fix(spend-logs): trim logged response strings (#16654)

* fix(spend-logs): trim logged response strings

- route spend-log responses through the existing string sanitizer so oversized base64/text fields are truncated before persistence
- add unit tests covering the truncation path and the feature flag

Note: embeddings-specific truncation (numeric vectors) is still pending and will be handled separately.

* remove unnecessary comment

* add: sanitization unit test for embeddings

* fix: simplify sanatization logic

I overcomplicated a simple change for lack of understanding, fixed.

litellm/proxy/spend_tracking/spend_tracking_utils.py

@@ -651,7 +651,20 @@ def _get_response_for_spend_logs_payload(
     if payload is None:
         return "{}"
     if _should_store_prompts_and_responses_in_spend_logs():
-        return json.dumps(payload.get("response", {}))
+        response_obj: Any = payload.get("response")
+        if response_obj is None:
+            return "{}"
+
+        sanitized_wrapper = _sanitize_request_body_for_spend_logs_payload(
+            {"response": response_obj}
+        )
+        sanitized_response = sanitized_wrapper.get("response", response_obj)
+
+        if sanitized_response is None:
+            return "{}"
+        if isinstance(sanitized_response, str):
+            return sanitized_response
+        return safe_dumps(sanitized_response)
     return "{}"
 
 

tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py

@@ -4,7 +4,7 @@ import json
 import os
 import sys
 from datetime import timezone
-from typing import Any
+from typing import Any, cast
 
 import pytest
 from fastapi.testclient import TestClient
@@ -19,10 +19,12 @@ import litellm
 from litellm.constants import LITELLM_TRUNCATED_PAYLOAD_FIELD, REDACTED_BY_LITELM_STRING
 from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
 from litellm.proxy.spend_tracking.spend_tracking_utils import (
+    _get_response_for_spend_logs_payload,
     _get_vector_store_request_for_spend_logs_payload,
     _sanitize_request_body_for_spend_logs_payload,
     get_logging_payload,
 )
+from litellm.types.utils import StandardLoggingPayload
 
 
 def test_sanitize_request_body_for_spend_logs_payload_basic():
@@ -235,6 +237,71 @@ def test_get_vector_store_request_for_spend_logs_payload_null_input(mock_should_
     assert result is None
 
 
+@patch(
+    "litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
+)
+def test_get_response_for_spend_logs_payload_truncates_large_base64(mock_should_store):
+    from litellm.constants import MAX_STRING_LENGTH_PROMPT_IN_DB
+
+    mock_should_store.return_value = True
+    large_text = "A" * (MAX_STRING_LENGTH_PROMPT_IN_DB + 500)
+    payload = cast(
+        StandardLoggingPayload,
+        {
+        "response": {
+            "data": [
+                {
+                    "b64_json": large_text,
+                    "other_field": "value",
+                }
+            ]
+        }
+        },
+    )
+
+    response_json = _get_response_for_spend_logs_payload(payload)
+    parsed = json.loads(response_json)
+    truncated_value = parsed["data"][0]["b64_json"]
+    assert len(truncated_value) < len(large_text)
+    assert LITELLM_TRUNCATED_PAYLOAD_FIELD in truncated_value
+    assert parsed["data"][0]["other_field"] == "value"
+
+
+@patch(
+    "litellm.proxy.spend_tracking.spend_tracking_utils._should_store_prompts_and_responses_in_spend_logs"
+)
+def test_get_response_for_spend_logs_payload_truncates_large_embedding(mock_should_store):
+    from litellm.constants import MAX_STRING_LENGTH_PROMPT_IN_DB
+
+    mock_should_store.return_value = True
+    embedding_values = [
+        round(i * 0.0001, 6) for i in range(MAX_STRING_LENGTH_PROMPT_IN_DB + 500)
+    ]
+    large_embedding = json.dumps(embedding_values)
+    payload = cast(
+        StandardLoggingPayload,
+        {
+            "response": {
+                "data": [
+                    {
+                        "embedding": large_embedding,
+                        "other_field": "value",
+                    }
+                ]
+            }
+        },
+    )
+
+    response_json = _get_response_for_spend_logs_payload(payload)
+    parsed = json.loads(response_json)
+    truncated_value = parsed["data"][0]["embedding"]
+    
+    assert isinstance(truncated_value, str)
+    assert len(truncated_value) < len(large_embedding)
+    assert LITELLM_TRUNCATED_PAYLOAD_FIELD in truncated_value
+    assert parsed["data"][0]["other_field"] == "value"
+
+
 def test_safe_dumps_handles_circular_references():
     """Test that safe_dumps can handle circular references without raising exceptions"""