feat(fal): don't scale by default on deploy

fal-ai · Oct 18, 2024 · e2e50dc · e2e50dc
1 parent 0777cd5
commit e2e50dc
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 3 deletions.
diff --git a/projects/fal/src/fal/api.py b/projects/fal/src/fal/api.py
@@ -430,6 +430,7 @@ def register(
         application_auth_mode: Literal["public", "shared", "private"] | None = None,
         metadata: dict[str, Any] | None = None,
         deployment_strategy: Literal["recreate", "rolling"] = "recreate",
+        scale: bool = False,
     ) -> str | None:
         environment_options = options.environment.copy()
         environment_options.setdefault("python_version", active_python())
@@ -439,15 +440,21 @@ def register(
             "machine_type", FAL_SERVERLESS_DEFAULT_MACHINE_TYPE
         )
         keep_alive = options.host.get("keep_alive", FAL_SERVERLESS_DEFAULT_KEEP_ALIVE)
-        max_concurrency = options.host.get("max_concurrency")
-        min_concurrency = options.host.get("min_concurrency")
-        max_multiplexing = options.host.get("max_multiplexing")
         base_image = options.host.get("_base_image", None)
         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")
 
+        if scale:
+            max_concurrency = options.host.get("max_concurrency")
+            min_concurrency = options.host.get("min_concurrency")
+            max_multiplexing = options.host.get("max_multiplexing")
+        else:
+            max_concurrency = None
+            min_concurrency = None
+            max_multiplexing = None
+
         machine_requirements = MachineRequirements(
             machine_types=machine_type,  # type: ignore
             num_gpus=options.host.get("num_gpus"),

diff --git a/projects/fal/src/fal/cli/deploy.py b/projects/fal/src/fal/cli/deploy.py
@@ -106,6 +106,7 @@ def _deploy_from_reference(
         application_auth_mode=app_auth,
         metadata=isolated_function.options.host.get("metadata", {}),
         deployment_strategy=deployment_strategy,
+        scale=args.scale,
     )
 
     if app_id:
@@ -219,5 +220,10 @@ def valid_auth_option(option):
         help="Deployment strategy.",
         default="recreate",
     )
+    parser.add_argument(
+        "--scale",
+        action="store_true",
+        help="Scale the application.",
+    )
 
     parser.set_defaults(func=_deploy)