modal 1.1.3.dev2__tar.gz → 1.1.3.dev4__tar.gz

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.1.3.dev2
+Version: 1.1.3.dev4
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/modal/_functions.py

@@ -100,6 +100,10 @@ if TYPE_CHECKING:
     import modal.partial_function
 
 MAX_INTERNAL_FAILURE_COUNT = 8
+TERMINAL_STATUSES = (
+    api_pb2.GenericResult.GENERIC_STATUS_SUCCESS,
+    api_pb2.GenericResult.GENERIC_STATUS_TERMINATED,
+)
 
 
 @dataclasses.dataclass
@@ -300,11 +304,7 @@ class _Invocation:
 
         while True:
             item = await self._get_single_output(ctx.input_jwt)
-            if item.result.status in (
-                api_pb2.GenericResult.GENERIC_STATUS_SUCCESS,
-                api_pb2.GenericResult.GENERIC_STATUS_TERMINATED,
-            ):
-                # success or cancellations are "final" results
+            if item.result.status in TERMINAL_STATUSES:
                 return await _process_result(item.result, item.data_format, self.stub, self.client)
 
             if item.result.status != api_pb2.GenericResult.GENERIC_STATUS_INTERNAL_FAILURE:
@@ -411,6 +411,7 @@ class _InputPlaneInvocation:
         client: _Client,
         input_item: api_pb2.FunctionPutInputsItem,
         function_id: str,
+        retry_policy: api_pb2.FunctionRetryPolicy,
         input_plane_region: str,
     ):
         self.stub = stub
@@ -418,6 +419,7 @@ class _InputPlaneInvocation:
         self.attempt_token = attempt_token
         self.input_item = input_item
         self.function_id = function_id
+        self.retry_policy = retry_policy
         self.input_plane_region = input_plane_region
 
     @staticmethod
@@ -453,11 +455,15 @@ class _InputPlaneInvocation:
         response = await retry_transient_errors(stub.AttemptStart, request, metadata=metadata)
         attempt_token = response.attempt_token
 
-        return _InputPlaneInvocation(stub, attempt_token, client, input_item, function_id, input_plane_region)
+        return _InputPlaneInvocation(
+            stub, attempt_token, client, input_item, function_id, response.retry_policy, input_plane_region
+        )
 
     async def run_function(self) -> Any:
+        # User errors including timeouts are managed by the user-specified retry policy.
+        user_retry_manager = RetryManager(self.retry_policy)
+
         # This will retry when the server returns GENERIC_STATUS_INTERNAL_FAILURE, i.e. lost inputs or worker preemption
-        # TODO(ryan): add logic to retry for user defined retry policy
         internal_failure_count = 0
         while True:
             await_request = api_pb2.AttemptAwaitRequest(
@@ -474,32 +480,48 @@ class _InputPlaneInvocation:
             )
 
             if await_response.HasField("output"):
+                if await_response.output.result.status in TERMINAL_STATUSES:
+                    return await _process_result(
+                        await_response.output.result, await_response.output.data_format, self.client.stub, self.client
+                    )
+
                 if await_response.output.result.status == api_pb2.GenericResult.GENERIC_STATUS_INTERNAL_FAILURE:
                     internal_failure_count += 1
                     # Limit the number of times we retry
                     if internal_failure_count < MAX_INTERNAL_FAILURE_COUNT:
                         # For system failures on the server, we retry immediately,
                         # and the failure does not count towards the retry policy.
-                        retry_request = api_pb2.AttemptRetryRequest(
-                            function_id=self.function_id,
-                            parent_input_id=current_input_id() or "",
-                            input=self.input_item,
-                            attempt_token=self.attempt_token,
-                        )
-                        # TODO(ryan): Add exponential backoff?
-                        retry_response = await retry_transient_errors(
-                            self.stub.AttemptRetry,
-                            retry_request,
-                            metadata=metadata,
-                        )
-                        self.attempt_token = retry_response.attempt_token
+                        self.attempt_token = await self._retry_input(metadata)
                         continue
 
-                control_plane_stub = self.client.stub
-                # Note: Blob download is done on the control plane stub, not the input plane stub!
-                return await _process_result(
-                    await_response.output.result, await_response.output.data_format, control_plane_stub, self.client
-                )
+                # We add delays between retries for non-internal failures.
+                delay_ms = user_retry_manager.get_delay_ms()
+                if delay_ms is None:
+                    # No more retries either because we reached the retry limit or user didn't set a retry policy
+                    # and the limit defaulted to 0.
+                    # An unsuccessful status should raise an error when it's converted to an exception.
+                    # Note: Blob download is done on the control plane stub not the input plane stub!
+                    return await _process_result(
+                        await_response.output.result, await_response.output.data_format, self.client.stub, self.client
+                    )
+                await asyncio.sleep(delay_ms / 1000)
+
+            await self._retry_input(metadata)
+
+    async def _retry_input(self, metadata: list[tuple[str, str]]) -> str:
+        retry_request = api_pb2.AttemptRetryRequest(
+            function_id=self.function_id,
+            parent_input_id=current_input_id() or "",
+            input=self.input_item,
+            attempt_token=self.attempt_token,
+        )
+        # TODO(ryan): Add exponential backoff?
+        retry_response = await retry_transient_errors(
+            self.stub.AttemptRetry,
+            retry_request,
+            metadata=metadata,
+        )
+        return retry_response.attempt_token
 
     async def run_generator(self):
         items_received = 0

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/modal/client.pyi

@@ -33,7 +33,7 @@ class _Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.1.3.dev2",
+        version: str = "1.1.3.dev4",
     ):
         """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 = "1.1.3.dev2",
+        version: str = "1.1.3.dev4",
     ):
         """mdmd:hidden
         The Modal client object is not intended to be instantiated directly by users.

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/modal/functions.pyi

@@ -433,7 +433,7 @@ 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."""
             ...
@@ -442,7 +442,7 @@ 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]:
@@ -469,7 +469,7 @@ 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.
 
@@ -493,7 +493,7 @@ 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]):
@@ -502,7 +502,7 @@ 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.
 
@@ -523,7 +523,7 @@ 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-1.1.3.dev2 → modal-1.1.3.dev4}/modal/parallel_map.py

@@ -673,9 +673,11 @@ async def _map_invocation_inputplane(
     # any reason).
     max_inputs_outstanding = MAX_INPUTS_OUTSTANDING_DEFAULT
 
-    # Input plane does not yet return a retry policy. So we currently disable retries.
+    # Set a default retry policy to construct an instance of _MapItemsManager.
+    # We'll update the retry policy with the actual user-specified retry policy
+    # from the server in the first MapStartOrContinue response.
     retry_policy = api_pb2.FunctionRetryPolicy(
-        retries=0,  # Input plane does not yet return a retry policy. So only retry server failures for now.
+        retries=0,
         initial_delay_ms=1000,
         max_delay_ms=1000,
         backoff_coefficient=1.0,
@@ -771,10 +773,17 @@ async def _map_invocation_inputplane(
 
             map_items_manager.handle_put_continue_response(response_items_idx_tuple)
 
+            # Set the function call id and actual retry policy with the data from the first response.
+            # This conditional is skipped for subsequent iterations of this for-loop.
             if function_call_id is None:
                 function_call_id = response.function_call_id
                 function_call_id_received.set()
                 max_inputs_outstanding = response.max_inputs_outstanding or MAX_INPUTS_OUTSTANDING_DEFAULT
+                map_items_manager.set_retry_policy(response.retry_policy)
+                # Update the retry policy for the first batch of inputs.
+                # Subsequent batches will have the correct user-specified retry policy
+                # set by the updated _MapItemsManager.
+                map_items_manager.update_items_retry_policy(response.retry_policy)
         yield
 
     async def check_lost_inputs():
@@ -1475,6 +1484,9 @@ class _MapItemContext:
             retry_count=self.retry_manager.retry_count,
         )
 
+    def set_retry_policy(self, retry_policy: api_pb2.FunctionRetryPolicy):
+        self.retry_manager = RetryManager(retry_policy)
+
     def handle_retry_response(self, input_jwt: str):
         self.input_jwt.set_result(input_jwt)
         self.state = _MapItemState.WAITING_FOR_OUTPUT
@@ -1511,6 +1523,9 @@ class _MapItemsManager:
         self._sync_client_retries_enabled = sync_client_retries_enabled
         self._is_input_plane_instance = is_input_plane_instance
 
+    def set_retry_policy(self, retry_policy: api_pb2.FunctionRetryPolicy):
+        self._retry_policy = retry_policy
+
     async def add_items(self, items: list[api_pb2.FunctionPutInputsItem]):
         for item in items:
             # acquire semaphore to limit the number of inputs in progress
@@ -1540,6 +1555,10 @@ class _MapItemsManager:
     async def prepare_items_for_retry(self, retriable_idxs: list[int]) -> list[api_pb2.FunctionRetryInputsItem]:
         return [await self._item_context[idx].prepare_item_for_retry() for idx in retriable_idxs]
 
+    def update_items_retry_policy(self, retry_policy: api_pb2.FunctionRetryPolicy):
+        for ctx in self._item_context.values():
+            ctx.set_retry_policy(retry_policy)
+
     def get_input_jwts_waiting_for_output(self) -> list[str]:
         """
         Returns a list of input_jwts for inputs that are waiting for output.

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/modal/parallel_map.pyi

@@ -406,6 +406,7 @@ class _MapItemContext:
         ...
 
     async def prepare_item_for_retry(self) -> modal_proto.api_pb2.FunctionRetryInputsItem: ...
+    def set_retry_policy(self, retry_policy: modal_proto.api_pb2.FunctionRetryPolicy): ...
     def handle_retry_response(self, input_jwt: str): ...
     async def create_map_start_or_continue_item(self, idx: int) -> modal_proto.api_pb2.MapStartOrContinueItem: ...
 
@@ -422,11 +423,13 @@ class _MapItemsManager:
         """Initialize self.  See help(type(self)) for accurate signature."""
         ...
 
+    def set_retry_policy(self, retry_policy: modal_proto.api_pb2.FunctionRetryPolicy): ...
     async def add_items(self, items: list[modal_proto.api_pb2.FunctionPutInputsItem]): ...
     async def add_items_inputplane(self, items: list[modal_proto.api_pb2.MapStartOrContinueItem]): ...
     async def prepare_items_for_retry(
         self, retriable_idxs: list[int]
     ) -> list[modal_proto.api_pb2.FunctionRetryInputsItem]: ...
+    def update_items_retry_policy(self, retry_policy: modal_proto.api_pb2.FunctionRetryPolicy): ...
     def get_input_jwts_waiting_for_output(self) -> list[str]:
         """Returns a list of input_jwts for inputs that are waiting for output."""
         ...

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/modal.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.1.3.dev2
+Version: 1.1.3.dev4
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/modal_proto/api.proto

@@ -618,7 +618,7 @@ message AttemptStartRequest {
 
 message AttemptStartResponse {
   string attempt_token = 1;
-  FunctionRetryPolicy retry_policy = 2; // TODO(ben-okeefe) TODO(nathan): Not currently used
+  FunctionRetryPolicy retry_policy = 2;
 }
 
 message AuthTokenGetRequest {
@@ -2167,7 +2167,7 @@ message MapStartOrContinueResponse {
   string function_call_id = 2;
   uint32 max_inputs_outstanding = 3;
   repeated string attempt_tokens = 4;
-  FunctionRetryPolicy retry_policy = 5; // TODO(ben-okeefe): Not currently used
+  FunctionRetryPolicy retry_policy = 5;
 }
 
 message MethodDefinition {

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/modal_proto/api_pb2.pyi

@@ -1959,8 +1959,7 @@ class AttemptStartResponse(google.protobuf.message.Message):
     RETRY_POLICY_FIELD_NUMBER: builtins.int
     attempt_token: builtins.str
     @property
-    def retry_policy(self) -> global___FunctionRetryPolicy:
-        """TODO(ben-okeefe) TODO(nathan): Not currently used"""
+    def retry_policy(self) -> global___FunctionRetryPolicy: ...
     def __init__(
         self,
         *,
@@ -7007,8 +7006,7 @@ class MapStartOrContinueResponse(google.protobuf.message.Message):
     @property
     def attempt_tokens(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]: ...
     @property
-    def retry_policy(self) -> global___FunctionRetryPolicy:
-        """TODO(ben-okeefe): Not currently used"""
+    def retry_policy(self) -> global___FunctionRetryPolicy: ...
     def __init__(
         self,
         *,

{modal-1.1.3.dev2 → modal-1.1.3.dev4}/modal_version/__init__.py

@@ -1,4 +1,4 @@
 # Copyright Modal Labs 2025
 """Supplies the current version of the modal client library."""
 
-__version__ = "1.1.3.dev2"
+__version__ = "1.1.3.dev4"