modal 1.0.4.dev10__py3-none-any.whl → 1.0.5.dev1__py3-none-any.whl

modal/_utils/async_utils.py

@@ -396,7 +396,7 @@ class _WarnIfGeneratorIsNotConsumed:
         return await self.gen.aclose()
 
 
-synchronize_api(_WarnIfGeneratorIsNotConsumed)
+_BlockingWarnIfGeneratorIsNotConsumed = synchronize_api(_WarnIfGeneratorIsNotConsumed)
 
 
 class _WarnIfNonWrappedGeneratorIsNotConsumed(_WarnIfGeneratorIsNotConsumed):

modal/client.pyi

@@ -31,7 +31,7 @@ class _Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.0.4.dev10",
+        version: str = "1.0.5.dev1",
     ): ...
     def is_closed(self) -> bool: ...
     @property
@@ -94,7 +94,7 @@ class Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.0.4.dev10",
+        version: str = "1.0.5.dev1",
     ): ...
     def is_closed(self) -> bool: ...
     @property

modal/cls.py

@@ -75,6 +75,7 @@ def _get_class_constructor_signature(user_cls: type) -> inspect.Signature:
 
 @dataclasses.dataclass()
 class _ServiceOptions:
+    # Note that default values should always be "untruthy" so we can detect when they are not set
     secrets: typing.Collection[_Secret] = ()
     validated_volumes: typing.Sequence[tuple[str, _Volume]] = ()
     resources: Optional[api_pb2.Resources] = None
@@ -88,6 +89,25 @@ class _ServiceOptions:
     batch_max_size: Optional[int] = None
     batch_wait_ms: Optional[int] = None
 
+    def merge_options(self, new_options: "_ServiceOptions") -> "_ServiceOptions":
+        """Implement protobuf-like MergeFrom semantics for this dataclass.
+
+        This mostly exists to support "stacking" of `.with_options()` calls.
+        """
+        new_options_dict = dataclasses.asdict(new_options)
+
+        # Resources needs special merge handling because individual fields are parameters in the public API
+        merged_resources = api_pb2.Resources()
+        if self.resources:
+            merged_resources.MergeFrom(self.resources)
+        if new_resources := new_options_dict.pop("resources"):
+            merged_resources.MergeFrom(new_resources)
+        self.resources = merged_resources
+
+        for key, value in new_options_dict.items():
+            if value:  # Only overwrite data when the value was set in the new options
+                setattr(self, key, value)
+
 
 def _bind_instance_method(cls: "_Cls", service_function: _Function, method_name: str):
     """Binds an "instance service function" to a specific method using metadata for that method
@@ -664,15 +684,32 @@ More information on class parameterization can be found here: https://modal.com/
         container_idle_timeout: Optional[int] = None,  # Now called `scaledown_window`
         allow_concurrent_inputs: Optional[int] = None,  # See `.with_concurrency`
     ) -> "_Cls":
-        """Create an instance of the Cls with configuration options overridden with new values.
+        """Override the static Function configuration at runtime.
+
+        This method will return a new instance of the cls that will autoscale independently of the
+        original instance. Note that options cannot be "unset" with this method (i.e., if a GPU
+        is configured in the `@app.cls()` decorator, passing `gpu=None` here will not create a
+        CPU-only instance).
 
         **Usage:**
 
+        You can use this method after looking up the Cls from a deployed App or if you have a
+        direct reference to a Cls from another Function or local entrypoint on its App:
+
         ```python notest
         Model = modal.Cls.from_name("my_app", "Model")
         ModelUsingGPU = Model.with_options(gpu="A100")
-        ModelUsingGPU().generate.remote(42)  # will run with an A100 GPU
+        ModelUsingGPU().generate.remote(input_prompt)  # Run with an A100 GPU
         ```
+
+        The method can be called multiple times to "stack" updates:
+
+        ```python notest
+        Model.with_options(gpu="A100").with_options(scaledown_window=300)  # Use an A100 with slow scaledown
+        ```
+
+        Note that container arguments (i.e. `volumes` and `secrets`) passed in subsequent calls
+        will not be merged.
         """
         retry_policy = _parse_retries(retries, f"Class {self.__name__}" if self._user_cls else "")
         if gpu or cpu or memory:
@@ -705,21 +742,23 @@ More information on class parameterization can be found here: https://modal.com/
 
         cls = _Cls._from_loader(_load_from_base, rep=f"{self._name}.with_options(...)", is_another_app=True, deps=_deps)
         cls._initialize_from_other(self)
-        cls._options = dataclasses.replace(
-            cls._options,
+
+        new_options = _ServiceOptions(
             secrets=secrets,
+            validated_volumes=validate_volumes(volumes),
             resources=resources,
             retry_policy=retry_policy,
             max_containers=max_containers,
             buffer_containers=buffer_containers,
             scaledown_window=scaledown_window,
             timeout_secs=timeout,
-            validated_volumes=validate_volumes(volumes),
             # Note: set both for backwards / forwards compatibility
             # But going forward `.with_concurrency` is the preferred method with distinct parameterization
             max_concurrent_inputs=allow_concurrent_inputs,
             target_concurrent_inputs=allow_concurrent_inputs,
         )
+
+        cls._options.merge_options(new_options)
         return cls
 
     def with_concurrency(self: "_Cls", *, max_inputs: int, target_inputs: Optional[int] = None) -> "_Cls":
@@ -746,9 +785,9 @@ More information on class parameterization can be found here: https://modal.com/
             _load_from_base, rep=f"{self._name}.with_concurrency(...)", is_another_app=True, deps=_deps
         )
         cls._initialize_from_other(self)
-        cls._options = dataclasses.replace(
-            cls._options, max_concurrent_inputs=max_inputs, target_concurrent_inputs=target_inputs
-        )
+
+        concurrency_options = _ServiceOptions(max_concurrent_inputs=max_inputs, target_concurrent_inputs=target_inputs)
+        cls._options.merge_options(concurrency_options)
         return cls
 
     def with_batching(self: "_Cls", *, max_batch_size: int, wait_ms: int) -> "_Cls":
@@ -775,7 +814,9 @@ More information on class parameterization can be found here: https://modal.com/
             _load_from_base, rep=f"{self._name}.with_concurrency(...)", is_another_app=True, deps=_deps
         )
         cls._initialize_from_other(self)
-        cls._options = dataclasses.replace(cls._options, batch_max_size=max_batch_size, batch_wait_ms=wait_ms)
+
+        batching_options = _ServiceOptions(batch_max_size=max_batch_size, batch_wait_ms=wait_ms)
+        cls._options.merge_options(batching_options)
         return cls
 
     @staticmethod

modal/cls.pyi

@@ -37,6 +37,7 @@ class _ServiceOptions:
     batch_max_size: typing.Optional[int]
     batch_wait_ms: typing.Optional[int]
 
+    def merge_options(self, new_options: _ServiceOptions) -> _ServiceOptions: ...
     def __init__(
         self,
         secrets: typing.Collection[modal.secret._Secret] = (),

modal/parallel_map.py

@@ -278,17 +278,19 @@ async def _map_invocation(
                 )
             )
             map_done_task = asyncio.create_task(map_done_event.wait())
-            done, pending = await asyncio.wait([get_response_task, map_done_task], return_when=FIRST_COMPLETED)
-            if get_response_task in done:
-                map_done_task.cancel()
-                response = get_response_task.result()
-            else:
-                assert map_done_event.is_set()
-                # map is done, cancel the pending call
+            try:
+                done, pending = await asyncio.wait([get_response_task, map_done_task], return_when=FIRST_COMPLETED)
+                if get_response_task in done:
+                    map_done_task.cancel()
+                    response = get_response_task.result()
+                else:
+                    assert map_done_event.is_set()
+                    # map is done - no more outputs, so return early
+                    return
+            finally:
+                # clean up tasks, in case of cancellations etc.
                 get_response_task.cancel()
-                # not strictly necessary - don't leave dangling task
-                await asyncio.gather(get_response_task, return_exceptions=True)
-                return
+                map_done_task.cancel()
 
             last_entry_id = response.last_entry_id
             now_seconds = int(time.time())
@@ -716,6 +718,7 @@ class _MapItemContext:
         Return True if input state was changed to COMPLETE, otherwise False.
         """
         # If the item is already complete, this is a duplicate output and can be ignored.
+
         if self.state == _MapItemState.COMPLETE:
             logger.debug(
                 f"Received output for input marked as complete. Must be duplicate, so ignoring. "

{modal-1.0.4.dev10.dist-info → modal-1.0.5.dev1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.0.4.dev10
+Version: 1.0.5.dev1
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0
@@ -22,7 +22,7 @@ Requires-Dist: click~=8.1.0
 Requires-Dist: grpclib==0.4.7
 Requires-Dist: protobuf!=4.24.0,<7.0,>=3.19
 Requires-Dist: rich>=12.0.0
-Requires-Dist: synchronicity~=0.9.14
+Requires-Dist: synchronicity~=0.9.15
 Requires-Dist: toml
 Requires-Dist: typer>=0.9
 Requires-Dist: types-certifi

{modal-1.0.4.dev10.dist-info → modal-1.0.5.dev1.dist-info}/RECORD

@@ -22,11 +22,11 @@ modal/app.py,sha256=NZ_rJ9TuMfiNiLg8-gOFgufD5flGtXWPHOZI0gdD3hE,46585
 modal/app.pyi,sha256=4-b_vbe3lNAqQPcMRpQCEDsE1zsVkQRJGUql9B7HvbM,22659
 modal/call_graph.py,sha256=1g2DGcMIJvRy-xKicuf63IVE98gJSnQsr8R_NVMptNc,2581
 modal/client.py,sha256=OwISJvkgMb-rHm9Gc4i-7YcDgGiZgwJ7F_PzwZH7a6Q,16847
-modal/client.pyi,sha256=rX7aILLKNY7gU3DXvvGNT9Hi5IheqO6l0iFjSfDZMCw,8459
+modal/client.pyi,sha256=mEQpk7s6oZXanSwdpeUki1HdXyxXk-Wu26jnOKGYDqY,8457
 modal/cloud_bucket_mount.py,sha256=YOe9nnvSr4ZbeCn587d7_VhE9IioZYRvF9VYQTQux08,5914
 modal/cloud_bucket_mount.pyi,sha256=30T3K1a89l6wzmEJ_J9iWv9SknoGqaZDx59Xs-ZQcmk,1607
-modal/cls.py,sha256=dBbeARwOWftlKd1cwtM0cHFtQWSWkwVXwVmOV4w0SyI,37907
-modal/cls.pyi,sha256=-gSV02T5S1La1dvmHHB_f139nNySk5MeXRSF4yp8z4s,12660
+modal/cls.py,sha256=JjxAeVudXKMp-co9YUnsGG6COJOv91K0ylf4f3mR67o,39792
+modal/cls.pyi,sha256=EL1mQuyf8nAL2AjVZVnZhDzhakUmQt1ZtOVMy_aYWJI,12742
 modal/config.py,sha256=e8sQ4RgwgJ_45S302vWUWs_wqRlKyEt3tU898RiaDKE,12073
 modal/container_process.py,sha256=PDvjcyZ6eeN8foKQgR0WJ66Sg3lt7OFhK7Y_Akz6k5w,5846
 modal/container_process.pyi,sha256=pPIUxVV_TY4huO2jF5cSSjb6L_EN7Es4xRvuwZ5sa5M,2802
@@ -52,7 +52,7 @@ modal/network_file_system.pyi,sha256=58DiUqHGlARmI3cz-Yo7IFObKKFIiGh5UIU5JxGNFfc
 modal/object.py,sha256=bTeskuY8JFrESjU4_UL_nTwYlBQdOLmVaOX3X6EMxsg,164
 modal/object.pyi,sha256=UkR8NQ1jCIaw3hBUPxBRc6vvrOqtV37G_hsW2O5-4wE,5378
 modal/output.py,sha256=q4T9uHduunj4NwY-YSwkHGgjZlCXMuJbfQ5UFaAGRAc,1968
-modal/parallel_map.py,sha256=1SOOUzKGVItv9fP2WvYR5vl49Y5q2GdupsnJMLIUDUw,37434
+modal/parallel_map.py,sha256=QkDxaFQwzOCU6-Vhb-Iq78eEHKqqlNnGMqjk-BdhI08,37469
 modal/parallel_map.pyi,sha256=mhYGQmufQEJbjNrX7vNhBS2gUdfBrpmuWNUHth_Dz6U,6140
 modal/partial_function.py,sha256=SwuAAj2wj4SO6F6nkSnwNZrczEmm9w9YdlQTHh6hr04,1195
 modal/partial_function.pyi,sha256=NFWz1aCAs2B3-GnPf1cTatWRZOLnYpFKCnjP_X9iNRs,6411
@@ -91,7 +91,7 @@ modal/_runtime/telemetry.py,sha256=T1RoAGyjBDr1swiM6pPsGRSITm7LI5FDK18oNXxY08U,5
 modal/_runtime/user_code_imports.py,sha256=78wJyleqY2RVibqcpbDQyfWVBVT9BjyHPeoV9WdwV5Y,17720
 modal/_utils/__init__.py,sha256=waLjl5c6IPDhSsdWAm9Bji4e2PVxamYABKAze6CHVXY,28
 modal/_utils/app_utils.py,sha256=88BT4TPLWfYAQwKTHcyzNQRHg8n9B-QE2UyJs96iV-0,108
-modal/_utils/async_utils.py,sha256=wgOSjofVaQAyP8Oq_xTSET5mYP5Y5W_eaQCPmE63xRA,29321
+modal/_utils/async_utils.py,sha256=MhSCsCL8GqIVFWoHubU_899IH-JBZAiiqadG9Wri2l4,29361
 modal/_utils/blob_utils.py,sha256=IexC2Jbtqp_Tkmy62ayfgzTYte0UPCNufB_v-DO21g8,18585
 modal/_utils/bytes_io_segment_payload.py,sha256=vaXPq8b52-x6G2hwE7SrjS58pg_aRm7gV3bn3yjmTzQ,4261
 modal/_utils/deprecation.py,sha256=EXP1beU4pmEqEzWMLw6E3kUfNfpmNA_VOp6i0EHi93g,4856
@@ -147,7 +147,7 @@ modal/requirements/2024.10.txt,sha256=qD-5cVIVM9wXesJ6JC89Ew-3m2KjEElUz3jaw_MddR
 modal/requirements/PREVIEW.txt,sha256=KxDaVTOwatHvboDo4lorlgJ7-n-MfAwbPwxJ0zcJqrs,312
 modal/requirements/README.md,sha256=9tK76KP0Uph7O0M5oUgsSwEZDj5y-dcUPsnpR0Sc-Ik,854
 modal/requirements/base-images.json,sha256=f1bwyp2UkM844eoO9Qk30gQw_xrMqKpMSeJ6MErXnEk,995
-modal-1.0.4.dev10.dist-info/licenses/LICENSE,sha256=psuoW8kuDP96RQsdhzwOqi6fyWv0ct8CR6Jr7He_P_k,10173
+modal-1.0.5.dev1.dist-info/licenses/LICENSE,sha256=psuoW8kuDP96RQsdhzwOqi6fyWv0ct8CR6Jr7He_P_k,10173
 modal_docs/__init__.py,sha256=svYKtV8HDwDCN86zbdWqyq5T8sMdGDj0PVlzc2tIxDM,28
 modal_docs/gen_cli_docs.py,sha256=c1yfBS_x--gL5bs0N4ihMwqwX8l3IBWSkBAKNNIi6bQ,3801
 modal_docs/gen_reference_docs.py,sha256=d_CQUGQ0rfw28u75I2mov9AlS773z9rG40-yq5o7g2U,6359
@@ -155,10 +155,10 @@ modal_docs/mdmd/__init__.py,sha256=svYKtV8HDwDCN86zbdWqyq5T8sMdGDj0PVlzc2tIxDM,2
 modal_docs/mdmd/mdmd.py,sha256=Irx49MCCTlBOP4FBdLR--JrpA3-WhsVeriq0LGgsRic,6232
 modal_docs/mdmd/signatures.py,sha256=XJaZrK7Mdepk5fdX51A8uENiLFNil85Ud0d4MH8H5f0,3218
 modal_proto/__init__.py,sha256=MIEP8jhXUeGq_eCjYFcqN5b1bxBM4fdk0VESpjWR0fc,28
-modal_proto/api.proto,sha256=nwh8RmhypmnFMKQ-WwHolz-1NvKr4JgRSc2As_cZ0JE,96407
+modal_proto/api.proto,sha256=aBHQSEFm1Os9ciCl6N9N5bhumjdJIpl7VPSbP2Chjk4,96432
 modal_proto/api_grpc.py,sha256=iY5o_Tm4VDP-Wa1JgA_NpQa_Y-4FYB_RN9wdSUExjwI,117469
-modal_proto/api_pb2.py,sha256=UeQg7ABeXbNCBLcft3pTZtV6EoLMifZ_O_BPxYadm0w,338603
-modal_proto/api_pb2.pyi,sha256=QQp5_zlv3j6_v7CaqM4yFKW4xun1yzy4lxwkn4_O3Ew,463819
+modal_proto/api_pb2.py,sha256=PTGsaddASU8iqRoaTFmV5bNg-TtGGUGvf8tTVWjeR1Q,338641
+modal_proto/api_pb2.pyi,sha256=hMvUMcYnpt7tR0fdkL-sDuOM-r2vistuds_rL03xY-k,463959
 modal_proto/api_pb2_grpc.py,sha256=NL5prOS_hh_pA1hVvQP_ZRE1w49N-PR8iNPRZ65i6nA,254089
 modal_proto/api_pb2_grpc.pyi,sha256=Xxgdcnv1mBnu5_AQxJ6fo0yz7GnqVU0HVObNfZWHVfM,59440
 modal_proto/modal_api_grpc.py,sha256=0ir2lnwT3-IgPcAWw98yWMAiqZPkjvNro9UBk4u8hnk,17763
@@ -170,10 +170,10 @@ modal_proto/options_pb2.pyi,sha256=l7DBrbLO7q3Ir-XDkWsajm0d0TQqqrfuX54i4BMpdQg,1
 modal_proto/options_pb2_grpc.py,sha256=1oboBPFxaTEXt9Aw7EAj8gXHDCNMhZD2VXqocC9l_gk,159
 modal_proto/options_pb2_grpc.pyi,sha256=CImmhxHsYnF09iENPoe8S4J-n93jtgUYD2JPAc0yJSI,247
 modal_proto/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-modal_version/__init__.py,sha256=LNXQQhlCrbhgsIePbjkDp4B4zhO1mSW6NEDtVxtRmu4,121
+modal_version/__init__.py,sha256=IBJQ2mcZZY6WLV44CaQFHUQqyZlvclRpvrY6CtGf5gs,120
 modal_version/__main__.py,sha256=2FO0yYQQwDTh6udt1h-cBnGd1c4ZyHnHSI4BksxzVac,105
-modal-1.0.4.dev10.dist-info/METADATA,sha256=4tXSMwgkVzvte0dFfa9VIzNOYw3u0L4DGTh5oDHfw8U,2455
-modal-1.0.4.dev10.dist-info/WHEEL,sha256=1tXe9gY0PYatrMPMDd6jXqjfpz_B-Wqm32CPfRC58XU,91
-modal-1.0.4.dev10.dist-info/entry_points.txt,sha256=An-wYgeEUnm6xzrAP9_NTSTSciYvvEWsMZILtYrvpAI,46
-modal-1.0.4.dev10.dist-info/top_level.txt,sha256=4BWzoKYREKUZ5iyPzZpjqx4G8uB5TWxXPDwibLcVa7k,43
-modal-1.0.4.dev10.dist-info/RECORD,,
+modal-1.0.5.dev1.dist-info/METADATA,sha256=HkVebF9LFr4MAW7KocR0W-AVfGfw5U6eUiU5gxTPPHI,2454
+modal-1.0.5.dev1.dist-info/WHEEL,sha256=1tXe9gY0PYatrMPMDd6jXqjfpz_B-Wqm32CPfRC58XU,91
+modal-1.0.5.dev1.dist-info/entry_points.txt,sha256=An-wYgeEUnm6xzrAP9_NTSTSciYvvEWsMZILtYrvpAI,46
+modal-1.0.5.dev1.dist-info/top_level.txt,sha256=4BWzoKYREKUZ5iyPzZpjqx4G8uB5TWxXPDwibLcVa7k,43
+modal-1.0.5.dev1.dist-info/RECORD,,

modal_proto/api.proto

@@ -2796,6 +2796,7 @@ message TaskInfo {
   modal.client.GenericResult result = 4;
   double enqueued_at = 5;
   string gpu_type = 6;
+  string sandbox_id = 7;
 }
 
 message TaskListRequest {