fal 1.25.0__tar.gz → 1.26.1__tar.gz

{fal-1.25.0/fal.egg-info → fal-1.26.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fal
-Version: 1.25.0
+Version: 1.26.1
 Summary: fal is an easy-to-use Serverless Python Framework
 Author: Features & Labels <support@fal.ai>
 Requires-Python: >=3.8

{fal-1.25.0 → fal-1.26.1/fal.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fal
-Version: 1.25.0
+Version: 1.26.1
 Summary: fal is an easy-to-use Serverless Python Framework
 Author: Features & Labels <support@fal.ai>
 Requires-Python: >=3.8

{fal-1.25.0 → fal-1.26.1}/src/fal/_fal_version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '1.25.0'
-__version_tuple__ = version_tuple = (1, 25, 0)
+__version__ = version = '1.26.1'
+__version_tuple__ = version_tuple = (1, 26, 1)

{fal-1.25.0 → fal-1.26.1}/src/fal/api.py

@@ -319,24 +319,26 @@ def _handle_grpc_error():
             try:
                 return fn(*args, **kwargs)
             except grpc.RpcError as e:
+                msg = e.details() or str(e)
                 if e.code() == grpc.StatusCode.UNAVAILABLE:
                     raise FalServerlessError(
                         "Could not reach fal host. "
                         "This is most likely a transient problem. "
-                        "Please, try again."
+                        "If it persists, please reach out to support@fal.ai with the following details: "  # noqa: E501
+                        f"{msg}"
                     )
-                elif e.details().endswith("died with <Signals.SIGKILL: 9>.`."):
+                elif msg.endswith("died with <Signals.SIGKILL: 9>.`."):
                     raise FalServerlessError(
                         "Isolated function crashed. "
                         "This is likely due to resource overflow. "
                         "You can try again by setting a bigger `machine_type`"
                     )
                 elif e.code() == grpc.StatusCode.INVALID_ARGUMENT and (
-                    "The function function could not be deserialized" in e.details()
+                    "The function function could not be deserialized" in msg
                 ):
-                    raise FalMissingDependencyError(e.details()) from None
+                    raise FalMissingDependencyError(msg) from None
                 else:
-                    raise FalServerlessError(e.details())
+                    raise FalServerlessError(msg)
 
         return handler
 
@@ -498,7 +500,7 @@ class FalServerlessHost(Host):
             partial_func,
             environments,
             application_name=application_name,
-            application_auth_mode=application_auth_mode,
+            auth_mode=application_auth_mode,
             machine_requirements=machine_requirements,
             metadata=metadata,
             deployment_strategy=deployment_strategy,

{fal-1.25.0 → fal-1.26.1}/src/fal/app.py

@@ -197,7 +197,14 @@ class AppClient:
 
     @classmethod
     @contextmanager
-    def connect(cls, app_cls):
+    def connect(
+        cls,
+        app_cls,
+        *,
+        health_request_timeout: int = 30,
+        startup_timeout: int = 60,
+        health_check_interval: float = 0.5,
+    ):
         app = wrap_app(app_cls)
         info = app.spawn()
         _shutdown_event = threading.Event()
@@ -214,20 +221,50 @@ class AppClient:
         _log_printer.start()
 
         try:
+            if info.url is None:
+                raise AppClientError(
+                    "App spawn failed: no URL returned",
+                    status_code=500,
+                )
+
+            start_time = time.perf_counter()
+            url = info.url + "/health"
+            last_error = None
+            attempt = 0
+
             with httpx.Client() as client:
-                retries = 100
-                for _ in range(retries):
-                    url = info.url + "/health"
-                    resp = client.get(url, timeout=60)
-
-                    if resp.is_success:
-                        break
-                    elif resp.status_code not in (500, 404):
-                        raise AppClientError(
-                            f"Failed to GET {url}: {resp.status_code} {resp.text}",
-                            status_code=resp.status_code,
+                while time.perf_counter() - start_time < startup_timeout:
+                    attempt += 1
+
+                    try:
+                        resp = client.get(url, timeout=health_request_timeout)
+                    except httpx.TimeoutException:
+                        last_error = (
+                            f"Request timed out after {health_request_timeout} seconds"
                         )
-                    time.sleep(0.1)
+                    except httpx.TransportError as e:
+                        last_error = f"Network error: {e}"
+                    else:
+                        if resp.is_success:
+                            break
+
+                        if resp.status_code in (500, 404):
+                            last_error = f"Server not ready (HTTP {resp.status_code})"
+                        else:
+                            raise AppClientError(
+                                "Health check failed with non-retryable error: "
+                                f"{resp.status_code} {resp.text}",
+                                status_code=resp.status_code,
+                            )
+
+                    time.sleep(health_check_interval)
+                else:
+                    # retry loop completed without success
+                    raise AppClientError(
+                        f"Health check failed after {startup_timeout}s "
+                        f"({attempt} attempts). Last error: {last_error}",
+                        status_code=500,
+                    )
 
             client = cls(app_cls, info.url)
             yield client

{fal-1.25.0 → fal-1.26.1}/src/fal/cli/apps.py

@@ -270,7 +270,7 @@ def _add_set_rev_parser(subparsers, parents):
     parser.add_argument(
         "--auth",
         choices=ALIAS_AUTH_MODES,
-        default="private",
+        default=None,
         help="Application authentication mode.",
     )
     parser.set_defaults(func=_set_rev)
@@ -310,7 +310,11 @@ def _add_runners_parser(subparsers, parents):
 def _delete(args):
     client = get_client(args.host, args.team)
     with client.connect() as connection:
-        connection.delete_alias(args.app_name)
+        res = connection.delete_alias(args.app_name)
+        if res is None:
+            args.console.print(f"Application {args.app_name!r} not found.")
+        else:
+            args.console.print(f"Application {args.app_name!r} deleted ({res})")
 
 
 def _add_delete_parser(subparsers, parents):

{fal-1.25.0 → fal-1.26.1}/src/fal/files.py

@@ -1,5 +1,8 @@
+import concurrent.futures
+import math
 import os
 import posixpath
+from concurrent.futures import ThreadPoolExecutor
 from functools import cached_property
 from typing import TYPE_CHECKING
 
@@ -9,6 +12,19 @@ if TYPE_CHECKING:
     import httpx
 
 USER_AGENT = "fal-sdk/1.14.0 (python)"
+MULTIPART_THRESHOLD = 10 * 1024 * 1024  # 10MB
+MULTIPART_CHUNK_SIZE = 10 * 1024 * 1024  # 10MB
+MULTIPART_WORKERS = 2  # only 2 because our REST is currently struggling with more
+
+
+def _compute_md5(lpath, chunk_size=8192):
+    import hashlib
+
+    hasher = hashlib.md5()
+    with open(lpath, "rb") as fobj:
+        for chunk in iter(lambda: fobj.read(chunk_size), b""):
+            hasher.update(chunk)
+    return hasher.hexdigest()
 
 
 class FalFileSystem(AbstractFileSystem):
@@ -95,16 +111,90 @@ class FalFileSystem(AbstractFileSystem):
             response = self._request("GET", f"/files/file/{rpath}")
             fobj.write(response.content)
 
+    def _put_file_part(self, rpath, lpath, upload_id, part_number, chunk_size):
+        offset = (part_number - 1) * chunk_size
+        with open(lpath, "rb") as fobj:
+            fobj.seek(offset)
+            chunk = fobj.read(chunk_size)
+            response = self._request(
+                "PUT",
+                f"/files/file/multipart/{rpath}/{upload_id}/{part_number}",
+                files={"file_upload": (posixpath.basename(lpath), chunk)},
+            )
+            data = response.json()
+            return {
+                "part_number": data["part_number"],
+                "etag": data["etag"],
+            }
+
+    def _put_file_multipart(self, lpath, rpath, size, progress):
+        response = self._request(
+            "POST",
+            f"/files/file/multipart/{rpath}/initiate",
+        )
+        upload_id = response.json()["upload_id"]
+
+        parts = []
+        num_parts = math.ceil(size / MULTIPART_CHUNK_SIZE)
+        md5 = _compute_md5(lpath)
+
+        task = progress.add_task(f"{os.path.basename(lpath)}", total=num_parts)
+
+        with ThreadPoolExecutor(max_workers=MULTIPART_WORKERS) as executor:
+            futures = []
+
+            for part_number in range(1, num_parts + 1):
+                futures.append(
+                    executor.submit(
+                        self._put_file_part,
+                        rpath,
+                        lpath,
+                        upload_id,
+                        part_number,
+                        MULTIPART_CHUNK_SIZE,
+                    )
+                )
+
+            for future in concurrent.futures.as_completed(futures):
+                parts.append(future.result())
+                progress.advance(task)
+
+        response = self._request(
+            "POST",
+            f"/files/file/multipart/{rpath}/{upload_id}/complete",
+            json={"parts": parts},
+        )
+        data = response.json()
+        if data["etag"] != md5:
+            raise RuntimeError(
+                f"MD5 mismatch on {rpath}: {data['etag']} != {md5}, "
+                "please contact support"
+            )
+
     def put_file(self, lpath, rpath, mode="overwrite", **kwargs):
+        from rich.progress import BarColumn, Progress, SpinnerColumn, TextColumn
+
         if os.path.isdir(lpath):
             return
 
-        with open(lpath, "rb") as fobj:
-            self._request(
-                "POST",
-                f"/files/file/local/{rpath}",
-                files={"file_upload": (posixpath.basename(lpath), fobj, "text/plain")},
-            )
+        size = os.path.getsize(lpath)
+        with Progress(
+            SpinnerColumn(),
+            TextColumn("[progress.description]{task.description}"),
+            BarColumn(),
+            TextColumn("[progress.percentage]{task.percentage:>3.0f}%"),
+        ) as progress:
+            if size > MULTIPART_THRESHOLD:
+                self._put_file_multipart(lpath, rpath, size, progress)
+            else:
+                task = progress.add_task(f"{os.path.basename(lpath)}", total=1)
+                with open(lpath, "rb") as fobj:
+                    self._request(
+                        "POST",
+                        f"/files/file/local/{rpath}",
+                        files={"file_upload": (posixpath.basename(lpath), fobj)},
+                    )
+                progress.advance(task)
         self.dircache.clear()
 
     def put_file_from_url(self, url, rpath, mode="overwrite", **kwargs):

{fal-1.25.0 → fal-1.26.1}/src/fal/sdk.py

@@ -45,6 +45,9 @@ logger = get_logger(__name__)
 patch_pickle()
 
 
+AuthMode = Optional[Literal["public", "private", "shared"]]
+
+
 class ServerCredentials:
     def to_grpc(self) -> grpc.ChannelCredentials:
         raise NotImplementedError
@@ -565,7 +568,7 @@ class FalServerlessConnection:
         function: Callable[..., ResultT],
         environments: list[isolate_proto.EnvironmentDefinition],
         application_name: str | None = None,
-        application_auth_mode: Literal["public", "private", "shared"] | None = None,
+        auth_mode: AuthMode = None,
         *,
         serialization_method: str = _DEFAULT_SERIALIZATION_METHOD,
         machine_requirements: MachineRequirements | None = None,
@@ -598,13 +601,14 @@ class FalServerlessConnection:
         else:
             wrapped_requirements = None
 
-        auth_mode = None
-        if application_auth_mode == "public":
-            auth_mode = isolate_proto.ApplicationAuthMode.PUBLIC
-        elif application_auth_mode == "shared":
-            auth_mode = isolate_proto.ApplicationAuthMode.SHARED
-        elif application_auth_mode == "private":
-            auth_mode = isolate_proto.ApplicationAuthMode.PRIVATE
+        if auth_mode == "public":
+            auth = isolate_proto.ApplicationAuthMode.PUBLIC
+        elif auth_mode == "shared":
+            auth = isolate_proto.ApplicationAuthMode.SHARED
+        elif auth_mode == "private":
+            auth = isolate_proto.ApplicationAuthMode.PRIVATE
+        else:
+            auth = None
 
         struct_metadata = None
         if metadata:
@@ -620,7 +624,7 @@ class FalServerlessConnection:
             environments=environments,
             machine_requirements=wrapped_requirements,
             application_name=application_name,
-            auth_mode=auth_mode,
+            auth_mode=auth,
             metadata=struct_metadata,
             deployment_strategy=deployment_strategy_proto,
             scale=scale,
@@ -730,14 +734,16 @@ class FalServerlessConnection:
         self,
         alias: str,
         revision: str,
-        auth_mode: Literal["public", "private", "shared"],
+        auth_mode: AuthMode,
     ):
         if auth_mode == "public":
             auth = isolate_proto.ApplicationAuthMode.PUBLIC
         elif auth_mode == "shared":
             auth = isolate_proto.ApplicationAuthMode.SHARED
-        else:
+        elif auth_mode == "private":
             auth = isolate_proto.ApplicationAuthMode.PRIVATE
+        else:
+            auth = None
 
         request = isolate_proto.SetAliasRequest(
             alias=alias,
@@ -746,10 +752,15 @@ class FalServerlessConnection:
         )
         self.stub.SetAlias(request)
 
-    def delete_alias(self, alias: str) -> str:
+    def delete_alias(self, alias: str) -> str | None:
         request = isolate_proto.DeleteAliasRequest(alias=alias)
-        res: isolate_proto.DeleteAliasResult = self.stub.DeleteAlias(request)
-        return res.revision
+        try:
+            res: isolate_proto.DeleteAliasResult = self.stub.DeleteAlias(request)
+            return res.revision
+        except grpc.RpcError as e:
+            if e.code() == grpc.StatusCode.NOT_FOUND:
+                return None
+            raise
 
     def list_aliases(self) -> list[AliasInfo]:
         request = isolate_proto.ListAliasesRequest()

{fal-1.25.0 → fal-1.26.1}/tests/test_apps.py

@@ -6,7 +6,7 @@ import time
 import uuid
 from contextlib import contextmanager, suppress
 from datetime import datetime, timedelta, timezone
-from typing import Generator, List, Tuple
+from typing import Generator, List, Tuple, Union
 
 import httpx
 import pytest
@@ -20,7 +20,7 @@ from pydantic import __version__ as pydantic_version
 import fal
 import fal.api as api
 from fal import apps
-from fal.app import AppClient, AppClientError
+from fal.app import AppClient, AppClientError, wrap_app
 from fal.cli.deploy import User, _get_user
 from fal.container import ContainerImage
 from fal.exceptions import (
@@ -235,7 +235,7 @@ class ExceptionApp(fal.App, keep_alive=300, max_concurrency=1):
         raise RuntimeError("cuDNN error: CUDNN_STATUS_INTERNAL_ERROR")
 
 
-class CancellableApp(fal.App, keep_alive=300, max_concurrency=1, request_timeout=10):
+class CancellableApp(fal.App, keep_alive=300, max_concurrency=1, request_timeout=2):
     task = None
     running = 0
 
@@ -332,7 +332,7 @@ class BrokenApp(fal.App, keep_alive=300, max_concurrency=1):
 
 
 @pytest.fixture(scope="module")
-def host() -> Generator[api.FalServerlessHost, None, None]:
+def host() -> Generator[api.Host, None, None]:
     yield addition_app.host
 
 
@@ -343,7 +343,11 @@ def user() -> Generator[User, None, None]:
 
 
 @contextmanager
-def register_app(host: api.FalServerlessHost, app: fal.App, suffix: str = ""):
+def register_app(
+    host: api.FalServerlessHost,
+    app: Union[api.ServedIsolatedFunction, api.IsolatedFunction],
+    suffix: str = "",
+):
     app_alias = str(uuid.uuid4()) + "-test-alias" + ("-" + suffix if suffix else "")
     app_revision = host.register(
         func=app.func,
@@ -398,7 +402,7 @@ def test_fastapi_app(host: api.FalServerlessHost, user: User):
 
 @pytest.fixture(scope="module")
 def test_stateful_app(host: api.FalServerlessHost, user: User):
-    stateful_app = fal.wrap_app(StatefulAdditionApp)
+    stateful_app = wrap_app(StatefulAdditionApp)
     with register_app(host, stateful_app, "stateful") as (app_alias, _):
         yield f"{user.username}/{app_alias}"
 
@@ -411,7 +415,7 @@ def test_pydantic_validation_error():
 
 @pytest.fixture(scope="module")
 def test_cancellable_app(host: api.FalServerlessHost, user: User):
-    cancellable_app = fal.wrap_app(CancellableApp)
+    cancellable_app = wrap_app(CancellableApp)
     with register_app(host, cancellable_app, "cancellable") as (app_alias, _):
         yield f"{user.username}/{app_alias}"
 
@@ -424,21 +428,21 @@ def test_exception_app():
 
 @pytest.fixture(scope="module")
 def test_sleep_app(host: api.FalServerlessHost, user: User):
-    sleep_app = fal.wrap_app(SleepApp)
+    sleep_app = wrap_app(SleepApp)
     with register_app(host, sleep_app, "sleep") as (app_alias, _):
         yield f"{user.username}/{app_alias}"
 
 
 @pytest.fixture(scope="module")
 def test_realtime_app(host: api.FalServerlessHost, user: User):
-    realtime_app = fal.wrap_app(RealtimeApp)
+    realtime_app = wrap_app(RealtimeApp)
     with register_app(host, realtime_app, "realtime") as (app_alias, _):
         yield f"{user.username}/{app_alias}"
 
 
 def test_broken_app_failure(host: api.FalServerlessHost, user: User):
     with pytest.raises(FalServerlessException) as e:
-        fal.wrap_app(BrokenApp)
+        wrap_app(BrokenApp)
 
     assert "Failed to generate OpenAPI" in str(e)
 
@@ -551,7 +555,7 @@ def test_app_disconnect_behavior(test_app: str, test_cancellable_app: str):
     with pytest.raises(HTTPStatusError) as e:
         apps.run(
             test_cancellable_app,
-            arguments={"lhs": 1, "rhs": 2, "wait_time": 20},
+            arguments={"lhs": 1, "rhs": 2, "wait_time": 6},
             path="/well-handled",
         )
     assert (
@@ -571,19 +575,12 @@ def test_app_disconnect_behavior(test_app: str, test_cancellable_app: str):
     with pytest.raises(HTTPStatusError) as e:
         apps.run(
             test_cancellable_app,
-            arguments={"lhs": 1, "rhs": 2, "wait_time": 20},
+            arguments={"lhs": 1, "rhs": 2, "wait_time": 6},
         )
     assert (
         e.value.response.status_code == 504
     ), "Expected Gateway Timeout even though the app handled it"
 
-    with pytest.raises(HTTPStatusError) as e:
-        apps.run(
-            test_cancellable_app,
-            arguments={"lhs": 1, "rhs": 2, "wait_time": 1},
-        )
-    assert e.value.response.status_code == 500
-
 
 @pytest.mark.xfail(
     reason="Temporary disabled while investigating backend issue. Ping @efiop"