modal 1.0.6.dev58__py3-none-any.whl → 1.2.3.dev7__py3-none-any.whl

modal/experimental/flash.pyi

@@ -1,13 +1,27 @@
 import modal.client
+import subprocess
 import typing
 import typing_extensions
 
 class _FlashManager:
-    def __init__(self, client: modal.client._Client, port: int, health_check_url: typing.Optional[str] = None):
+    def __init__(
+        self,
+        client: modal.client._Client,
+        port: int,
+        process: typing.Optional[subprocess.Popen] = None,
+        health_check_url: typing.Optional[str] = None,
+    ):
         """Initialize self.  See help(type(self)) for accurate signature."""
         ...
 
+    async def is_port_connection_healthy(
+        self, process: typing.Optional[subprocess.Popen], timeout: float = 0.5
+    ) -> tuple[bool, typing.Optional[Exception]]: ...
     async def _start(self): ...
+    async def _drain_container(self):
+        """Background task that checks if we've encountered too many failures and drains the container if so."""
+        ...
+
     async def _run_heartbeat(self, host: str, port: int): ...
     def get_container_url(self): ...
     async def stop(self): ...
@@ -16,7 +30,23 @@ class _FlashManager:
 SUPERSELF = typing.TypeVar("SUPERSELF", covariant=True)
 
 class FlashManager:
-    def __init__(self, client: modal.client.Client, port: int, health_check_url: typing.Optional[str] = None): ...
+    def __init__(
+        self,
+        client: modal.client.Client,
+        port: int,
+        process: typing.Optional[subprocess.Popen] = None,
+        health_check_url: typing.Optional[str] = None,
+    ): ...
+
+    class __is_port_connection_healthy_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(
+            self, /, process: typing.Optional[subprocess.Popen], timeout: float = 0.5
+        ) -> tuple[bool, typing.Optional[Exception]]: ...
+        async def aio(
+            self, /, process: typing.Optional[subprocess.Popen], timeout: float = 0.5
+        ) -> tuple[bool, typing.Optional[Exception]]: ...
+
+    is_port_connection_healthy: __is_port_connection_healthy_spec[typing_extensions.Self]
 
     class ___start_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, /): ...
@@ -24,6 +54,17 @@ class FlashManager:
 
     _start: ___start_spec[typing_extensions.Self]
 
+    class ___drain_container_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(self, /):
+            """Background task that checks if we've encountered too many failures and drains the container if so."""
+            ...
+
+        async def aio(self, /):
+            """Background task that checks if we've encountered too many failures and drains the container if so."""
+            ...
+
+    _drain_container: ___drain_container_spec[typing_extensions.Self]
+
     class ___run_heartbeat_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, /, host: str, port: int): ...
         async def aio(self, /, host: str, port: int): ...
@@ -45,17 +86,27 @@ class FlashManager:
     close: __close_spec[typing_extensions.Self]
 
 class __flash_forward_spec(typing_extensions.Protocol):
-    def __call__(self, /, port: int, health_check_url: typing.Optional[str] = None) -> FlashManager:
+    def __call__(
+        self,
+        /,
+        port: int,
+        process: typing.Optional[subprocess.Popen] = None,
+        health_check_url: typing.Optional[str] = None,
+    ) -> FlashManager:
         """Forward a port to the Modal Flash service, exposing that port as a stable web endpoint.
-
         This is a highly experimental method that can break or be removed at any time without warning.
         Do not use this method unless explicitly instructed to do so by Modal support.
         """
         ...
 
-    async def aio(self, /, port: int, health_check_url: typing.Optional[str] = None) -> FlashManager:
+    async def aio(
+        self,
+        /,
+        port: int,
+        process: typing.Optional[subprocess.Popen] = None,
+        health_check_url: typing.Optional[str] = None,
+    ) -> FlashManager:
         """Forward a port to the Modal Flash service, exposing that port as a stable web endpoint.
-
         This is a highly experimental method that can break or be removed at any time without warning.
         Do not use this method unless explicitly instructed to do so by Modal support.
         """
@@ -74,6 +125,7 @@ class _FlashPrometheusAutoscaler:
         target_metric_value: float,
         min_containers: typing.Optional[int],
         max_containers: typing.Optional[int],
+        buffer_containers: typing.Optional[int],
         scale_up_tolerance: float,
         scale_down_tolerance: float,
         scale_up_stabilization_window_seconds: int,
@@ -85,9 +137,28 @@ class _FlashPrometheusAutoscaler:
 
     async def start(self): ...
     async def _run_autoscaler_loop(self): ...
-    async def _compute_target_containers(self, current_replicas: int) -> int: ...
+    async def _compute_target_containers(self, current_replicas: int) -> int:
+        """Gets metrics from container to autoscale up or down."""
+        ...
+
+    def _calculate_desired_replicas(
+        self,
+        n_current_replicas: int,
+        sum_metric: float,
+        n_containers_with_metrics: int,
+        n_total_containers: int,
+        target_metric_value: float,
+    ) -> int:
+        """Calculate the desired number of replicas to autoscale to."""
+        ...
+
+    async def _get_scaling_info(self, containers) -> tuple[float, int]:
+        """Get metrics using container exposed metrics endpoints."""
+        ...
+
     async def _get_metrics(self, url: str) -> typing.Optional[dict[str, list[typing.Any]]]: ...
     async def _get_all_containers(self): ...
+    async def _set_target_slots(self, target_slots: int): ...
     def _make_scaling_decision(
         self,
         current_replicas: int,
@@ -96,6 +167,7 @@ class _FlashPrometheusAutoscaler:
         scale_down_stabilization_window_seconds: int = 300,
         min_containers: typing.Optional[int] = None,
         max_containers: typing.Optional[int] = None,
+        buffer_containers: typing.Optional[int] = None,
     ) -> int:
         """Return the target number of containers following (simplified) Kubernetes HPA
         stabilization-window semantics.
@@ -128,6 +200,7 @@ class FlashPrometheusAutoscaler:
         target_metric_value: float,
         min_containers: typing.Optional[int],
         max_containers: typing.Optional[int],
+        buffer_containers: typing.Optional[int],
         scale_up_tolerance: float,
         scale_down_tolerance: float,
         scale_up_stabilization_window_seconds: int,
@@ -148,11 +221,38 @@ class FlashPrometheusAutoscaler:
     _run_autoscaler_loop: ___run_autoscaler_loop_spec[typing_extensions.Self]
 
     class ___compute_target_containers_spec(typing_extensions.Protocol[SUPERSELF]):
-        def __call__(self, /, current_replicas: int) -> int: ...
-        async def aio(self, /, current_replicas: int) -> int: ...
+        def __call__(self, /, current_replicas: int) -> int:
+            """Gets metrics from container to autoscale up or down."""
+            ...
+
+        async def aio(self, /, current_replicas: int) -> int:
+            """Gets metrics from container to autoscale up or down."""
+            ...
 
     _compute_target_containers: ___compute_target_containers_spec[typing_extensions.Self]
 
+    def _calculate_desired_replicas(
+        self,
+        n_current_replicas: int,
+        sum_metric: float,
+        n_containers_with_metrics: int,
+        n_total_containers: int,
+        target_metric_value: float,
+    ) -> int:
+        """Calculate the desired number of replicas to autoscale to."""
+        ...
+
+    class ___get_scaling_info_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(self, /, containers) -> tuple[float, int]:
+            """Get metrics using container exposed metrics endpoints."""
+            ...
+
+        async def aio(self, /, containers) -> tuple[float, int]:
+            """Get metrics using container exposed metrics endpoints."""
+            ...
+
+    _get_scaling_info: ___get_scaling_info_spec[typing_extensions.Self]
+
     class ___get_metrics_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, /, url: str) -> typing.Optional[dict[str, list[typing.Any]]]: ...
         async def aio(self, /, url: str) -> typing.Optional[dict[str, list[typing.Any]]]: ...
@@ -165,6 +265,12 @@ class FlashPrometheusAutoscaler:
 
     _get_all_containers: ___get_all_containers_spec[typing_extensions.Self]
 
+    class ___set_target_slots_spec(typing_extensions.Protocol[SUPERSELF]):
+        def __call__(self, /, target_slots: int): ...
+        async def aio(self, /, target_slots: int): ...
+
+    _set_target_slots: ___set_target_slots_spec[typing_extensions.Self]
+
     def _make_scaling_decision(
         self,
         current_replicas: int,
@@ -173,6 +279,7 @@ class FlashPrometheusAutoscaler:
         scale_down_stabilization_window_seconds: int = 300,
         min_containers: typing.Optional[int] = None,
         max_containers: typing.Optional[int] = None,
+        buffer_containers: typing.Optional[int] = None,
     ) -> int:
         """Return the target number of containers following (simplified) Kubernetes HPA
         stabilization-window semantics.
@@ -214,6 +321,7 @@ class __flash_prometheus_autoscaler_spec(typing_extensions.Protocol):
         scale_up_stabilization_window_seconds: int = 0,
         scale_down_stabilization_window_seconds: int = 300,
         autoscaling_interval_seconds: int = 15,
+        buffer_containers: typing.Optional[int] = None,
     ) -> FlashPrometheusAutoscaler:
         """Autoscale a Flash service based on containers' Prometheus metrics.
 
@@ -239,6 +347,7 @@ class __flash_prometheus_autoscaler_spec(typing_extensions.Protocol):
         scale_up_stabilization_window_seconds: int = 0,
         scale_down_stabilization_window_seconds: int = 300,
         autoscaling_interval_seconds: int = 15,
+        buffer_containers: typing.Optional[int] = None,
     ) -> FlashPrometheusAutoscaler:
         """Autoscale a Flash service based on containers' Prometheus metrics.
 
@@ -250,3 +359,22 @@ class __flash_prometheus_autoscaler_spec(typing_extensions.Protocol):
         ...
 
 flash_prometheus_autoscaler: __flash_prometheus_autoscaler_spec
+
+class __flash_get_containers_spec(typing_extensions.Protocol):
+    def __call__(self, /, app_name: str, cls_name: str) -> list[dict[str, typing.Any]]:
+        """Return a list of flash containers for a deployed Flash service.
+
+        This is a highly experimental method that can break or be removed at any time without warning.
+        Do not use this method unless explicitly instructed to do so by Modal support.
+        """
+        ...
+
+    async def aio(self, /, app_name: str, cls_name: str) -> list[dict[str, typing.Any]]:
+        """Return a list of flash containers for a deployed Flash service.
+
+        This is a highly experimental method that can break or be removed at any time without warning.
+        Do not use this method unless explicitly instructed to do so by Modal support.
+        """
+        ...
+
+flash_get_containers: __flash_get_containers_spec

modal/file_io.py

@@ -13,7 +13,6 @@ import json
 from grpclib.exceptions import GRPCError, StreamTerminatedError
 
 from modal._utils.async_utils import TaskContext
-from modal._utils.grpc_utils import retry_transient_errors
 from modal.exception import ClientClosed
 from modal_proto import api_pb2
 
@@ -57,8 +56,7 @@ async def _delete_bytes(file: "_FileIO", start: Optional[int] = None, end: Optio
     if start is not None and end is not None:
         if start >= end:
             raise ValueError("start must be less than end")
-    resp = await retry_transient_errors(
-        file._client.stub.ContainerFilesystemExec,
+    resp = await file._client.stub.ContainerFilesystemExec(
         api_pb2.ContainerFilesystemExecRequest(
             file_delete_bytes_request=api_pb2.ContainerFileDeleteBytesRequest(
                 file_descriptor=file._file_descriptor,
@@ -85,8 +83,7 @@ async def _replace_bytes(file: "_FileIO", data: bytes, start: Optional[int] = No
             raise InvalidError("start must be less than end")
     if len(data) > WRITE_CHUNK_SIZE:
         raise InvalidError("Write request payload exceeds 16 MiB limit")
-    resp = await retry_transient_errors(
-        file._client.stub.ContainerFilesystemExec,
+    resp = await file._client.stub.ContainerFilesystemExec(
         api_pb2.ContainerFilesystemExecRequest(
             file_write_replace_bytes_request=api_pb2.ContainerFileWriteReplaceBytesRequest(
                 file_descriptor=file._file_descriptor,
@@ -128,7 +125,7 @@ class _FileIO(Generic[T]):
 
     **Usage**
 
-    ```python
+    ```python notest
     import modal
 
     app = modal.App.lookup("my-app", create_if_missing=True)
@@ -261,8 +258,7 @@ class _FileIO(Generic[T]):
             raise TypeError("Expected str when in text mode")
 
     async def _open_file(self, path: str, mode: str) -> None:
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_open_request=api_pb2.ContainerFileOpenRequest(path=path, mode=mode),
                 task_id=self._task_id,
@@ -285,8 +281,7 @@ class _FileIO(Generic[T]):
         return self
 
     async def _make_read_request(self, n: Optional[int]) -> bytes:
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_read_request=api_pb2.ContainerFileReadRequest(file_descriptor=self._file_descriptor, n=n),
                 task_id=self._task_id,
@@ -309,8 +304,7 @@ class _FileIO(Generic[T]):
         """Read a single line from the current position."""
         self._check_closed()
         self._check_readable()
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_read_line_request=api_pb2.ContainerFileReadLineRequest(file_descriptor=self._file_descriptor),
                 task_id=self._task_id,
@@ -351,8 +345,7 @@ class _FileIO(Generic[T]):
             raise ValueError("Write request payload exceeds 1 GiB limit")
         for i in range(0, len(data), WRITE_CHUNK_SIZE):
             chunk = data[i : i + WRITE_CHUNK_SIZE]
-            resp = await retry_transient_errors(
-                self._client.stub.ContainerFilesystemExec,
+            resp = await self._client.stub.ContainerFilesystemExec(
                 api_pb2.ContainerFilesystemExecRequest(
                     file_write_request=api_pb2.ContainerFileWriteRequest(
                         file_descriptor=self._file_descriptor,
@@ -367,8 +360,7 @@ class _FileIO(Generic[T]):
         """Flush the buffer to disk."""
         self._check_closed()
         self._check_writable()
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_flush_request=api_pb2.ContainerFileFlushRequest(file_descriptor=self._file_descriptor),
                 task_id=self._task_id,
@@ -393,8 +385,7 @@ class _FileIO(Generic[T]):
         (relative to the current position) and 2 (relative to the file's end).
         """
         self._check_closed()
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_seek_request=api_pb2.ContainerFileSeekRequest(
                     file_descriptor=self._file_descriptor,
@@ -410,8 +401,7 @@ class _FileIO(Generic[T]):
     async def ls(cls, path: str, client: _Client, task_id: str) -> list[str]:
         """List the contents of the provided directory."""
         self = _FileIO(client, task_id)
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_ls_request=api_pb2.ContainerFileLsRequest(path=path),
                 task_id=task_id,
@@ -427,8 +417,7 @@ class _FileIO(Generic[T]):
     async def mkdir(cls, path: str, client: _Client, task_id: str, parents: bool = False) -> None:
         """Create a new directory."""
         self = _FileIO(client, task_id)
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_mkdir_request=api_pb2.ContainerFileMkdirRequest(path=path, make_parents=parents),
                 task_id=self._task_id,
@@ -440,8 +429,7 @@ class _FileIO(Generic[T]):
     async def rm(cls, path: str, client: _Client, task_id: str, recursive: bool = False) -> None:
         """Remove a file or directory in the Sandbox."""
         self = _FileIO(client, task_id)
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_rm_request=api_pb2.ContainerFileRmRequest(path=path, recursive=recursive),
                 task_id=self._task_id,
@@ -460,8 +448,7 @@ class _FileIO(Generic[T]):
         timeout: Optional[int] = None,
     ) -> AsyncIterator[FileWatchEvent]:
         self = _FileIO(client, task_id)
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_watch_request=api_pb2.ContainerFileWatchRequest(
                     path=path,
@@ -503,8 +490,7 @@ class _FileIO(Generic[T]):
 
     async def _close(self) -> None:
         # Buffer is flushed by the runner on close
-        resp = await retry_transient_errors(
-            self._client.stub.ContainerFilesystemExec,
+        resp = await self._client.stub.ContainerFilesystemExec(
             api_pb2.ContainerFilesystemExecRequest(
                 file_close_request=api_pb2.ContainerFileCloseRequest(file_descriptor=self._file_descriptor),
                 task_id=self._task_id,

modal/file_io.pyi

@@ -62,7 +62,7 @@ class _FileIO(typing.Generic[T]):
 
     **Usage**
 
-    ```python
+    ```python notest
     import modal
 
     app = modal.App.lookup("my-app", create_if_missing=True)
@@ -232,7 +232,7 @@ class FileIO(typing.Generic[T]):
 
     **Usage**
 
-    ```python
+    ```python notest
     import modal
 
     app = modal.App.lookup("my-app", create_if_missing=True)

modal/file_pattern_matcher.py

@@ -11,6 +11,7 @@ then asking it whether file paths match any of its patterns.
 
 import os
 from abc import abstractmethod
+from functools import cached_property
 from pathlib import Path
 from typing import Callable, Optional, Sequence, Union
 
@@ -99,11 +100,11 @@ class FilePatternMatcher(_AbstractPatternMatcher):
     ```
     """
 
-    patterns: list[Pattern]
-    _delayed_init: Callable[[], None] = None
+    _file_path: Optional[Union[str, Path]]
+    _pattern_strings: Optional[Sequence[str]]
 
-    def _set_patterns(self, patterns: Sequence[str]) -> None:
-        self.patterns = []
+    def _parse_patterns(self, patterns: Sequence[str]) -> list[Pattern]:
+        parsed_patterns = []
         for pattern in list(patterns):
             pattern = pattern.strip().strip(os.path.sep)
             if not pattern:
@@ -118,7 +119,8 @@ class FilePatternMatcher(_AbstractPatternMatcher):
             # In Python, we can proceed without explicit syntax checking
             new_pattern.cleaned_pattern = pattern
             new_pattern.dirs = pattern.split(os.path.sep)
-            self.patterns.append(new_pattern)
+            parsed_patterns.append(new_pattern)
+        return parsed_patterns
 
     def __init__(self, *pattern: str) -> None:
         """Initialize a new FilePatternMatcher instance.
@@ -129,7 +131,8 @@ class FilePatternMatcher(_AbstractPatternMatcher):
         Raises:
             ValueError: If an illegal exclusion pattern is provided.
         """
-        self._set_patterns(pattern)
+        self._pattern_strings = pattern
+        self._file_path = None
 
     @classmethod
     def from_file(cls, file_path: Union[str, Path]) -> "FilePatternMatcher":
@@ -148,14 +151,10 @@ class FilePatternMatcher(_AbstractPatternMatcher):
         ```
 
         """
-        uninitialized = cls.__new__(cls)
-
-        def _delayed_init():
-            uninitialized._set_patterns(Path(file_path).read_text("utf8").splitlines())
-            uninitialized._delayed_init = None
-
-        uninitialized._delayed_init = _delayed_init
-        return uninitialized
+        instance = cls.__new__(cls)
+        instance._file_path = file_path
+        instance._pattern_strings = None
+        return instance
 
     def _matches(self, file_path: str) -> bool:
         """Check if the file path or any of its parent directories match the patterns.
@@ -194,6 +193,18 @@ class FilePatternMatcher(_AbstractPatternMatcher):
 
         return matched
 
+    @cached_property
+    def patterns(self) -> list[Pattern]:
+        """Get the patterns, loading from file if necessary."""
+        if self._file_path is not None:
+            # Lazy load from file
+            pattern_strings = Path(self._file_path).read_text("utf8").splitlines()
+        else:
+            # Use patterns provided in __init__
+            pattern_strings = list(self._pattern_strings)
+
+        return self._parse_patterns(pattern_strings)
+
     def can_prune_directories(self) -> bool:
         """
         Returns True if this pattern matcher allows safe early directory pruning.
@@ -205,8 +216,6 @@ class FilePatternMatcher(_AbstractPatternMatcher):
         return not any(pattern.exclusion for pattern in self.patterns)
 
     def __call__(self, file_path: Path) -> bool:
-        if self._delayed_init:
-            self._delayed_init()
         return self._matches(str(file_path))