modal 0.73.94__tar.gz → 0.73.96__tar.gz

{modal-0.73.94 → modal-0.73.96}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: modal
-Version: 0.73.94
+Version: 0.73.96
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-0.73.94 → modal-0.73.96}/modal/_utils/async_utils.py

@@ -385,6 +385,10 @@ class AsyncOrSyncIterable:
         except NestedEventLoops:
             raise InvalidError(self.nested_async_message)
 
+    async def aclose(self):
+        if hasattr(self._async_iterable, "aclose"):
+            await self._async_iterable.aclose()
+
 
 _shutdown_tasks = []
 

{modal-0.73.94 → modal-0.73.96}/modal/_utils/grpc_utils.py

@@ -8,6 +8,7 @@ import typing
 import urllib.parse
 import uuid
 from collections.abc import AsyncIterator
+from dataclasses import dataclass
 from typing import (
     Any,
     Optional,
@@ -68,6 +69,11 @@ RETRYABLE_GRPC_STATUS_CODES = [
     Status.INTERNAL,
 ]
 
+@dataclass
+class RetryWarningMessage:
+    message: str
+    warning_interval: int
+    errors_to_warn_for: typing.List[Status]
 
 def create_channel(
     server_url: str,
@@ -144,6 +150,7 @@ async def retry_transient_errors(
     attempt_timeout: Optional[float] = None,  # timeout for each attempt
     total_timeout: Optional[float] = None,  # timeout for the entire function call
     attempt_timeout_floor=2.0,  # always have at least this much timeout (only for total_timeout)
+    retry_warning_message: Optional[RetryWarningMessage] = None
 ) -> ResponseType:
     """Retry on transient gRPC failures with back-off until max_retries is reached.
     If max_retries is None, retry forever."""
@@ -212,6 +219,9 @@ async def retry_transient_errors(
 
             n_retries += 1
 
+            if retry_warning_message and n_retries % retry_warning_message.warning_interval == 0:
+                logger.warning(retry_warning_message.message)
+
             await asyncio.sleep(delay)
             delay = min(delay * delay_factor, max_delay)
 

{modal-0.73.94 → modal-0.73.96}/modal/client.pyi

@@ -27,7 +27,7 @@ class _Client:
     _snapshotted: bool
 
     def __init__(
-        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.73.94"
+        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.73.96"
     ): ...
     def is_closed(self) -> bool: ...
     @property
@@ -85,7 +85,7 @@ class Client:
     _snapshotted: bool
 
     def __init__(
-        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.73.94"
+        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.73.96"
     ): ...
     def is_closed(self) -> bool: ...
     @property

{modal-0.73.94 → modal-0.73.96}/modal/parallel_map.py

@@ -5,7 +5,7 @@ import typing
 from dataclasses import dataclass
 from typing import Any, Callable, Optional
 
-from grpclib import GRPCError, Status
+from grpclib import Status
 
 from modal._runtime.execution_context import current_input_id
 from modal._utils.async_utils import (
@@ -27,13 +27,17 @@ from modal._utils.function_utils import (
     _create_input,
     _process_result,
 )
-from modal._utils.grpc_utils import retry_transient_errors
+from modal._utils.grpc_utils import RETRYABLE_GRPC_STATUS_CODES, RetryWarningMessage, retry_transient_errors
 from modal.config import logger
 from modal_proto import api_pb2
 
 if typing.TYPE_CHECKING:
     import modal.client
 
+# pump_inputs should retry if it receives any of the standard retryable codes plus RESOURCE_EXHAUSTED.
+PUMP_INPUTS_RETRYABLE_GRPC_STATUS_CODES = RETRYABLE_GRPC_STATUS_CODES + [Status.RESOURCE_EXHAUSTED]
+PUMP_INPUTS_MAX_RETRIES = 8
+PUMP_INPUTS_MAX_RETRY_DELAY=15
 
 class _SynchronizedQueue:
     """mdmd:hidden"""
@@ -136,25 +140,19 @@ async def _map_invocation(
             logger.debug(
                 f"Pushing {len(items)} inputs to server. Num queued inputs awaiting push is {input_queue.qsize()}."
             )
-            while True:
-                try:
-                    resp = await retry_transient_errors(
-                        client.stub.FunctionPutInputs,
-                        request,
-                        # with 8 retries we log the warning below about every 30 secondswhich isn't too spammy.
-                        max_retries=8,
-                        max_delay=15,
-                        additional_status_codes=[Status.RESOURCE_EXHAUSTED],
-                    )
-                    break
-                except GRPCError as err:
-                    if err.status != Status.RESOURCE_EXHAUSTED:
-                        raise err
-                    logger.warning(
-                        f"Warning: map progress for function {function._function_name} is limited."
-                        " Common bottlenecks include slow iteration over results, or function backlogs."
-                    )
-
+            # with 8 retries we log the warning below about every 30 seconds which isn't too spammy.
+            retry_warning_message = RetryWarningMessage(
+                message=f"Warning: map progress for function {function._function_name} is limited."
+                " Common bottlenecks include slow iteration over results, or function backlogs.",
+                warning_interval=8,
+                errors_to_warn_for=[Status.RESOURCE_EXHAUSTED])
+            resp = await retry_transient_errors(
+                client.stub.FunctionPutInputs,
+                request,
+                max_retries=None,
+                max_delay=PUMP_INPUTS_MAX_RETRY_DELAY,
+                additional_status_codes=[Status.RESOURCE_EXHAUSTED],
+                retry_warning_message=retry_warning_message)
             count_update()
             for item in resp.inputs:
                 pending_outputs.setdefault(item.input_id, 0)
@@ -312,7 +310,7 @@ def _map_sync(
 
 @warn_if_generator_is_not_consumed(function_name="Function.map.aio")
 async def _map_async(
-    self,
+    self: "modal.functions.Function",
     *input_iterators: typing.Union[
         typing.Iterable[Any], typing.AsyncIterable[Any]
     ],  # one input iterator per argument in the mapped-over function/generator
@@ -339,19 +337,20 @@ async def _map_async(
             async for args in streamer:
                 await raw_input_queue.put.aio((args, kwargs))
         await raw_input_queue.put.aio(None)  # end-of-input sentinel
-
-    feed_input_task = asyncio.create_task(feed_queue())
-
-    try:
-        # note that `map()` and `map.aio()` are not synchronicity-wrapped, since
-        # they accept executable code in the form of
-        # iterators that we don't want to run inside the synchronicity thread.
-        # Instead, we delegate to `._map()` with a safer Queue as input
-        async with aclosing(self._map.aio(raw_input_queue, order_outputs, return_exceptions)) as map_output_stream:
-            async for output in map_output_stream:
-                yield output
-    finally:
-        feed_input_task.cancel()  # should only be needed in case of exceptions
+        if False:
+            # make this a never yielding generator so we can async_merge it below
+            # this is important so any exception raised in feed_queue will be propagated
+            yield
+
+    # note that `map()` and `map.aio()` are not synchronicity-wrapped, since
+    # they accept executable code in the form of
+    # iterators that we don't want to run inside the synchronicity thread.
+    # Instead, we delegate to `._map()` with a safer Queue as input
+    async with aclosing(
+        async_merge(self._map.aio(raw_input_queue, order_outputs, return_exceptions), feed_queue())
+    ) as map_output_stream:
+        async for output in map_output_stream:
+            yield output
 
 
 def _for_each_sync(self, *input_iterators, kwargs={}, ignore_exceptions: bool = False):

{modal-0.73.94 → modal-0.73.96}/modal/parallel_map.pyi

@@ -2,6 +2,7 @@ import collections.abc
 import modal._functions
 import modal._utils.async_utils
 import modal.client
+import modal.functions
 import typing
 import typing_extensions
 
@@ -52,7 +53,7 @@ def _map_sync(
     self, *input_iterators, kwargs={}, order_outputs: bool = True, return_exceptions: bool = False
 ) -> modal._utils.async_utils.AsyncOrSyncIterable: ...
 def _map_async(
-    self,
+    self: modal.functions.Function,
     *input_iterators: typing.Union[typing.Iterable[typing.Any], typing.AsyncIterable[typing.Any]],
     kwargs={},
     order_outputs: bool = True,

{modal-0.73.94 → modal-0.73.96}/modal.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: modal
-Version: 0.73.94
+Version: 0.73.96
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-0.73.94 → modal-0.73.96}/modal_version/_version_generated.py

@@ -1,4 +1,4 @@
 # Copyright Modal Labs 2025
 
 # Note: Reset this value to -1 whenever you make a minor `0.X` release of the client.
-build_number = 94  # git: 3f8ec72
+build_number = 96  # git: 9b83eca