[Fix] Router - cooldown time, allow using dynamic cooldown time for a specific deployment (#12037)

* fixes header_cooldown

* test_deployment_callback_respects_cooldown_time

litellm/router.py

@@ -4108,20 +4108,26 @@ class Router:
                 original_exception=exception
             )
 
-            _time_to_cooldown = kwargs.get("litellm_params", {}).get(
-                "cooldown_time", self.cooldown_time
-            )
-
+            # Determine cooldown time with priority: deployment config > response header > router default
+            deployment_cooldown = kwargs.get("litellm_params", {}).get("cooldown_time", None)
+            
+            header_cooldown = None
             if exception_headers is not None:
-                _time_to_cooldown = (
-                    litellm.utils._get_retry_after_from_exception_header(
-                        response_headers=exception_headers
-                    )
+                header_cooldown = litellm.utils._get_retry_after_from_exception_header(
+                    response_headers=exception_headers
                 )
-
-                if _time_to_cooldown is None or _time_to_cooldown < 0:
-                    # if the response headers did not read it -> set to default cooldown time
-                    _time_to_cooldown = self.cooldown_time
+            ##############################################
+            # Logic to determine cooldown time
+            # 1. Check if a cooldown time is set in the deployment config
+            # 2. Check if a cooldown time is set in the response header
+            # 3. If no cooldown time is set, use the router default cooldown time
+            ##############################################
+            if deployment_cooldown is not None and deployment_cooldown >= 0:
+                _time_to_cooldown = deployment_cooldown
+            elif header_cooldown is not None and header_cooldown >= 0:
+                _time_to_cooldown = header_cooldown
+            else:
+                _time_to_cooldown = self.cooldown_time
 
             if isinstance(_model_info, dict):
                 deployment_id = _model_info.get("id", None)

tests/router_unit_tests/test_router_helper_utils.py

@@ -446,6 +446,40 @@ async def test_deployment_callback_on_failure(model_list):
     )
 
 
+def test_deployment_callback_respects_cooldown_time(model_list):
+    """Ensure per-model cooldown_time is honored even when exception headers are present."""
+    import httpx
+    import time
+    from unittest.mock import patch
+
+    router = Router(model_list=model_list)
+
+    class FakeException(Exception):
+        def __init__(self):
+            self.status_code = 429
+            self.headers = httpx.Headers({"x-test": "1"})
+
+    kwargs = {
+        "exception": FakeException(),
+        "litellm_params": {
+            "metadata": {"model_group": "gpt-3.5-turbo"},
+            "model_info": {"id": 100},
+            "cooldown_time": 0,
+        },
+    }
+
+    with patch("litellm.router._set_cooldown_deployments") as mock_set:
+        router.deployment_callback_on_failure(
+            kwargs=kwargs,
+            completion_response=None,
+            start_time=time.time(),
+            end_time=time.time(),
+        )
+
+        mock_set.assert_called_once()
+        assert mock_set.call_args.kwargs["time_to_cooldown"] == 0
+
+
 def test_log_retry(model_list):
     """Test if the '_log_retry' function is working correctly"""
     import time