecated. Please set `single_use_containers=True` instead.",
+                pending=True,
+            )
+            single_use_containers = max_inputs == 1
+
         if _experimental_scheduler_placement is not None:
             deprecation_warning(
                 (2025, 11, 17),
@@ -1139,7 +1165,7 @@ class _App:
                 enable_memory_snapshot=enable_memory_snapshot,
                 block_network=block_network,
                 restrict_modal_access=restrict_modal_access,
-                max_inputs=max_inputs,
+                single_use_containers=single_use_containers,
                 http_config=http_config,
                 i6pn_enabled=i6pn_enabled,
                 cluster_size=cluster_size,

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/app.pyi

@@ -491,7 +491,7 @@ class _App:
         enable_memory_snapshot: bool = False,
         block_network: bool = False,
         restrict_modal_access: bool = False,
-        max_inputs: typing.Optional[int] = None,
+        single_use_containers: bool = False,
         i6pn: typing.Optional[bool] = None,
         include_source: typing.Optional[bool] = None,
         experimental_options: typing.Optional[dict[str, typing.Any]] = None,
@@ -502,6 +502,7 @@ class _App:
         concurrency_limit: typing.Optional[int] = None,
         container_idle_timeout: typing.Optional[int] = None,
         allow_concurrent_inputs: typing.Optional[int] = None,
+        max_inputs: typing.Optional[int] = None,
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_scheduler_placement: typing.Optional[modal.scheduler_placement.SchedulerPlacement] = None,
     ) -> _FunctionDecoratorType:
@@ -545,7 +546,7 @@ class _App:
         enable_memory_snapshot: bool = False,
         block_network: bool = False,
         restrict_modal_access: bool = False,
-        max_inputs: typing.Optional[int] = None,
+        single_use_containers: bool = False,
         i6pn: typing.Optional[bool] = None,
         include_source: typing.Optional[bool] = None,
         experimental_options: typing.Optional[dict[str, typing.Any]] = None,
@@ -556,6 +557,7 @@ class _App:
         concurrency_limit: typing.Optional[int] = None,
         container_idle_timeout: typing.Optional[int] = None,
         allow_concurrent_inputs: typing.Optional[int] = None,
+        max_inputs: typing.Optional[int] = None,
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_scheduler_placement: typing.Optional[modal.scheduler_placement.SchedulerPlacement] = None,
     ) -> collections.abc.Callable[[typing.Union[CLS_T, modal._partial_function._PartialFunction]], CLS_T]:
@@ -1154,7 +1156,7 @@ class App:
         enable_memory_snapshot: bool = False,
         block_network: bool = False,
         restrict_modal_access: bool = False,
-        max_inputs: typing.Optional[int] = None,
+        single_use_containers: bool = False,
         i6pn: typing.Optional[bool] = None,
         include_source: typing.Optional[bool] = None,
         experimental_options: typing.Optional[dict[str, typing.Any]] = None,
@@ -1165,6 +1167,7 @@ class App:
         concurrency_limit: typing.Optional[int] = None,
         container_idle_timeout: typing.Optional[int] = None,
         allow_concurrent_inputs: typing.Optional[int] = None,
+        max_inputs: typing.Optional[int] = None,
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_scheduler_placement: typing.Optional[modal.scheduler_placement.SchedulerPlacement] = None,
     ) -> _FunctionDecoratorType:
@@ -1208,7 +1211,7 @@ class App:
         enable_memory_snapshot: bool = False,
         block_network: bool = False,
         restrict_modal_access: bool = False,
-        max_inputs: typing.Optional[int] = None,
+        single_use_containers: bool = False,
         i6pn: typing.Optional[bool] = None,
         include_source: typing.Optional[bool] = None,
         experimental_options: typing.Optional[dict[str, typing.Any]] = None,
@@ -1219,6 +1222,7 @@ class App:
         concurrency_limit: typing.Optional[int] = None,
         container_idle_timeout: typing.Optional[int] = None,
         allow_concurrent_inputs: typing.Optional[int] = None,
+        max_inputs: typing.Optional[int] = None,
         _experimental_buffer_containers: typing.Optional[int] = None,
         _experimental_scheduler_placement: typing.Optional[modal.scheduler_placement.SchedulerPlacement] = None,
     ) -> collections.abc.Callable[[typing.Union[CLS_T, modal.partial_function.PartialFunction]], CLS_T]:

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/client.pyi

@@ -32,7 +32,7 @@ class _Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.2.7.dev6",
+        version: str = "1.2.7.dev9",
     ):
         """mdmd:hidden
         The Modal client object is not intended to be instantiated directly by users.
@@ -163,7 +163,7 @@ class Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.2.7.dev6",
+        version: str = "1.2.7.dev9",
     ):
         """mdmd:hidden
         The Modal client object is not intended to be instantiated directly by users.

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/file_io.py

@@ -190,7 +190,7 @@ class _FileIO(Generic[T]):
 
     async def _wait(self, exec_id: str) -> bytes:
         # The logic here is similar to how output is read from `exec`
-        output = b""
+        output_buffer = io.BytesIO()
         completed = False
         retries_remaining = 10
         while not completed:
@@ -201,7 +201,7 @@ class _FileIO(Generic[T]):
                         break
                     if isinstance(data, Exception):
                         raise data
-                    output += data
+                    output_buffer.write(data)
             except (GRPCError, StreamTerminatedError) as exc:
                 if retries_remaining > 0:
                     retries_remaining -= 1
@@ -212,7 +212,7 @@ class _FileIO(Generic[T]):
                     elif isinstance(exc, StreamTerminatedError):
                         continue
                 raise
-        return output
+        return output_buffer.getvalue()
 
     def _validate_type(self, data: Union[bytes, str]) -> None:
         if self._binary and isinstance(data, str):
@@ -421,10 +421,22 @@ class _FileIO(Generic[T]):
                 task_id=self._task_id,
             ),
         )
+
+        def end_of_event(item: bytes, buffer: io.BytesIO, boundary_token: bytes) -> bool:
+            if not item.endswith(b"\n"):
+                return False
+            boundary_token_size = len(boundary_token)
+            if buffer.tell() < boundary_token_size:
+                return False
+            buffer.seek(-boundary_token_size, io.SEEK_END)
+            if buffer.read(boundary_token_size) == boundary_token:
+                return True
+            return False
+
         async with TaskContext() as tc:
             tc.create_task(self._consume_watch_output(resp.exec_id))
 
-            buffer = b""
+            item_buffer = io.BytesIO()
             while True:
                 if len(self._watch_output_buffer) > 0:
                     item = self._watch_output_buffer.pop(0)
@@ -432,12 +444,12 @@ class _FileIO(Generic[T]):
                         break
                     if isinstance(item, Exception):
                         raise item
-                    buffer += item
-                    # a single event may be split across multiple messages
-                    # the end of an event is marked by two newlines
-                    if buffer.endswith(b"\n\n"):
+                    item_buffer.write(item)
+                    assert isinstance(item, bytes)
+                    # Single events may span multiple messages so we need to check for a special event boundary token
+                    if end_of_event(item, item_buffer, boundary_token=b"\n\n"):
                         try:
-                            event_json = json.loads(buffer.strip().decode())
+                            event_json = json.loads(item_buffer.getvalue().strip().decode())
                             event = FileWatchEvent(
                                 type=FileWatchEventType(event_json["event_type"]),
                                 paths=event_json["paths"],
@@ -447,7 +459,7 @@ class _FileIO(Generic[T]):
                         except (json.JSONDecodeError, KeyError, ValueError):
                             # skip invalid events
                             pass
-                        buffer = b""
+                        item_buffer = io.BytesIO()
                 else:
                     await asyncio.sleep(0.1)
 

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/functions.pyi

@@ -106,7 +106,7 @@ class Function(
         i6pn_enabled: bool = False,
         cluster_size: typing.Optional[int] = None,
         rdma: typing.Optional[bool] = None,
-        max_inputs: typing.Optional[int] = None,
+        single_use_containers: bool = False,
         ephemeral_disk: typing.Optional[int] = None,
         include_source: bool = True,
         experimental_options: typing.Optional[dict[str, str]] = None,

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal/io_streams.py

@@ -23,7 +23,7 @@ from grpclib.exceptions import GRPCError, StreamTerminatedError
 from modal.exception import ClientClosed, ExecTimeoutError, InvalidError
 from modal_proto import api_pb2
 
-from ._utils.async_utils import aclosing, synchronize_api
+from ._utils.async_utils import aclosing, synchronize_api, synchronizer
 from ._utils.grpc_utils import RETRYABLE_GRPC_STATUS_CODES
 from ._utils.task_command_router_client import TaskCommandRouterClient
 from .client import _Client
@@ -153,17 +153,17 @@ class _StreamReaderThroughServer(Generic[T]):
         """Fetch the entire contents of the stream until EOF."""
         logger.debug(f"{self._object_id} StreamReader fd={self._file_descriptor} read starting")
         if self._text:
-            data_str = ""
+            buffer = io.StringIO()
             async for message in _decode_bytes_stream_to_str(self._get_logs()):
-                data_str += message
+                buffer.write(message)
             logger.debug(f"{self._object_id} StreamReader fd={self._file_descriptor} read completed after EOF")
-            return cast(T, data_str)
+            return cast(T, buffer.getvalue())
         else:
-            data_bytes = b""
+            buffer = io.BytesIO()
             async for message in self._get_logs():
-                data_bytes += message
+                buffer.write(message)
             logger.debug(f"{self._object_id} StreamReader fd={self._file_descriptor} read completed after EOF")
-            return cast(T, data_bytes)
+            return cast(T, buffer.getvalue())
 
     async def _consume_container_process_stream(self):
         """Consume the container process stream and store messages in the buffer."""
@@ -566,6 +566,9 @@ class _StreamReader(Generic[T]):
         task_id: Optional[str] = None,
     ) -> None:
         """mdmd:hidden"""
+        # we can remove this once we ensure no constructors use async code
+        assert asyncio.get_running_loop() == synchronizer._get_loop(start=False)
+
         if by_line and not text:
             raise ValueError("line-buffering is only supported when text=True")
 

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.2.7.dev6
+Version: 1.2.7.dev9
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.2.7.dev6 → modal-1.2.7.dev9}/modal_proto/api.proto

@@ -1469,6 +1469,8 @@ message Function {
 
   bool block_network = 44;
 
+  // In the SDK, we've deprecated `max_inputs` (which only every implemented `max_inputs=1`)
+  // in favor of a boolean `single_use_containers` parameter.
   uint32 max_inputs = 46;
 
   repeated S3Mount s3_mounts = 47;
@@ -1556,6 +1558,9 @@ message Function {
   // Attribute on the module with the implementation, which may differ from function_name
   // when the user provided a custom name= for the Function inside the Application namespace
   string implementation_name = 90;
+
+  bool single_use_containers = 91; // When True, containers will shut down after handling a single input
+
 }
 
 message FunctionAsyncInvokeRequest {