[Spaces] ZeroGPU Queue fix (#5129)

* Properly set max_size and concurrency_count for ZeroGPU

* ZeroGPU concurrency_count warning

* Test

* add changeset

* Lint

* Sure, typing_extensions is a third-party ..

* Fix and refactor

* Black

* Empty commit

* add changeset

---------

Co-authored-by: gradio-pr-bot <gradio-pr-bot@users.noreply.github.com>

.changeset/true-camels-shave.md

@@ -0,0 +1,5 @@
+---
+"gradio": minor
+---
+
+fix:[Spaces] ZeroGPU Queue fix

gradio/blocks.py

@@ -58,6 +58,7 @@
     TupleNoPrint,
     check_function_inputs_match,
     component_or_layout_class,
+    concurrency_count_warning,
     delete_none,
     get_cancel_function,
     get_continuous_fn,
@@ -1583,6 +1584,7 @@ def clear(self):
         self.children = []
         return self
 
+    @concurrency_count_warning
     @document()
     def queue(
         self,
@@ -1624,12 +1626,14 @@ def queue(
             warn_deprecation(
                 "The client_position_to_load_data parameter is deprecated."
             )
-        max_size_default = self.max_threads if utils.is_zero_gpu_space() else None
+        if utils.is_zero_gpu_space():
+            concurrency_count = self.max_threads
+            max_size = 1 if max_size is None else max_size
         self._queue = queueing.Queue(
             live_updates=status_update_rate == "auto",
             concurrency_count=concurrency_count,
             update_intervals=status_update_rate if status_update_rate != "auto" else 1,
-            max_size=max_size_default if max_size is None else max_size,
+            max_size=max_size,
             blocks_dependencies=self.dependencies,
         )
         self.config = self.get_config_file()

gradio/utils.py

@@ -39,6 +39,7 @@
 from mdit_py_plugins.dollarmath.index import dollarmath_plugin
 from mdit_py_plugins.footnote.index import footnote_plugin
 from pydantic import BaseModel, parse_obj_as
+from typing_extensions import ParamSpec
 
 import gradio
 from gradio.context import Context
@@ -53,6 +54,7 @@
     (pkgutil.get_data(__name__, "version.txt") or b"").decode("ascii").strip()
 )
 
+P = ParamSpec("P")
 T = TypeVar("T")
 
 
@@ -843,6 +845,23 @@ def check_function_inputs_match(fn: Callable, inputs: list, inputs_as_dict: bool
         )
 
 
+def concurrency_count_warning(queue: Callable[P, T]) -> Callable[P, T]:
+    @functools.wraps(queue)
+    def wrapper(*args: P.args, **kwargs: P.kwargs) -> T:
+        _self, *positional = args
+        if is_zero_gpu_space() and (
+            len(positional) >= 1 or "concurrency_count" in kwargs
+        ):
+            warnings.warn(
+                "Queue concurrency_count on ZeroGPU Spaces cannot be overriden "
+                "and is always equal to Block's max_threads. "
+                "Consider setting max_threads value on the Block instead"
+            )
+        return queue(*args, **kwargs)
+
+    return wrapper
+
+
 class TupleNoPrint(tuple):
     # To remove printing function return in notebook
     def __repr__(self):

test/test_blocks.py

@@ -434,6 +434,16 @@ def test_raise_error_if_event_queued_but_queue_not_enabled(self):
 
         demo.close()
 
+    def test_concurrency_count_zero_gpu(self):
+        os.environ["SPACES_ZERO_GPU"] = "true"
+        demo = gr.Blocks()
+        with pytest.warns():
+            demo.queue(concurrency_count=42)
+        with pytest.warns():
+            demo.queue(42)
+        assert demo._queue.max_thread_count == demo.max_threads
+        del os.environ["SPACES_ZERO_GPU"]
+
 
 class TestTempFile:
     def test_pil_images_hashed(self, connect, gradio_temp_dir):