[Fix-Router] Don't cooldown when only 1 deployment exists (#5673)

* fix get model list

* fix test custom callback router

* fix embedding fallback test

* fix router retry policy on AuthErrors

* fix router test

* add test for single deployments no cooldown test prod

* add test test_single_deployment_no_cooldowns_test_prod_mock_completion_calls

litellm/proxy/proxy_config.yaml

@@ -6,11 +6,11 @@ model_list:
       vertex_project: "adroit-crow-413218"
       vertex_location: "us-central1"
       vertex_credentials: "/Users/ishaanjaffer/Downloads/adroit-crow-413218-a956eef1a2a8.json"
-  - model_name: fake-openai-endpoint
+  - model_name: fake-azure-endpoint
     litellm_params:
-      model: openai/fake
+      model: openai/429
       api_key: fake-key
-      api_base: https://exampleopenaiendpoint-production.up.railway.app/
+      api_base: https://exampleopenaiendpoint-production.up.railway.app
 
 general_settings: 
  master_key: sk-1234 

litellm/router.py

@@ -1130,7 +1130,7 @@ class Router:
         make_request = False
 
         while curr_time < end_time:
-            _healthy_deployments = await self._async_get_healthy_deployments(
+            _healthy_deployments, _ = await self._async_get_healthy_deployments(
                 model=model
             )
             make_request = await self.scheduler.poll(  ## POLL QUEUE ## - returns 'True' if there's healthy deployments OR if request is at top of queue
@@ -3060,14 +3060,17 @@ class Router:
             Retry Logic
              
             """
-            _healthy_deployments = await self._async_get_healthy_deployments(
+            _healthy_deployments, _all_deployments = (
+                await self._async_get_healthy_deployments(
                     model=kwargs.get("model") or "",
                 )
+            )
 
             # raises an exception if this error should not be retries
             self.should_retry_this_error(
                 error=e,
                 healthy_deployments=_healthy_deployments,
+                all_deployments=_all_deployments,
                 context_window_fallbacks=context_window_fallbacks,
                 regular_fallbacks=fallbacks,
                 content_policy_fallbacks=content_policy_fallbacks,
@@ -3114,7 +3117,7 @@ class Router:
                     ## LOGGING
                     kwargs = self.log_retry(kwargs=kwargs, e=e)
                     remaining_retries = num_retries - current_attempt
-                    _healthy_deployments = await self._async_get_healthy_deployments(
+                    _healthy_deployments, _ = await self._async_get_healthy_deployments(
                         model=kwargs.get("model"),
                     )
                     _timeout = self._time_to_sleep_before_retry(
@@ -3135,6 +3138,7 @@ class Router:
         self,
         error: Exception,
         healthy_deployments: Optional[List] = None,
+        all_deployments: Optional[List] = None,
         context_window_fallbacks: Optional[List] = None,
         content_policy_fallbacks: Optional[List] = None,
         regular_fallbacks: Optional[List] = None,
@@ -3150,6 +3154,9 @@ class Router:
         _num_healthy_deployments = 0
         if healthy_deployments is not None and isinstance(healthy_deployments, list):
             _num_healthy_deployments = len(healthy_deployments)
+        _num_all_deployments = 0
+        if all_deployments is not None and isinstance(all_deployments, list):
+            _num_all_deployments = len(all_deployments)
 
         ### CHECK IF RATE LIMIT / CONTEXT WINDOW ERROR / CONTENT POLICY VIOLATION ERROR w/ fallbacks available / Bad Request Error
         if (
@@ -3180,7 +3187,9 @@ class Router:
             - if other deployments available -> retry
             - else -> raise error
             """
-            if _num_healthy_deployments <= 0:  # if no healthy deployments
+            if (
+                _num_all_deployments <= 1
+            ):  # if there is only 1 deployment for this model group then don't retry
                 raise error  # then raise error
 
         # Do not retry if there are no healthy deployments
@@ -3390,7 +3399,7 @@ class Router:
             current_attempt = None
             original_exception = e
             ### CHECK IF RATE LIMIT / CONTEXT WINDOW ERROR
-            _healthy_deployments = self._get_healthy_deployments(
+            _healthy_deployments, _all_deployments = self._get_healthy_deployments(
                 model=kwargs.get("model"),
             )
 
@@ -3398,6 +3407,7 @@ class Router:
             self.should_retry_this_error(
                 error=e,
                 healthy_deployments=_healthy_deployments,
+                all_deployments=_all_deployments,
                 context_window_fallbacks=context_window_fallbacks,
                 regular_fallbacks=fallbacks,
                 content_policy_fallbacks=content_policy_fallbacks,
@@ -3428,7 +3438,7 @@ class Router:
                 except Exception as e:
                     ## LOGGING
                     kwargs = self.log_retry(kwargs=kwargs, e=e)
-                    _healthy_deployments = self._get_healthy_deployments(
+                    _healthy_deployments, _ = self._get_healthy_deployments(
                         model=kwargs.get("model"),
                     )
                     remaining_retries = num_retries - current_attempt
@@ -3881,7 +3891,7 @@ class Router:
             else:
                 healthy_deployments.append(deployment)
 
-        return healthy_deployments
+        return healthy_deployments, _all_deployments
 
     async def _async_get_healthy_deployments(self, model: str):
         _all_deployments: list = []
@@ -3901,7 +3911,7 @@ class Router:
                 continue
             else:
                 healthy_deployments.append(deployment)
-        return healthy_deployments
+        return healthy_deployments, _all_deployments
 
     def routing_strategy_pre_call_checks(self, deployment: dict):
         """
@@ -4679,10 +4689,7 @@ class Router:
                 returned_models += self.model_list
 
                 return returned_models
-
-            for model in self.model_list:
             returned_models.extend(self._get_all_deployments(model_name=model_name))
-
             return returned_models
         return None
 

litellm/tests/test_custom_callback_router.py

@@ -533,6 +533,7 @@ async def test_async_chat_azure_with_fallbacks():
     try:
         customHandler_fallbacks = CompletionCustomHandler()
         litellm.callbacks = [customHandler_fallbacks]
+        litellm.set_verbose = True
         # with fallbacks
         model_list = [
             {
@@ -555,7 +556,13 @@ async def test_async_chat_azure_with_fallbacks():
                 "rpm": 1800,
             },
         ]
-        router = Router(model_list=model_list, fallbacks=[{"gpt-3.5-turbo": ["gpt-3.5-turbo-16k"]}])  # type: ignore
+        router = Router(
+            model_list=model_list,
+            fallbacks=[{"gpt-3.5-turbo": ["gpt-3.5-turbo-16k"]}],
+            retry_policy=litellm.router.RetryPolicy(
+                AuthenticationErrorRetries=0,
+            ),
+        )  # type: ignore
         response = await router.acompletion(
             model="gpt-3.5-turbo",
             messages=[{"role": "user", "content": "Hi 👋 - i'm openai"}],

litellm/tests/test_router_cooldowns.py

@@ -150,3 +150,100 @@ def test_single_deployment_no_cooldowns(num_deployments):
             mock_client.assert_not_called()
         else:
             mock_client.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_single_deployment_no_cooldowns_test_prod():
+    """
+    Do not cooldown on single deployment.
+
+    """
+    router = Router(
+        model_list=[
+            {
+                "model_name": "gpt-3.5-turbo",
+                "litellm_params": {
+                    "model": "gpt-3.5-turbo",
+                },
+            },
+            {
+                "model_name": "gpt-5",
+                "litellm_params": {
+                    "model": "openai/gpt-5",
+                },
+            },
+            {
+                "model_name": "gpt-12",
+                "litellm_params": {
+                    "model": "openai/gpt-12",
+                },
+            },
+        ],
+        allowed_fails=0,
+        num_retries=0,
+    )
+
+    with patch.object(
+        router.cooldown_cache, "add_deployment_to_cooldown", new=MagicMock()
+    ) as mock_client:
+        try:
+            await router.acompletion(
+                model="gpt-3.5-turbo",
+                messages=[{"role": "user", "content": "Hey, how's it going?"}],
+                mock_response="litellm.RateLimitError",
+            )
+        except litellm.RateLimitError:
+            pass
+
+        await asyncio.sleep(2)
+
+        mock_client.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_single_deployment_no_cooldowns_test_prod_mock_completion_calls():
+    """
+    Do not cooldown on single deployment.
+
+    """
+    router = Router(
+        model_list=[
+            {
+                "model_name": "gpt-3.5-turbo",
+                "litellm_params": {
+                    "model": "gpt-3.5-turbo",
+                },
+            },
+            {
+                "model_name": "gpt-5",
+                "litellm_params": {
+                    "model": "openai/gpt-5",
+                },
+            },
+            {
+                "model_name": "gpt-12",
+                "litellm_params": {
+                    "model": "openai/gpt-12",
+                },
+            },
+        ],
+    )
+
+    for _ in range(20):
+        try:
+            await router.acompletion(
+                model="gpt-3.5-turbo",
+                messages=[{"role": "user", "content": "Hey, how's it going?"}],
+                mock_response="litellm.RateLimitError",
+            )
+        except litellm.RateLimitError:
+            pass
+
+    cooldown_list = await router._async_get_cooldown_deployments()
+    assert len(cooldown_list) == 0
+
+    healthy_deployments, _ = await router._async_get_healthy_deployments(
+        model="gpt-3.5-turbo"
+    )
+
+    print("healthy_deployments: ", healthy_deployments)