modal 1.1.4.dev16__tar.gz → 1.1.4.dev18__tar.gz

{modal-1.1.4.dev16 → modal-1.1.4.dev18}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.1.4.dev16
+Version: 1.1.4.dev18
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.1.4.dev16 → modal-1.1.4.dev18}/modal/_functions.py

@@ -1471,6 +1471,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
         self._info = None
         self._serve_mounts = frozenset()
         self._metadata = None
+        self._experimental_flash_urls = None
 
     def _hydrate_metadata(self, metadata: Optional[Message]):
         # Overridden concrete implementation of base class method
@@ -1498,6 +1499,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
         self._max_object_size_bytes = (
             metadata.max_object_size_bytes if metadata.HasField("max_object_size_bytes") else MAX_OBJECT_SIZE_BYTES
         )
+        self._experimental_flash_urls = metadata._experimental_flash_urls
 
     def _get_metadata(self):
         # Overridden concrete implementation of base class method
@@ -1515,6 +1517,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
             input_plane_url=self._input_plane_url,
             input_plane_region=self._input_plane_region,
             max_object_size_bytes=self._max_object_size_bytes,
+            _experimental_flash_urls=self._experimental_flash_urls,
         )
 
     def _check_no_web_url(self, fn_name: str):
@@ -1545,6 +1548,11 @@ Use the `Function.get_web_url()` method instead.
         """URL of a Function running as a web endpoint."""
         return self._web_url
 
+    @live_method
+    async def _experimental_get_flash_urls(self) -> Optional[list[str]]:
+        """URL of the flash service for the function."""
+        return list(self._experimental_flash_urls) if self._experimental_flash_urls else None
+
     @property
     async def is_generator(self) -> bool:
         """mdmd:hidden"""

{modal-1.1.4.dev16 → modal-1.1.4.dev18}/modal/client.pyi

@@ -33,7 +33,7 @@ class _Client:
         server_url: str,
         client_type: int,
         credentials: typing.Optional[tuple[str, str]],
-        version: str = "1.1.4.dev16",
+        version: str = "1.1.4.dev18",
     ):
         """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 = "1.1.4.dev16",
+        version: str = "1.1.4.dev18",
     ):
         """mdmd:hidden
         The Modal client object is not intended to be instantiated directly by users.

{modal-1.1.4.dev16 → modal-1.1.4.dev18}/modal/cls.py

@@ -12,7 +12,7 @@ from grpclib import GRPCError, Status
 from modal_proto import api_pb2
 
 from ._functions import _Function, _parse_retries
-from ._object import _Object
+from ._object import _Object, live_method
 from ._partial_function import (
     _find_callables_for_obj,
     _find_partial_methods_for_user_cls,
@@ -510,6 +510,11 @@ class _Cls(_Object, type_prefix="cs"):
         # returns method names for a *local* class only for now (used by cli)
         return self._method_partials.keys()
 
+    @live_method
+    async def _experimental_get_flash_urls(self) -> Optional[list[str]]:
+        """URL of the flash service for the class."""
+        return await self._get_class_service_function()._experimental_get_flash_urls()
+
     def _hydrate_metadata(self, metadata: Message):
         assert isinstance(metadata, api_pb2.ClassHandleMetadata)
         class_service_function = self._get_class_service_function()

{modal-1.1.4.dev16 → modal-1.1.4.dev18}/modal/cls.pyi

@@ -354,6 +354,10 @@ class _Cls(modal._object._Object):
     def _get_name(self) -> str: ...
     def _get_class_service_function(self) -> modal._functions._Function: ...
     def _get_method_names(self) -> collections.abc.Collection[str]: ...
+    async def _experimental_get_flash_urls(self) -> typing.Optional[list[str]]:
+        """URL of the flash service for the class."""
+        ...
+
     def _hydrate_metadata(self, metadata: google.protobuf.message.Message): ...
     @staticmethod
     def validate_construction_mechanism(user_cls):
@@ -520,6 +524,18 @@ class Cls(modal.object.Object):
     def _get_name(self) -> str: ...
     def _get_class_service_function(self) -> modal.functions.Function: ...
     def _get_method_names(self) -> collections.abc.Collection[str]: ...
+
+    class ___experimental_get_flash_urls_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(self, /) -> typing.Optional[list[str]]:
+            """URL of the flash service for the class."""
+            ...
+
+        async def aio(self, /) -> typing.Optional[list[str]]:
+            """URL of the flash service for the class."""
+            ...
+
+    _experimental_get_flash_urls: ___experimental_get_flash_urls_spec[typing_extensions.Self]
+
     def _hydrate_metadata(self, metadata: google.protobuf.message.Message): ...
     @staticmethod
     def validate_construction_mechanism(user_cls):

{modal-1.1.4.dev16 → modal-1.1.4.dev18}/modal/experimental/flash.py

@@ -213,8 +213,8 @@ class _FlashPrometheusAutoscaler:
                 )
 
                 logger.warning(
-                    f"[Modal Flash] Scaling to {actual_target_containers} containers. Autoscaling decision "
-                    f"made in {time.time() - autoscaling_time} seconds."
+                    f"[Modal Flash] Scaling to {actual_target_containers=} containers. "
+                    f" Autoscaling decision made in {time.time() - autoscaling_time} seconds."
                 )
 
                 await self.autoscaling_decisions_dict.put(
@@ -223,9 +223,7 @@ class _FlashPrometheusAutoscaler:
                 )
                 await self.autoscaling_decisions_dict.put("current_replicas", actual_target_containers)
 
-                await self.cls.update_autoscaler(
-                    min_containers=actual_target_containers,
-                )
+                await self.cls.update_autoscaler(min_containers=actual_target_containers)
 
                 if time.time() - autoscaling_time < self.autoscaling_interval_seconds:
                     await asyncio.sleep(self.autoscaling_interval_seconds - (time.time() - autoscaling_time))
@@ -239,6 +237,8 @@ class _FlashPrometheusAutoscaler:
                 await asyncio.sleep(self.autoscaling_interval_seconds)
 
     async def _compute_target_containers(self, current_replicas: int) -> int:
+        # current_replicas is the number of live containers + cold starting containers (not yet live)
+        # containers is the number of live containers that are registered in flash dns
         containers = await self._get_all_containers()
         if len(containers) > current_replicas:
             logger.info(
@@ -271,11 +271,17 @@ class _FlashPrometheusAutoscaler:
             sum_metric += container_metrics[target_metric][0].value
             containers_with_metrics += 1
 
+        # n_containers_missing_metric is the number of unhealthy containers + number of cold starting containers
         n_containers_missing_metric = current_replicas - containers_with_metrics
+        # n_containers_unhealthy is the number of live containers that are not emitting metrics i.e. unhealthy
+        n_containers_unhealthy = len(containers) - containers_with_metrics
+
+        # Scale up assuming that every unhealthy container is at 2x the target metric value.
+        scale_up_target_metric_value = (sum_metric + n_containers_unhealthy * target_metric_value) / (
+            (containers_with_metrics + n_containers_unhealthy) or 1
+        )
 
-        # Scale up / down conservatively: Any container that is missing the metric is assumed to be at the minimum
-        # value of the metric when scaling up and the maximum value of the metric when scaling down.
-        scale_up_target_metric_value = sum_metric / (containers_with_metrics or 1)
+        # Scale down assuming that every container (including cold starting containers) are at the target metric value.
         scale_down_target_metric_value = (
             sum_metric + n_containers_missing_metric * target_metric_value
         ) / current_replicas
@@ -290,9 +296,14 @@ class _FlashPrometheusAutoscaler:
             desired_replicas = math.ceil(current_replicas * scale_down_ratio)
 
         logger.warning(
-            f"[Modal Flash] Current replicas: {current_replicas}, target metric value: {target_metric_value}, "
-            f"current sum of metric values: {sum_metric}, number of containers missing metric: "
-            f"{n_containers_missing_metric}, scale up ratio: {scale_up_ratio}, scale down ratio: {scale_down_ratio}, "
+            f"[Modal Flash] Current replicas: {current_replicas}, "
+            f"target metric value: {target_metric_value}, "
+            f"current sum of metric values: {sum_metric}, "
+            f"number of containers with metrics: {containers_with_metrics}, "
+            f"number of containers unhealthy: {n_containers_unhealthy}, "
+            f"number of containers missing metric (includes unhealthy): {n_containers_missing_metric}, "
+            f"scale up ratio: {scale_up_ratio}, "
+            f"scale down ratio: {scale_down_ratio}, "
             f"desired replicas: {desired_replicas}"
         )
 
@@ -312,9 +323,19 @@ class _FlashPrometheusAutoscaler:
             logger.warning(f"[Modal Flash] Error getting metrics from {url}: {e}")
             return None
 
+        # Read body with timeout/error handling and parse Prometheus metrics
+        try:
+            text_body = await response.text()
+        except asyncio.TimeoutError:
+            logger.warning(f"[Modal Flash] Timeout reading metrics body from {url}")
+            return None
+        except Exception as e:
+            logger.warning(f"[Modal Flash] Error reading metrics body from {url}: {e}")
+            return None
+
         # Parse the text-based Prometheus metrics format
         metrics: dict[str, list[Sample]] = defaultdict(list)
-        for family in text_string_to_metric_families(await response.text()):
+        for family in text_string_to_metric_families(text_body):
             for sample in family.samples:
                 metrics[sample.name] += [sample]
 

{modal-1.1.4.dev16 → modal-1.1.4.dev18}/modal/functions.pyi

@@ -359,6 +359,17 @@ class Function(
 
     get_web_url: __get_web_url_spec[typing_extensions.Self]
 
+    class ___experimental_get_flash_urls_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(self, /) -> typing.Optional[list[str]]:
+            """URL of the flash service for the function."""
+            ...
+
+        async def aio(self, /) -> typing.Optional[list[str]]:
+            """URL of the flash service for the function."""
+            ...
+
+    _experimental_get_flash_urls: ___experimental_get_flash_urls_spec[typing_extensions.Self]
+
     @property
     def is_generator(self) -> bool:
         """mdmd:hidden"""
@@ -433,7 +444,7 @@ 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 +453,7 @@ 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 +480,7 @@ 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 +504,7 @@ 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 +513,7 @@ 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 +534,7 @@ 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-1.1.4.dev16 → modal-1.1.4.dev18}/modal.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 1.1.4.dev16
+Version: 1.1.4.dev18
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-1.1.4.dev16 → modal-1.1.4.dev18}/modal_version/__init__.py

@@ -1,4 +1,4 @@
 # Copyright Modal Labs 2025
 """Supplies the current version of the modal client library."""
 
-__version__ = "1.1.4.dev16"
+__version__ = "1.1.4.dev18"