Merge pull request #42118 from BerriAI/litellm_migrate_tests_p3

test(unit): migrate wave 1 phase 3 anthropic, apiserpent, azure and azure_ai legacy tests

tests/unit/conftest.py

@@ -1,4 +1,5 @@
 import os
+from collections.abc import Iterator
 from typing import Final
 
 import pytest
@@ -6,7 +7,19 @@ from pytest_socket import enable_socket, socket_allow_hosts
 
 os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
 
+import litellm  # noqa: E402  # litellm reads LITELLM_LOCAL_MODEL_COST_MAP at import
+import litellm.router as litellm_router_module  # noqa: E402  # same import-time dependency
+import litellm.utils as litellm_utils_module  # noqa: E402  # same import-time dependency
+
 LOOPBACK_HOSTS: Final = ["127.0.0.1", "::1"]
+AMBIENT_AZURE_CREDENTIAL_ENV_VARS: Final = (
+    "AZURE_AD_TOKEN",
+    "AZURE_TENANT_ID",
+    "AZURE_CLIENT_ID",
+    "AZURE_CLIENT_SECRET",
+    "AZURE_USERNAME",
+    "AZURE_PASSWORD",
+)
 
 
 def _allow_loopback_only() -> None:
@@ -21,5 +34,38 @@ def pytest_runtest_setup() -> None:
     _allow_loopback_only()
 
 
+@pytest.fixture(autouse=True)
+def isolate_router_model_cost_state() -> Iterator[None]:
+    original_live_routers: Final = frozenset(litellm_router_module._live_routers)
+    original_runtime_registered_model_cost: Final = {
+        model_key: dict(model_value)
+        for model_key, model_value in litellm_utils_module._runtime_registered_model_cost.items()
+    }
+    yield
+    for router in tuple(litellm_router_module._live_routers):
+        litellm_router_module._live_routers.discard(router)
+    for router in original_live_routers:
+        litellm_router_module._live_routers.add(router)
+    litellm_utils_module._runtime_registered_model_cost.clear()
+    litellm_utils_module._runtime_registered_model_cost.update(original_runtime_registered_model_cost)
+    litellm_utils_module._invalidate_model_cost_lowercase_map()
+    litellm.get_model_info.cache_clear()
+
+
+@pytest.fixture
+def local_model_cost_map(monkeypatch: pytest.MonkeyPatch) -> Iterator[None]:
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url=""))
+    litellm.get_model_info.cache_clear()
+    yield
+    litellm.get_model_info.cache_clear()
+
+
+@pytest.fixture
+def no_ambient_azure_credentials(monkeypatch: pytest.MonkeyPatch) -> None:
+    for name in AMBIENT_AZURE_CREDENTIAL_ENV_VARS:
+        monkeypatch.delenv(name, raising=False)
+
+
 def pytest_sessionfinish() -> None:
     enable_socket()

tests/unit/llms/azure/image_generation/test_azure_image_generation_init.py

@@ -133,88 +133,6 @@ def test_azure_image_generation_flattens_extra_body():
     assert data["size"] == "1024x1024"
 
 
-def test_azure_image_generation_creates_token_provider_from_credentials():
-    """
-    Test that azure_ad_token_provider is created from tenant_id, client_id, client_secret.
-
-    This test verifies the fix in images/main.py where we now create the
-    azure_ad_token_provider from credentials in litellm_params if it's not already provided.
-    """
-    # Simulate the fix in images/main.py
-    litellm_params_dict = {
-        "tenant_id": "test-tenant-id",
-        "client_id": "test-client-id",
-        "client_secret": "test-client-secret",
-        "azure_scope": None,
-    }
-
-    azure_ad_token_provider = None
-
-    # This is the logic we added in images/main.py
-    if azure_ad_token_provider is None:
-        tenant_id = litellm_params_dict.get("tenant_id")
-        client_id = litellm_params_dict.get("client_id")
-        client_secret = litellm_params_dict.get("client_secret")
-        azure_scope = (
-            litellm_params_dict.get("azure_scope")
-            or "https://cognitiveservices.azure.com/.default"
-        )
-
-        # Verify the credentials are extracted correctly
-        assert tenant_id == "test-tenant-id"
-        assert client_id == "test-client-id"
-        assert client_secret == "test-client-secret"
-        assert azure_scope == "https://cognitiveservices.azure.com/.default"
-
-        # Verify the condition to create token provider is met
-        assert (
-            tenant_id and client_id and client_secret
-        ), "Credentials should be present to create token provider"
-
-
-def test_azure_image_generation_headers_without_api_key():
-    """
-    Test that when api_key is None, the api-key header is not added to headers.
-
-    This prevents the httpx TypeError: "Header value must be str or bytes, not <class 'NoneType'>"
-    that was occurring when api_key was None and being set in headers.
-
-    This is a unit test for the fix in images/main.py where we now check:
-    if api_key is not None:
-        default_headers["api-key"] = api_key
-    """
-    from litellm.images.main import image_generation
-
-    # Test the header building logic directly
-    api_key = None
-
-    default_headers = {
-        "Content-Type": "application/json",
-    }
-
-    # This is the fix: only add api-key if it's not None
-    if api_key is not None:
-        default_headers["api-key"] = api_key
-
-    # Verify api-key is not in headers when api_key is None
-    assert "api-key" not in default_headers
-
-    # Verify Content-Type is still there
-    assert default_headers["Content-Type"] == "application/json"
-
-    # Test with a valid api_key
-    api_key = "valid-key-123"
-    default_headers_with_key = {
-        "Content-Type": "application/json",
-    }
-    if api_key is not None:
-        default_headers_with_key["api-key"] = api_key
-
-    # Verify api-key is added when api_key is valid
-    assert "api-key" in default_headers_with_key
-    assert default_headers_with_key["api-key"] == "valid-key-123"
-
-
 def test_azure_image_generation_drop_params_response_format():
     """
     Test that unsupported params like response_format are dropped when drop_params=True.

tests/unit/llms/azure/realtime/test_azure_realtime_handler.py

@@ -426,41 +426,6 @@ async def test_async_realtime_beta_without_api_version_raises():
         )
 
 
-@pytest.mark.asyncio
-async def test_realtime_protocol_env_var_fallback():
-    """
-    Test that LITELLM_AZURE_REALTIME_PROTOCOL env var is used as fallback.
-    Fixes #22127: no way to set realtime_protocol from config.
-    """
-    from litellm.realtime_api.main import _arealtime
-    from litellm.types.router import GenericLiteLLMParams
-
-    with patch.dict(os.environ, {"LITELLM_AZURE_REALTIME_PROTOCOL": "v1"}):
-        # Create a GenericLiteLLMParams without realtime_protocol
-        litellm_params = GenericLiteLLMParams()
-        # The env var should be picked up as fallback
-        realtime_protocol = (
-            {}.get("realtime_protocol")
-            or litellm_params.get("realtime_protocol")
-            or os.environ.get("LITELLM_AZURE_REALTIME_PROTOCOL")
-            or "beta"
-        )
-        assert realtime_protocol == "v1"
-
-
-@pytest.mark.asyncio
-async def test_realtime_protocol_from_litellm_params():
-    """
-    Test that realtime_protocol is read from litellm_params (config.yaml extra field).
-    Fixes #22127: realtime_protocol in litellm_params was not used.
-    """
-    from litellm.types.router import GenericLiteLLMParams
-
-    # Simulate config.yaml with realtime_protocol as an extra field
-    litellm_params = GenericLiteLLMParams(realtime_protocol="GA")
-    assert litellm_params.get("realtime_protocol") == "GA"
-
-
 @pytest.mark.asyncio
 async def test_arealtime_transcription_intent_defaults_to_ga(monkeypatch):
     """
@@ -742,7 +707,7 @@ async def test_realtime_health_check_uses_bearer_token_when_no_api_key(monkeypat
 
 
 @pytest.mark.asyncio
-async def test_arealtime_forwards_deployment_azure_ad_token(monkeypatch):
+async def test_arealtime_forwards_deployment_azure_ad_token(monkeypatch, no_ambient_azure_credentials):
     """
     The router binds a deployment's `azure_ad_token` to `_arealtime`'s named parameter rather than
     **kwargs, so it must still reach the handler.

tests/unit/llms/azure_ai/chat/test_azure_ai_transformation.py

@@ -352,21 +352,6 @@ def test_azure_model_router_stamps_selected_model_on_hidden_params():
     )
 
 
-def test_azure_model_router_stamp_does_not_leak_across_responses():
-    """
-    ModelResponse declares _hidden_params as a class-level dict, so the stamp has to be written
-    as a fresh dict. Mutating in place would bleed the selected model into unrelated responses.
-    """
-    from litellm.llms.azure_ai.common_utils import (
-        AZURE_MODEL_ROUTER_SELECTED_MODEL_KEY,
-    )
-    from litellm.types.utils import ModelResponse
-
-    untouched = ModelResponse()
-
-    assert AZURE_MODEL_ROUTER_SELECTED_MODEL_KEY not in (untouched._hidden_params or {})
-
-
 def test_drop_tool_level_extra_fields_strips_copilot_mcp_server_name():
     """
     Regression test: Azure AI returns 400 when tools contain copilot_mcp_server_name.

tests/unit/llms/azure_ai/image_edit/test_azure_ai_image_edit_transformation.py

@@ -41,7 +41,7 @@ def test_azure_ai_url_generation():
     assert complete_url == expected_url
 
 
-def test_azure_ai_validate_environment_with_entra_token(monkeypatch):
+def test_azure_ai_validate_environment_with_entra_token(monkeypatch, no_ambient_azure_credentials):
     monkeypatch.delenv("AZURE_AI_API_KEY", raising=False)
     monkeypatch.setattr(litellm, "api_key", None)
     config = AzureFoundryFluxImageEditConfig()
@@ -55,7 +55,7 @@ def test_azure_ai_validate_environment_with_entra_token(monkeypatch):
     assert headers == {"Authorization": "Bearer entra-token"}
 
 
-def test_flux2_validate_environment_with_entra_token(monkeypatch):
+def test_flux2_validate_environment_with_entra_token(monkeypatch, no_ambient_azure_credentials):
     monkeypatch.delenv("AZURE_AI_API_KEY", raising=False)
     monkeypatch.setattr(litellm, "api_key", None)
     config = AzureFoundryFlux2ImageEditConfig()

tests/unit/llms/azure_ai/image_edit/test_mai_image_edit_transformation.py

@@ -174,7 +174,7 @@ class TestAzureMAIImageEdit:
         assert image_response.usage.total_tokens == 1024
 
 
-def test_mai_validate_environment_with_entra_token(monkeypatch):
+def test_mai_validate_environment_with_entra_token(monkeypatch, no_ambient_azure_credentials):
     monkeypatch.delenv("AZURE_AI_API_KEY", raising=False)
     monkeypatch.setattr(litellm, "api_key", None)
 

tests/unit/llms/azure_ai/passthrough/test_azure_ai_passthrough_transformation.py

@@ -256,13 +256,13 @@ def test_serverless_host_gets_a_bearer_token():
     assert "api-key" not in headers
 
 
-def test_entra_token_is_used_when_the_deployment_has_no_api_key():
+def test_entra_token_is_used_when_the_deployment_has_no_api_key(no_ambient_azure_credentials):
     headers = _auth_headers(api_key=None, api_base=FOUNDRY_BASE, litellm_params={"azure_ad_token": "entra-token"})
 
     assert headers["Authorization"] == "Bearer entra-token"
 
 
-def test_no_credentials_at_all_raises():
+def test_no_credentials_at_all_raises(no_ambient_azure_credentials):
     with pytest.raises(ValueError, match="Missing Azure AI credentials"):
         _auth_headers(api_key=None, api_base=FOUNDRY_BASE)
 

tests/unit/llms/azure_ai/rerank/test_azure_ai_rerank_transformation.py

@@ -105,7 +105,7 @@ class TestAzureAIRerankConfigValidateEnvironment:
 
         assert headers["Authorization"] == "Bearer my-key"
 
-    def test_falls_back_to_entra_token(self, monkeypatch):
+    def test_falls_back_to_entra_token(self, monkeypatch, no_ambient_azure_credentials):
         monkeypatch.delenv("AZURE_AI_API_KEY", raising=False)
         monkeypatch.setattr(litellm, "azure_key", None)