[Fix] handle metadata=None in SDK path retry/error logic (utils.py) (#20873)

* [Fix] handle metadata=None in SDK path retry/error logic (utils.py)

Fixes #20871

Same class of bug as #9717 (fixed by #9764 for the proxy path).
The SDK path in utils.py has the same fragile pattern at 7 locations.

Replace `kwargs.get("metadata", {})` with `(kwargs.get("metadata") or {})`
to handle the case where metadata key exists with value None (e.g. from
Azure OpenAI streaming responses).

This is consistent with the existing correct pattern at line 602:
`metadata = kwargs.get("metadata") or {}`

Adds TestMetadataNoneHandling with 6 unit tests in test_utils.py.

* fix: remove duplicate PerplexityResponsesConfig key in lazy imports registry

Removes duplicate dictionary key added in commit be0ebb15 (PR #20860).
The entry at line 1042 is identical to the existing entry at line 906.
This causes ruff F601 lint failure on all PRs targeting main.

litellm/utils.py

@@ -1406,7 +1406,7 @@ def client(original_function):  # noqa: PLR0915
             # [OPTIONAL] CHECK MAX RETRIES / REQUEST
             if litellm.num_retries_per_request is not None:
                 # check if previous_models passed in as ['litellm_params']['metadata]['previous_models']
-                previous_models = kwargs.get("metadata", {}).get(
+                previous_models = (kwargs.get("metadata") or {}).get(
                     "previous_models", None
                 )
                 if previous_models is not None:
@@ -1483,7 +1483,7 @@ def client(original_function):  # noqa: PLR0915
             # [OPTIONAL] CHECK MAX RETRIES / REQUEST
             if litellm.num_retries_per_request is not None:
                 # check if previous_models passed in as ['litellm_params']['metadata]['previous_models']
-                previous_models = kwargs.get("metadata", {}).get(
+                previous_models = (kwargs.get("metadata") or {}).get(
                     "previous_models", None
                 )
                 if previous_models is not None:
@@ -1678,8 +1678,8 @@ def client(original_function):  # noqa: PLR0915
                     "context_window_fallback_dict", {}
                 )
 
-                _is_litellm_router_call = "model_group" in kwargs.get(
-                    "metadata", {}
+                _is_litellm_router_call = "model_group" in (
+                    kwargs.get("metadata") or {}
                 )  # check if call from litellm.router/proxy
                 if (
                     num_retries and not _is_litellm_router_call
@@ -1724,8 +1724,8 @@ def client(original_function):  # noqa: PLR0915
                     None  # set retries to None to prevent infinite loops
                 )
 
-                _is_litellm_router_call = "model_group" in kwargs.get(
-                    "metadata", {}
+                _is_litellm_router_call = "model_group" in (
+                    kwargs.get("metadata") or {}
                 )  # check if call from litellm.router/proxy
                 if (
                     num_retries and not _is_litellm_router_call
@@ -1974,8 +1974,8 @@ def client(original_function):  # noqa: PLR0915
                     "context_window_fallback_dict", {}
                 )
 
-                _is_litellm_router_call = "model_group" in kwargs.get(
-                    "metadata", {}
+                _is_litellm_router_call = "model_group" in (
+                    kwargs.get("metadata") or {}
                 )  # check if call from litellm.router/proxy
 
                 if (
@@ -2008,8 +2008,8 @@ def client(original_function):  # noqa: PLR0915
                         kwargs["model"] = context_window_fallback_dict[model]
                     return await original_function(*args, **kwargs)
             elif call_type == CallTypes.aresponses.value:
-                _is_litellm_router_call = "model_group" in kwargs.get(
-                    "metadata", {}
+                _is_litellm_router_call = "model_group" in (
+                    kwargs.get("metadata") or {}
                 )  # check if call from litellm.router/proxy
 
                 if (
@@ -7320,7 +7320,7 @@ def _get_base_model_from_metadata(model_call_details=None):
         _base_model = litellm_params.get("base_model", None)
         if _base_model is not None:
             return _base_model
-        metadata = litellm_params.get("metadata", {})
+        metadata = litellm_params.get("metadata") or {}
 
         _get_base_model_from_litellm_call_metadata = getattr(
             sys.modules[__name__], "_get_base_model_from_litellm_call_metadata"

tests/test_litellm/test_utils.py

@@ -3305,3 +3305,73 @@ class TestIsStreamingRequest:
 
     def test_stream_true_overrides_non_streaming_call_type(self):
         assert _is_streaming_request(kwargs={"stream": True}, call_type=CallTypes.acompletion) is True
+
+
+class TestMetadataNoneHandling:
+    """
+    Test that metadata=None in kwargs doesn't cause TypeError.
+
+    When metadata key exists with value None (e.g., from Azure OpenAI streaming),
+    dict.get("metadata", {}) returns None (key exists, so default is ignored).
+    The fix uses (kwargs.get("metadata") or {}) which handles both missing key
+    and explicit None value.
+
+    Related: #20871
+    """
+
+    def test_metadata_none_get_previous_models(self):
+        """kwargs.get("metadata") or {} should return {} when metadata is None."""
+        kwargs = {"metadata": None}
+        previous_models = (kwargs.get("metadata") or {}).get(
+            "previous_models", None
+        )
+        assert previous_models is None
+
+    def test_metadata_none_model_group_check(self):
+        """'model_group' in (kwargs.get("metadata") or {}) should not raise TypeError."""
+        kwargs = {"metadata": None}
+        _is_litellm_router_call = "model_group" in (
+            kwargs.get("metadata") or {}
+        )
+        assert _is_litellm_router_call is False
+
+    def test_metadata_missing_key(self):
+        """Should work when metadata key is completely absent."""
+        kwargs = {}
+        previous_models = (kwargs.get("metadata") or {}).get(
+            "previous_models", None
+        )
+        assert previous_models is None
+
+    def test_metadata_present_with_values(self):
+        """Should work when metadata has actual values."""
+        kwargs = {"metadata": {"previous_models": ["model1"], "model_group": "test"}}
+        previous_models = (kwargs.get("metadata") or {}).get(
+            "previous_models", None
+        )
+        assert previous_models == ["model1"]
+        _is_litellm_router_call = "model_group" in (
+            kwargs.get("metadata") or {}
+        )
+        assert _is_litellm_router_call is True
+
+    def test_metadata_none_causes_error_with_old_pattern(self):
+        """Demonstrate the bug: dict.get('metadata', {}) returns None when key exists with None value."""
+        kwargs = {"metadata": None}
+        # Old pattern: kwargs.get("metadata", {}) returns None because key exists
+        result = kwargs.get("metadata", {})
+        assert result is None  # This is the root cause of the bug
+
+        # Attempting to use .get() on None raises AttributeError or TypeError
+        with pytest.raises((TypeError, AttributeError)):
+            kwargs.get("metadata", {}).get("previous_models", None)
+
+        # Attempting 'in' on None raises TypeError
+        with pytest.raises(TypeError):
+            "model_group" in kwargs.get("metadata", {})
+
+    def test_litellm_params_metadata_none(self):
+        """litellm_params.get("metadata") or {} should handle None value."""
+        litellm_params = {"metadata": None}
+        metadata = litellm_params.get("metadata") or {}
+        assert metadata == {}