fix(search): validate search provider environment off the event loop

async_search called validate_environment on the event loop, so a Grounding with
Bing Search request that mints an Entra token blocked the loop for the token
endpoint round trip. It now runs in a worker thread like the Responses handlers

litellm/llms/custom_httpx/llm_http_handler.py

@@ -2084,8 +2084,8 @@ class BaseLLMHTTPHandler:
         if provider_config is None:
             raise ValueError(f"No provider config found for provider: {custom_llm_provider}")
 
-        # Validate environment and get headers
-        headers = provider_config.validate_environment(
+        validated_headers: Final = await asyncio.to_thread(
+            provider_config.validate_environment,
             api_key=api_key,
             api_base=api_base,
             headers=headers or {},
@@ -2097,7 +2097,7 @@ class BaseLLMHTTPHandler:
             optional_params=optional_params,
             api_key=api_key,
             api_base=api_base,
-            headers=headers or {},
+            headers=validated_headers,
         )
 
         # Get complete URL (pass data for providers that need request body for URL construction)
@@ -2111,7 +2111,7 @@ class BaseLLMHTTPHandler:
         signed_headers, signed_json_body = await sign_request_off_loop_if_aws(
             provider_config,
             provider_config.sign_request,
-            headers=headers,
+            headers=validated_headers,
             optional_params=optional_params,
             request_data=data,
             api_base=complete_url,
@@ -2125,7 +2125,7 @@ class BaseLLMHTTPHandler:
             additional_args={
                 "complete_input_dict": data,
                 "api_base": complete_url,
-                "headers": headers,
+                "headers": validated_headers,
             },
         )
 

tests/unit/llms/custom_httpx/test_llm_http_handler.py

@@ -24,6 +24,7 @@ from litellm.integrations.code_interpreter_interception.handler import (
 from litellm.integrations.custom_logger import CustomLogger
 from litellm.integrations.websearch_interception.handler import WebSearchInterceptionLogger
 from litellm.llms.azure.responses.transformation import AzureOpenAIResponsesAPIConfig
+from litellm.llms.azure.search.transformation import BingGroundingSearchConfig
 from litellm.llms.base_llm.audio_transcription.transformation import (
     AudioTranscriptionRequestData,
     BaseAudioTranscriptionConfig,
@@ -135,6 +136,48 @@ async def test_get_search_preserves_successful_results(is_async: bool, has_resul
                 assert response.results[0].snippet == "Example snippet"
 
 
+@pytest.mark.asyncio
+async def test_async_search_mints_the_bing_grounding_token_off_the_event_loop(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setenv("BING_GROUNDING_PROJECT_ENDPOINT", "https://acct.services.ai.azure.com/api/projects/proj")
+    monkeypatch.setenv("BING_GROUNDING_MODEL", "gpt-4.1")
+    monkeypatch.delenv("BING_GROUNDING_TOKEN", raising=False)
+    minting_threads: Final[list[int]] = []
+
+    def mint() -> str:
+        minting_threads.append(threading.get_ident())
+        return "minted-entra-token"
+
+    authorizations: Final[list[str | None]] = []
+
+    def foundry(request: httpx.Request) -> httpx.Response:
+        authorizations.append(request.headers.get("authorization"))
+        return httpx.Response(200, json={"output": []})
+
+    client: Final = AsyncHTTPHandler()
+    await client.close()
+    client.client = httpx.AsyncClient(transport=httpx.MockTransport(foundry))
+
+    response: Final = await BaseLLMHTTPHandler().search(
+        query="test",
+        optional_params={},
+        timeout=5,
+        logging_obj=Mock(),
+        api_key=None,
+        api_base=None,
+        custom_llm_provider="bing_grounding",
+        client=client,
+        asearch=True,
+        provider_config=BingGroundingSearchConfig(entra_token_minter=mint),
+    )
+
+    assert response.object == "search"
+    assert authorizations == ["Bearer minted-entra-token"]
+    assert len(minting_threads) == 1
+    assert threading.get_ident() not in minting_threads
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("is_async", (False, True))
 async def test_get_search_preserves_tinyfish_http_error_formatting(is_async: bool) -> None: