fal 1.4.0__tar.gz → 1.5.0__tar.gz

{fal-1.4.0 → fal-1.5.0}/PKG-INFO

@@ -1,12 +1,12 @@
 Metadata-Version: 2.1
 Name: fal
-Version: 1.4.0
+Version: 1.5.0
 Summary: fal is an easy-to-use Serverless Python Framework
 Author: Features & Labels <support@fal.ai>
 Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 Requires-Dist: isolate[build]<1.14.0,>=0.13.0
-Requires-Dist: isolate-proto==0.5.3
+Requires-Dist: isolate-proto==0.5.4
 Requires-Dist: grpcio==1.64.0
 Requires-Dist: dill==0.3.7
 Requires-Dist: cloudpickle==3.0.0

{fal-1.4.0 → fal-1.5.0}/fal.egg-info/PKG-INFO

@@ -1,12 +1,12 @@
 Metadata-Version: 2.1
 Name: fal
-Version: 1.4.0
+Version: 1.5.0
 Summary: fal is an easy-to-use Serverless Python Framework
 Author: Features & Labels <support@fal.ai>
 Requires-Python: >=3.8
 Description-Content-Type: text/markdown
 Requires-Dist: isolate[build]<1.14.0,>=0.13.0
-Requires-Dist: isolate-proto==0.5.3
+Requires-Dist: isolate-proto==0.5.4
 Requires-Dist: grpcio==1.64.0
 Requires-Dist: dill==0.3.7
 Requires-Dist: cloudpickle==3.0.0

{fal-1.4.0 → fal-1.5.0}/fal.egg-info/requires.txt

@@ -1,5 +1,5 @@
 isolate[build]<1.14.0,>=0.13.0
-isolate-proto==0.5.3
+isolate-proto==0.5.4
 grpcio==1.64.0
 dill==0.3.7
 cloudpickle==3.0.0

{fal-1.4.0 → fal-1.5.0}/pyproject.toml

@@ -23,7 +23,7 @@ readme = "README.md"
 requires-python = ">=3.8"
 dependencies = [
     "isolate[build]>=0.13.0,<1.14.0",
-    "isolate-proto==0.5.3",
+    "isolate-proto==0.5.4",
     "grpcio==1.64.0",
     "dill==0.3.7",
     "cloudpickle==3.0.0",

{fal-1.4.0 → fal-1.5.0}/src/fal/_fal_version.py

@@ -12,5 +12,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '1.4.0'
-__version_tuple__ = version_tuple = (1, 4, 0)
+__version__ = version = '1.5.0'
+__version_tuple__ = version_tuple = (1, 5, 0)

{fal-1.4.0 → fal-1.5.0}/src/fal/api.py

@@ -397,6 +397,7 @@ class FalServerlessHost(Host):
             "max_multiplexing",
             "setup_function",
             "metadata",
+            "request_timeout",
             "_base_image",
             "_scheduler",
             "_scheduler_options",
@@ -444,6 +445,7 @@ class FalServerlessHost(Host):
         scheduler = options.host.get("_scheduler", None)
         scheduler_options = options.host.get("_scheduler_options", None)
         exposed_port = options.get_exposed_port()
+        request_timeout = options.host.get("request_timeout")
 
         machine_requirements = MachineRequirements(
             machine_types=machine_type,  # type: ignore
@@ -456,6 +458,7 @@ class FalServerlessHost(Host):
             max_multiplexing=max_multiplexing,
             max_concurrency=max_concurrency,
             min_concurrency=min_concurrency,
+            request_timeout=request_timeout,
         )
 
         partial_func = _prepare_partial_func(func)
@@ -516,6 +519,7 @@ class FalServerlessHost(Host):
         scheduler_options = options.host.get("_scheduler_options", None)
         exposed_port = options.get_exposed_port()
         setup_function = options.host.get("setup_function", None)
+        request_timeout = options.host.get("request_timeout")
 
         machine_requirements = MachineRequirements(
             machine_types=machine_type,  # type: ignore
@@ -528,6 +532,7 @@ class FalServerlessHost(Host):
             max_multiplexing=max_multiplexing,
             max_concurrency=max_concurrency,
             min_concurrency=min_concurrency,
+            request_timeout=request_timeout,
         )
 
         return_value = _UNSET
@@ -693,6 +698,7 @@ def function(
     keep_alive: int = FAL_SERVERLESS_DEFAULT_KEEP_ALIVE,
     max_multiplexing: int = FAL_SERVERLESS_DEFAULT_MAX_MULTIPLEXING,
     min_concurrency: int = FAL_SERVERLESS_DEFAULT_MIN_CONCURRENCY,
+    request_timeout: int | None = None,
     setup_function: Callable[..., None] | None = None,
     _base_image: str | None = None,
     _scheduler: str | None = None,
@@ -719,6 +725,7 @@ def function(
     keep_alive: int = FAL_SERVERLESS_DEFAULT_KEEP_ALIVE,
     max_multiplexing: int = FAL_SERVERLESS_DEFAULT_MAX_MULTIPLEXING,
     min_concurrency: int = FAL_SERVERLESS_DEFAULT_MIN_CONCURRENCY,
+    request_timeout: int | None = None,
     setup_function: Callable[..., None] | None = None,
     _base_image: str | None = None,
     _scheduler: str | None = None,
@@ -795,6 +802,7 @@ def function(
     keep_alive: int = FAL_SERVERLESS_DEFAULT_KEEP_ALIVE,
     max_multiplexing: int = FAL_SERVERLESS_DEFAULT_MAX_MULTIPLEXING,
     min_concurrency: int = FAL_SERVERLESS_DEFAULT_MIN_CONCURRENCY,
+    request_timeout: int | None = None,
     setup_function: Callable[..., None] | None = None,
     _base_image: str | None = None,
     _scheduler: str | None = None,
@@ -826,6 +834,7 @@ def function(
     keep_alive: int = FAL_SERVERLESS_DEFAULT_KEEP_ALIVE,
     max_multiplexing: int = FAL_SERVERLESS_DEFAULT_MAX_MULTIPLEXING,
     min_concurrency: int = FAL_SERVERLESS_DEFAULT_MIN_CONCURRENCY,
+    request_timeout: int | None = None,
     setup_function: Callable[..., None] | None = None,
     _base_image: str | None = None,
     _scheduler: str | None = None,
@@ -851,6 +860,7 @@ def function(
     keep_alive: int = FAL_SERVERLESS_DEFAULT_KEEP_ALIVE,
     max_multiplexing: int = FAL_SERVERLESS_DEFAULT_MAX_MULTIPLEXING,
     min_concurrency: int = FAL_SERVERLESS_DEFAULT_MIN_CONCURRENCY,
+    request_timeout: int | None = None,
     setup_function: Callable[..., None] | None = None,
     _base_image: str | None = None,
     _scheduler: str | None = None,
@@ -876,6 +886,7 @@ def function(
     keep_alive: int = FAL_SERVERLESS_DEFAULT_KEEP_ALIVE,
     max_multiplexing: int = FAL_SERVERLESS_DEFAULT_MAX_MULTIPLEXING,
     min_concurrency: int = FAL_SERVERLESS_DEFAULT_MIN_CONCURRENCY,
+    request_timeout: int | None = None,
     setup_function: Callable[..., None] | None = None,
     _base_image: str | None = None,
     _scheduler: str | None = None,

{fal-1.4.0 → fal-1.5.0}/src/fal/app.py

@@ -189,11 +189,17 @@ class App(fal.api.BaseServable):
     }
     app_name: ClassVar[str]
     app_auth: ClassVar[Literal["private", "public", "shared"]] = "private"
+    version: ClassVar[str] = "unknown"
+    request_timeout: ClassVar[int | None] = None
 
     def __init_subclass__(cls, **kwargs):
         app_name = kwargs.pop("name", None) or _to_fal_app_name(cls.__name__)
         parent_settings = getattr(cls, "host_kwargs", {})
         cls.host_kwargs = {**parent_settings, **kwargs}
+
+        if cls.request_timeout is not None:
+            cls.host_kwargs["request_timeout"] = cls.request_timeout
+
         cls.app_name = getattr(cls, "app_name", app_name)
 
         if cls.__init__ is not App.__init__:
@@ -232,7 +238,7 @@ class App(fal.api.BaseServable):
             await _call_any_fn(self.teardown)
 
     def health(self):
-        return {}
+        return {"version": self.version}
 
     def setup(self):
         """Setup the application before serving."""

{fal-1.4.0 → fal-1.5.0}/src/fal/cli/_utils.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from fal.files import find_pyproject_toml, parse_pyproject_toml
+from fal.files import find_project_root, find_pyproject_toml, parse_pyproject_toml
 
 
 def is_app_name(app_ref: tuple[str, str | None]) -> bool:
@@ -29,6 +29,11 @@ def get_app_data_from_toml(app_name):
     except KeyError:
         raise ValueError(f"App {app_name} does not have a ref key in pyproject.toml")
 
+    # Convert the app_ref to a path relative to the project root
+    project_root, _ = find_project_root(None)
+    app_ref = str(project_root / app_ref)
+
     app_auth = app_data.get("auth", "private")
+    app_deployment_strategy = app_data.get("deployment_strategy", "recreate")
 
-    return app_ref, app_auth
+    return app_ref, app_auth, app_deployment_strategy

{fal-1.4.0 → fal-1.5.0}/src/fal/cli/deploy.py

@@ -65,8 +65,9 @@ def _get_user() -> User:
 def _deploy_from_reference(
     app_ref: Tuple[Optional[Union[Path, str]], ...],
     app_name: str,
-    auth: Literal["public", "shared", "private"],
     args,
+    auth: Optional[Literal["public", "shared", "private"]] = None,
+    deployment_strategy: Optional[Literal["recreate", "rolling"]] = None,
 ):
     from fal.api import FalServerlessError, FalServerlessHost
     from fal.utils import load_function_from
@@ -96,7 +97,7 @@ def _deploy_from_reference(
     isolated_function = loaded.function
     app_name = app_name or loaded.app_name  # type: ignore
     app_auth = auth or loaded.app_auth or "private"
-    deployment_strategy = args.strategy or "recreate"
+    deployment_strategy = deployment_strategy or "recreate"
 
     app_id = host.register(
         func=isolated_function.func,
@@ -137,7 +138,7 @@ def _deploy(args):
             raise ValueError("Cannot use --app-name or --auth with app name reference.")
 
         app_name = args.app_ref[0]
-        app_ref, app_auth = get_app_data_from_toml(app_name)
+        app_ref, app_auth, app_deployment_strategy = get_app_data_from_toml(app_name)
         file_path, func_name = RefAction.split_ref(app_ref)
 
     # path/to/myfile.py::MyApp
@@ -145,8 +146,15 @@ def _deploy(args):
         file_path, func_name = args.app_ref
         app_name = args.app_name
         app_auth = args.auth
-
-    _deploy_from_reference((file_path, func_name), app_name, app_auth, args)
+        app_deployment_strategy = args.strategy
+
+    _deploy_from_reference(
+        (file_path, func_name),
+        app_name,
+        args,
+        app_auth,
+        app_deployment_strategy,
+    )
 
 
 def add_parser(main_subparsers, parents):

{fal-1.4.0 → fal-1.5.0}/src/fal/cli/run.py

@@ -10,7 +10,7 @@ def _run(args):
 
     if is_app_name(args.func_ref):
         app_name = args.func_ref[0]
-        app_ref, _ = get_app_data_from_toml(app_name)
+        app_ref, _, _ = get_app_data_from_toml(app_name)
         file_path, func_name = RefAction.split_ref(app_ref)
     else:
         file_path, func_name = args.func_ref

{fal-1.4.0 → fal-1.5.0}/src/fal/sdk.py

@@ -399,6 +399,7 @@ class MachineRequirements:
     max_concurrency: int | None = None
     max_multiplexing: int | None = None
     min_concurrency: int | None = None
+    request_timeout: int | None = None
 
     def __post_init__(self):
         if isinstance(self.machine_types, str):
@@ -514,6 +515,7 @@ class FalServerlessConnection:
                 max_concurrency=machine_requirements.max_concurrency,
                 min_concurrency=machine_requirements.min_concurrency,
                 max_multiplexing=machine_requirements.max_multiplexing,
+                request_timeout=machine_requirements.request_timeout,
             )
         else:
             wrapped_requirements = None
@@ -607,6 +609,7 @@ class FalServerlessConnection:
                 max_concurrency=machine_requirements.max_concurrency,
                 max_multiplexing=machine_requirements.max_multiplexing,
                 min_concurrency=machine_requirements.min_concurrency,
+                request_timeout=machine_requirements.request_timeout,
             )
         else:
             wrapped_requirements = None

{fal-1.4.0 → fal-1.5.0}/src/fal/toolkit/file/providers/fal.py

@@ -10,6 +10,7 @@ from dataclasses import dataclass
 from datetime import datetime, timezone
 from pathlib import Path
 from urllib.error import HTTPError
+from urllib.parse import urlparse, urlunparse
 from urllib.request import Request, urlopen
 
 from fal.auth import key_credentials
@@ -24,7 +25,7 @@ _FAL_CDN = "https://fal.media"
 class FalV2Token:
     token: str
     token_type: str
-    base_url: str
+    base_upload_url: str
     expires_at: datetime
 
     def is_expired(self) -> bool:
@@ -36,7 +37,7 @@ class FalV2TokenManager:
         self._token: FalV2Token = FalV2Token(
             token="",
             token_type="",
-            base_url="",
+            base_upload_url="",
             expires_at=datetime.min.replace(tzinfo=timezone.utc),
         )
         self._lock: threading.Lock = threading.Lock()
@@ -72,10 +73,15 @@ class FalV2TokenManager:
         with urlopen(req) as response:
             result = json.load(response)
 
+        parsed_base_url = urlparse(result["base_url"])
+        base_upload_url = urlunparse(
+            parsed_base_url._replace(netloc="upload." + parsed_base_url.netloc)
+        )
+
         self._token = FalV2Token(
             token=result["token"],
             token_type=result["token_type"],
-            base_url=result["base_url"],
+            base_upload_url=base_upload_url,
             expires_at=datetime.fromisoformat(result["expires_at"]),
         )
 
@@ -179,7 +185,7 @@ class MultipartUpload:
         token = fal_v2_token_manager.get_token()
         try:
             req = Request(
-                f"{token.base_url}/upload/initiate-multipart",
+                f"{token.base_upload_url}/upload/initiate-multipart",
                 method="POST",
                 headers={
                     "Authorization": f"{token.token_type} {token.token}",
@@ -195,7 +201,7 @@ class MultipartUpload:
             )
             with urlopen(req) as response:
                 result = json.load(response)
-                self._upload_id = result["upload_id"]
+                self._upload_url = result["upload_url"]
                 self._file_url = result["file_url"]
         except HTTPError as exc:
             raise FileUploadException(
@@ -235,10 +241,7 @@ class MultipartUpload:
         ) as executor:
             futures = []
             for part_number in range(1, parts + 1):
-                upload_url = (
-                    f"{self._file_url}?upload_id={self._upload_id}"
-                    f"&part_number={part_number}"
-                )
+                upload_url = f"{self._upload_url}&part_number={part_number}"
                 futures.append(
                     executor.submit(self._upload_part, upload_url, part_number)
                 )
@@ -248,7 +251,7 @@ class MultipartUpload:
                 self._parts.append(entry)
 
     def complete(self):
-        url = f"{self._file_url}?upload_id={self._upload_id}"
+        url = self._upload_url
         try:
             req = Request(
                 url,
@@ -281,7 +284,7 @@ class FalFileRepositoryV2(FalFileRepositoryBase):
             "Content-Type": file.content_type,
         }
 
-        storage_url = f"{token.base_url}/upload"
+        storage_url = f"{token.base_upload_url}/upload"
 
         try:
             req = Request(

{fal-1.4.0 → fal-1.5.0}/tests/cli/test_deploy.py

@@ -4,6 +4,7 @@ from unittest.mock import MagicMock, patch
 import pytest
 from fal.cli.deploy import _deploy
 from fal.cli.main import parse_args
+from fal.files import find_project_root
 
 
 def test_deploy():
@@ -19,8 +20,8 @@ def mock_parse_pyproject_toml():
             "my-app": {
                 "ref": "src/my_app/inference.py::MyApp",
                 "auth": "shared",
+                "deployment_strategy": "rolling",
             },
-            # auth is not provided for another-app
             "another-app": {
                 "ref": "src/another_app/inference.py::AnotherApp",
             },
@@ -29,13 +30,17 @@ def mock_parse_pyproject_toml():
 
 
 def mock_args(
-    app_ref: tuple[str], app_name: Optional[str] = None, auth: Optional[str] = None
+    app_ref: tuple[str],
+    app_name: Optional[str] = None,
+    auth: Optional[str] = None,
+    strategy: Optional[str] = None,
 ):
     args = MagicMock()
 
     args.app_ref = app_ref
     args.app_name = app_name
     args.auth = auth
+    args.strategy = strategy
 
     return args
 
@@ -53,9 +58,15 @@ def test_deploy_with_toml_success(
 
     _deploy(args)
 
+    project_root, _ = find_project_root(None)
+
     # Ensure the correct app is deployed
     mock_deploy_ref.assert_called_once_with(
-        ("src/my_app/inference.py", "MyApp"), "my-app", "shared", args
+        (f"{project_root / 'src/my_app/inference.py'}", "MyApp"),
+        "my-app",
+        args,
+        "shared",
+        "rolling",
     )
 
 
@@ -72,9 +83,15 @@ def test_deploy_with_toml_no_auth(
 
     _deploy(args)
 
+    project_root, _ = find_project_root(None)
+
     # Since auth is not provided for "another-app", it should default to "private"
     mock_deploy_ref.assert_called_once_with(
-        ("src/another_app/inference.py", "AnotherApp"), "another-app", "private", args
+        (f"{project_root / 'src/another_app/inference.py'}", "AnotherApp"),
+        "another-app",
+        args,
+        "private",
+        "recreate",
     )
 
 
@@ -146,3 +163,91 @@ def test_deploy_with_toml_only_app_name_is_provided(
         ValueError, match="Cannot use --app-name or --auth with app name reference."
     ):
         _deploy(args)
+
+
+@patch("fal.cli._utils.find_pyproject_toml", return_value="pyproject.toml")
+@patch("fal.cli._utils.parse_pyproject_toml")
+@patch("fal.cli.deploy._deploy_from_reference")
+def test_deploy_with_toml_deployment_strategy(
+    mock_deploy_ref, mock_parse_toml, mock_find_toml, mock_parse_pyproject_toml
+):
+    mock_parse_toml.return_value = mock_parse_pyproject_toml
+
+    args = mock_args(app_ref=("my-app", None), strategy="rolling")
+
+    _deploy(args)
+
+    project_root, _ = find_project_root(None)
+
+    mock_deploy_ref.assert_called_once_with(
+        (f"{project_root / 'src/my_app/inference.py'}", "MyApp"),
+        "my-app",
+        args,
+        "shared",
+        "rolling",
+    )
+
+
+@patch("fal.cli._utils.find_pyproject_toml", return_value="pyproject.toml")
+@patch("fal.cli._utils.parse_pyproject_toml")
+@patch("fal.cli.deploy._deploy_from_reference")
+def test_deploy_with_toml_default_deployment_strategy(
+    mock_deploy_ref, mock_parse_toml, mock_find_toml, mock_parse_pyproject_toml
+):
+    mock_parse_toml.return_value = mock_parse_pyproject_toml
+
+    args = mock_args(app_ref=("another-app", None))
+
+    _deploy(args)
+
+    project_root, _ = find_project_root(None)
+
+    mock_deploy_ref.assert_called_once_with(
+        (f"{project_root / 'src/another_app/inference.py'}", "AnotherApp"),
+        "another-app",
+        args,
+        "private",
+        "recreate",
+    )
+
+
+@patch("fal.cli._utils.find_pyproject_toml", return_value="pyproject.toml")
+@patch("fal.cli._utils.parse_pyproject_toml")
+@patch("fal.cli.deploy._deploy_from_reference")
+def test_deploy_with_cli_auth(
+    mock_deploy_ref, mock_parse_toml, mock_find_toml, mock_parse_pyproject_toml
+):
+    mock_parse_toml.return_value = mock_parse_pyproject_toml
+
+    args = mock_args(app_ref=("src/my_app/inference.py", "MyApp"), auth="shared")
+
+    _deploy(args)
+
+    mock_deploy_ref.assert_called_once_with(
+        ("src/my_app/inference.py", "MyApp"),
+        None,
+        args,
+        "shared",
+        None,
+    )
+
+
+@patch("fal.cli._utils.find_pyproject_toml", return_value="pyproject.toml")
+@patch("fal.cli._utils.parse_pyproject_toml")
+@patch("fal.cli.deploy._deploy_from_reference")
+def test_deploy_with_cli_deployment_strategy(
+    mock_deploy_ref, mock_parse_toml, mock_find_toml, mock_parse_pyproject_toml
+):
+    mock_parse_toml.return_value = mock_parse_pyproject_toml
+
+    args = mock_args(app_ref=("src/my_app/inference.py", "MyApp"), strategy="rolling")
+
+    _deploy(args)
+
+    mock_deploy_ref.assert_called_once_with(
+        ("src/my_app/inference.py", "MyApp"),
+        None,
+        args,
+        None,
+        "rolling",
+    )

{fal-1.4.0 → fal-1.5.0}/tests/cli/test_run.py

@@ -3,6 +3,7 @@ from unittest.mock import MagicMock, patch
 import pytest
 from fal.cli.main import parse_args
 from fal.cli.run import _run
+from fal.files import find_project_root
 
 
 def test_run():
@@ -60,9 +61,11 @@ def test_run_with_toml_success(
 
     _run(args)
 
-    # Ensure the correct app is deployed
+    project_root, _ = find_project_root(None)
+
+    # Ensure the correct app is ran
     mock_load_function_from.assert_called_once_with(
-        host, "src/my_app/inference.py", "MyApp"
+        host, f"{project_root / 'src/my_app/inference.py'}", "MyApp"
     )